5 Clustering of Synoptic Systems During the Indian Summer Monsoon by In- ... Bibliography. 100 ... nal low frequency variations of the Indian summer monsoon?
INTRASEASONAL OSCILLATIONS AND INTERANNUAL VARIABILITY OF THE INDIAN SUMMER MONSOON
A thesis submitted for the award of the degree of
Doctor of Philosophy in the Faculty of Engineering by
R. S. AJAYA MOHAN
Centre for Atmospheric and Oceanic Sciences Indian Institute of Science Bangalore 560 012 INDIA NOVEMBER 2001
Dedicated to My Parents
”I looked forward to the coming of the monsoon and I became a watcher of skies, waiting to spot the heralds the preceded the attack. A few showers came. Oh! that was nothing, I was told; the monsoon has yet to come. Heavier rains followed, but I ignored them and waited for some extraordinary happening. While I waited I learnt from various people that the monsoon had definitely come and established itself. Where was the pomp and circumstance and the glory of the attack, and the combat between cloud and land, and the surging and lashing sea? Like a thief in the night the monsoon had come to Bombay, as well it might have done in Allahabad or elsewhere. Another illusion gone” Jawaharlal Nehru, The monsoon comes to Bombay, 1939
Contents Acknowledgements
i
Abstract
iii
Acronyms
viii
List of Figures 1
2
3
4
ix
Introduction 1.1 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Reanalysis Data . . . . . . . . . . . . . . . . . . . . . . . 1.2.2 NOAA Outgoing Long wave Radiation (OLR) Dataset 1.2.3 Precipitation Datasets . . . . . . . . . . . . . . . . . . . 1.2.4 Statistics of Low Pressure Systems . . . . . . . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
1 . 8 . 9 . 9 . 10 . 11 . 12
Basic Characteristics of Monsoon Intraseasonal Oscillations 2.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Propagation Characteristics . . . . . . . . . . . . . . . . . 2.3 A Circulation Criterion for ’Active’ and ’Break’ Phases . 2.4 Mean Structure of ISOs . . . . . . . . . . . . . . . . . . . . 2.5 Meridional Bimodality of ISO Spatial Structure . . . . . . 2.6 Discussions and Conclusions . . . . . . . . . . . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
13 13 16 19 21 30 32
Intraseasonal Oscillations and Interannual Variability of the Indian Summer Monsoon 3.1 A Common Spatial Mode of Intraseasonal and Interannual Variability . . 3.2 Probability of ’Active’/’Break’ Conditions and Seasonal Mean Monsoon . 3.3 Interannual Variations of ISO Activity and Seasonal Mean Monsoon . . . 3.4 Discussions and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . .
34 34 40 50 51
Estimate of Potential Predictability of Monthly and Seasonal Means in Tropics from Observations 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Estimation of Potential Predictability of Monthly Means . . . . . . . . . . . 4.2.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.2 Estimation of ’Internal’ and ’External’ Interannual Variances . . . . 4.2.3 Potential Predictability of Monthly Means . . . . . . . . . . . . . . 4.3 Potential Predictability of Seasonal means . . . . . . . . . . . . . . . . . . . 4.4 Discussions and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . .
55 55 58 58 65 68 76 82
. . . . . .
. . . . . .
. . . . . .
4.5 5
6
Appendix : Procedure for Estimating ’Climate Noise’ . . . . . . . . . . . . 85
Clustering of Synoptic Systems During the Indian Summer Monsoon traseasonal Oscillations 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Wet and Dry Spells and Clustering of LPS . . . . . . . . . . . . . . . 5.3 Monsoon Intraseasonal Oscillation Index . . . . . . . . . . . . . . . 5.4 Clustering of Genesis of LPS by Intraseasonal Oscillations . . . . . 5.5 Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . Conclusions
Bibliography
by In. . . . .
. . . . .
. . . . .
. . . . .
87 87 89 91 91 95 96 100
Acknowledgments I am fortunate to have got a chance to work with Prof B. N. Goswami. The vastness of his knowledge and his abundant enthusiasm to seek new results has always been a source of encouragement to me. His critical appraisals and encouragement of independent thought during discussions have contributed immensely to the course of this thesis. It helped smoothen many hurdles. His calm and collective approach was invaluable in putting things in the right perspective. He shall always remain in my mind as a model of an intelligent, enthusiastic and hardworking man-of-science. Many thanks are due to the faculty in CAOS, especially Prof Sengupta and Prof Srinivasan for many informal discussions and encouragement. The cheerful and helpful people in CAOS, supporting staff, project staff and students deserve appreciation. Rama, Padma, Mohan, Raja and Shiva were always obliging despite their strenuous work load. Thanks are due to Natraj, Rajasekhar and Srinivas for helping me in solving system related problems. It was a pleasure to work with Rajendran, Sajani, Anagani, Janakiraman, J V S Raju, Salil, Arindam, Chandru and Pallav. Informal discussions/chat with Retish, Prince, Francis, Vinoj and Simi both within the department and in Raffique’s Tea-kiosk is fondly acknowledged. No words can express my gratitude to Manu as she has been a constant source of support and encouragement. I would like to acknowledge interactions with Prof N. Balakrishnan for I have learned a lot from him. His vision and cool and collected approach was indeed impressive. I will be failing in my duty, if I do not thank Prof G. Padmanabhan - He taught me how simple and straightforward a human being should be. I am grateful to Prof M. Ghil, Dr M. Kimoto and Dr A. Robertson for providing information regarding probability density estimation. Thanks are due to the faculty in Department of Atmospheric Science, Cochin University of Science & Technology, especially Prof Mohan Kumar for encouraging me to take up a research career.
Acknowledgements
ii
I cherish the brief interaction with Bharataratna Dr A.P.J Abdul Kalam. His words ’Only strength respects strength’ gave me a fresh sense of enthusiasm to work for this great country. Working for Students’ council helped me understand problems of others and gave me ample opportunity to interact with all kinds of people. This was also an opportunity to befriend many. I would like to acknowledge the support and friendship of Saishankar, Rajkumar, Vasan, Ganesh, Guruprasad, Pratap Jayaprakash, John, Dhruba and Suresh. Special thanks are due to Brar and Suma for, they offered good company. Thanks are due to the the ’famous mallu gang’ as the friendship they offered is matchless. Cheers to Venu, Anoop (simplan) and Bijoy. It is pleasure to have friends like Anil (kunz), Hari (healy), Glomin (RC), Anil (aavi), Suresh (kumily), Manoj (neergosh), Prabhu, Randhir (kuru), Baiju, Ajayan, Sameen, John, Vinay and Vinod. Thanks are also due to Pappan and Pappy, Suresh (cobra) and Sridevi, Sunoj and Viji. Let me acknowledge the support offered by Sriram. Finally, my deepest sense of gratitude go to my parents and my family for their goodwill and blessings. I am grateful to my father for sharing all my worries and happiness. Without his constant encouragement this thesis would not have been possible. Innumerable phone calls gave me a feeling that I am at home, away from home. I shall always strive to rise up to his expectations. Thanks to all the unknown faces that continue to develop and strive for free software. Working with ”LINUX”, ”GrADS” and ”LATEX” and numerous other free software made life easier in the pursuit of this thesis. Thanks are due to SERC for high power computing. Last but not least, let me thank Indian Institute of Science and Council of Scientific and Industrial Research for providing financial support.
Abstract Several modeling studies show that the predictability of the seasonal mean Indian summer monsoon is limited due to a significant fraction of the interannual variability of the seasonal mean being governed by internal chaotic dynamics. What causes the internal low frequency variations of the Indian summer monsoon? One possible candidate is the monsoon intraseasonal oscillations (ISOs). Indian summer monsoon has vigorous intraseasonal oscillations in the form of ’active’ and weak (or ’break’) spells of monsoon rainfall within the summer monsoon season. These ’active’ and ’break’ spells of the monsoon are associated with fluctuations of the tropical convergence zone. Temporally ISOs of the Indian summer monsoon represent two preferred bands of periods, one between 10 and 20 days and the other between 30 and 60 days. As the separation between the dominant ISO periods and the season is not large, the statistics of the ISOs could, in principle, influence the seasonal mean monsoon and it’s interannual variability. To the extent that the ISOs are intrinsically chaotic and unpredictable, the predictability of the Indian summer monsoon would depend on relative contribution of the ISOs to the seasonal mean compared to the more predictable externally forced component. Therefore, it is of great importance to establish (a) whether there exits a physical basis for monsoon ISOs to influence the seasonal mean. (b) Even if there exits a physical basis for the ISOs to influence the seasonal mean, is there an empirical evidence of association between some statistics of the ISOs and interannual variability of the Indian summer monsoon? (c) If such an association between monsoon ISOs and the seasonal mean monsoon exits, it would be desirable to make quantitative estimate of the extent to which ISOs influence the seasonal mean and its interannual variability. The primary objectives of this study are to address these three issues using sufficiently long homogeneous daily circulation and convection data. Although spatial and temporal structures of the monsoon ISOs have been examined extensively, the relationship between ISOs and interannual variability has received little attention in the past. The existing literature on the subject is critically reviewed in Chapter 1.
Abstract
iv
In an attempt to establish the physical basis for the ISOs to influence the seasonal mean, we first examine the similarity between the spatial structure of the ISOs and the seasonal mean. The large scale nature of the Indian summer monsoon ISOs and relationship between circulation and convection on this time scale are investigated using 42-years (1956-1997) daily circulation data from NCEP/NCAR reanalysis and satellite derived outgoing long wave radiation data for the period 1974-1997. Traditionally, ’active’ and ’break’ conditions or the dry and wet spells of the monsoon ISO are defined based on continental precipitation. Arguing that the dry and wet spells are part of large scale fluctuations associated with the ISO, a circulation based criterion is devised to define ’active’ and ’break’ monsoon conditions using zonal winds at 850 hPa over the Bay of Bengal. Although the ISOs vary in intensity and period, it is shown that, the underlying spatial structure of a typical ISO cycle in circulation and convection is invariant over the years and is constructed using a composite technique. Typical ISOs have large scale horizontal structure similar to the seasonal mean and intensifies (weakens) the mean flow during it’s ’active’ (’break’) phase. A typical ’active’ (’break’) phase is also associated with enhanced (decreased) cyclonic low-level vorticity and convection and anomalous upward (downward) motion in the northern position of the tropical convergence zone (TCZ) and decreased (increased) convection and anomalous downward (upward) motion in the southern position of the TCZ. The cycle evolves with a northward propagation of the TCZ and convection from the southern to the northern position of the TCZ. Thus the ISOs result in spinning up (or spinning down) of the large scale mean monsoon circulation in it’s extreme phases. (Chapter 2) A physical basis for ISOs to influence the seasonal mean and it’s interannual variability is established when it is shown that the intraseasonal and interannual variations are governed by a common mode of spatial variability. The spatial pattern of standard deviation of intraseasonal and interannual variability of low-level vorticity is shown to be similar. The spatial pattern of the dominant mode of ISO variability of the low-level winds is also shown to be similar to that of the interannual variability of the seasonal mean winds. The similarity between the spatial patterns of the two variability indicates that higher frequency of occurrence of ’active’ (’break’) conditions would result in ’stronger’ (’weaker’) than normal seasonal mean. This possibility is tested by calculating probability density function (PDF) of the ISO activity in the low-level vorticity represented by the two dominant empirical orthogonal functions (EOFs). The PDF estimates for ’strong’ monsoon years and ’weak’ monsoon years are shown to be asymmetric in
Abstract
v
both the cases. It is seen that the ’strong’ (’weak’) monsoon years are associated with higher probability of occurrence of ’active’ (’break’) conditions. This result is further supported by calculation of PDF of ISO activity from combined vorticity and outgoing long wave radiation. This result, indicates that the frequency of intraseasonal pattern determine the seasonal mean. As the ISOs are essentially chaotic, it raises an important question on predictability of the Indian summer monsoon. (Chapter 3) Having shown that the ISOs can influence the seasonal mean and its interannual variability, the next objective is to make quantitative estimates of potential predictability of the monsoon climate. A measure of potential predictability of the monthly and seasonal means in a place could be obtained from the ratio of variances associated with the ’external’ to the ’internal’ components. A method of separating the ’external’ component arising from contributions from slowly varying boundary forcing from the ’internal’ components (e.g. intraseasonal oscillations) that determines the potential predictability of the monthly mean tropical climate is proposed. Based on 33 years of daily low-level wind observations and 24 years of satellite observations of outgoing long wave radiation, we show that the Indian monsoon climate is marginally predictable on monthly time scales as the contribution from the boundary forcing in this region is comparable to that from the internal dynamics. It is further shown that excluding the Indian monsoon region, the predictable region is larger and predictability is higher in the tropics during northern summer. Even though the boundary forced variance is large during northern winter, the predictable region is smaller as the internal variance is larger and covers a larger region during northern winter due to stronger intraseasonal activity. It is also shown that most of the internal low frequency variability in the Indian summer monsoon region arise from the ISOs. (Chapter 4) An estimate of potential predictability for the Northern Hemisphere summer and winter seasons in the tropics has also been made using an established method of estimating ’climate noise’. Even on seasonal mean time scales, we show that the Indian monsoon climate is only marginally predictable as the contribution of the boundary forcing in this region is relatively low and that of the internal dynamics is relatively large. (Chapter 4) While the monsoon ISOs seem to lead to decrease in the predictability of monthly or seasonal mean monsoon climate, it is possible that the same ISOs lead to extended range prediction of spells of synoptic activity. We recall that the seasonal mean monsoon is strengthened in one phase of the ISOs (active phase) while it is weakened in
Abstract
vi
another (’break’) phase of the monsoon. The main rain bearing system during the monsoon season are the Low Pressure Systems (LPS) consisting of lows and depressions. Since the genesis of the LPS depends on the horizontal shear and low-level vorticity, it is possible that more LPS form in the active phase relative to the break phase. In other words, large scale circulation associated with the ISOs could modulate the frequency of genesis of LPS. We examined how the LPS are modulated by the intraseasonal oscillations. Using more than 40 years of LPS genesis statistics and daily circulation data, here we show that the dry and wet spells are the result of clustering of lows and depressions caused by modulation of the large scale monsoon flow by the intraseasonal oscillations. The slow evolution of the ISOs may permit extended range prediction of the ISO phases and through them dry and wet spells of the Indian summer monsoon (Chapter 5). Major results and outstanding issues are discussed in Chapter 6.
Publications 1. B.N Goswami and R.S Ajaya Mohan, 2001: Intraseasonal Oscillations and Int erannual Variability of Indian Summer Monsoon. J.Climate, 14, 1180 -1198. 2. B.N Goswami and R.S. Ajaya Mohan, 2001: Estimate of Predictability of Monthly Means in Tropics from Observations. Curr.Sci., 80, 56-63. 3. R. S. Ajaya Mohan and B.N. Goswami, 2000: A Common spatial mode for intraseasonal and interannual variation and predictability of the Indian Summer Monsoon. Curr.Sci., 79, 1106-1111. 4. B.N Goswami and R.S. Ajaya Mohan, 2001: Intra-seasonal Oscillations and predictability of the Indian summer monsoon. Proc.Ind.Nat.Sci.Aca., 67A (3), 369-383. 5. B.N Goswami, R.S Ajaya Mohan, Prince K Xavier and D. Sengupta 2001: Clustering of low-pressure systems during the Indian summer monsoon by Intraseasonal Oscillations, Geophys.Res.Letts, 30, 1431, doi:10.1029/2002GL016734. 6. R. S. Ajaya Mohan and B. N. Goswami 2003: Potential predictability of the Asian Summer Monsoon on Monthly and Seasonal Time Scales, Meteorol.Atmos.Phys., 84, 83-100. 7. B.N Goswami and R.S Ajaya Mohan, 2005: Multi-scale interactions and predictability of the Indian summer monsoon, section 3 in Nonequillibrium Phenomena
Abstract
vii
in Plasmas, Eds. Sharma, A. Surjalal, Kaw Predhiman, K, IX, 347p, ISBN: 1-40203108-4, Springer, USA.
Acronyms AGCM
Atmospheric General Circulation Model
AMIP
Atmospheric General Circulation Model Intercomparison Project
CEOF
Combined Empirical Orthogonal Function
CMAP
Climate prediction center Merged Analysis of Precipitation
DJF
December-January-February
ENSO
˜ El Nino-Southern Oscillation
EOF
Empirical Orthogonal Function
JJA
June-July-August
JJAS
June-July-August-September
IMD
India Meteorological Department
IMR
All India Monsoon Rainfall Index
ISOs
Intraseasonal oscillations
LPS
Low Pressure Systems
MISI
Monsoon Intraseasonal Oscillation Index
MTV
Monsoon Trough Vorticity
NCAR
National Center for Atmospheric Research
NCEP
National Centers for Environmental Prediction
NH
Northern Hemisphere
OLR
Outgoing Long wave Radiation
PC
Principal Component
PDF
Probability Density Function
SD
Standard Deviation
TCZ
Tropical Convergence Zone
U850
Low level zonal winds (850 hPa)
U200
Upper level zonal winds (200 hPa)
Z700
Geopotential height (700 hPa)
List of Figures 2.1
Climatological mean (JJAS) monsoon winds (ms−1 ) and precipitation (mm.day−1 ). (a) 850 hPa vector winds, (b) Relative vorticity at 850 hPa (10−6 s−1 ), (c) 200 hPa vector winds, (d) Precipitation from Xie and Arkin [1997]. . . . . . . . . . . . . . 14
2.2
Some examples of raw time series of zonal winds at 850 hPa at a few selected points during 1990. (Left panels) Daily zonal winds (ms−1 ) with the annual cycle (annual and semi-annual harmonics, green lines). (Right panels) Anomalous daily zonal winds (ms−1 ).
2.3
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Examples of spectra of zonal winds and OLR for a typical year (1984) at a typical point (90◦ E, 10◦ N). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4
An example illustrating the horizontal scale and vertical structure of the dominant ISO mode. (a) Lag-zero correlations of the 850 hPa 30-60 day filtered zonal winds with respect to a reference point (85◦ E, 10◦ N). (b) Lag-zero correlations between 30-60 day filtered zonal winds at 850 hPa and 200 hPa at each grid point. Correlations are calculated between May 1 and October 31 of 1990. Correlations exceeding 0.2 are significant at 95% confidence level. . . . . . . . . . . 17
2.5
(a) Correlations between 30-60 day filtered zonal winds at 850 hPa with respect to that at a reference point (85◦ E, 10◦ N) at different lead/lags averaged over (80◦ E-90◦ E) for 1990. (b) Same as (a) but for 30-60 day filtered OLR. Starting contour is ±0.1 and contour interval is 0.2. . . . . . . . . . . . . . . . . . . . . . 18
2.6
(a) Correlations between 30-60 day filtered zonal winds at 850 hPa with respect to that at a reference point (85◦ E, 10◦ N) at different lead/lags averaged over (10◦ N-20◦ N) for 1990. (b) Same as (a) but for 30-60 day filtered OLR. Starting contour is ±0.1 and contour interval is 0.2.
. . . . . . . . . . . . . . . . . . . . 18
List of Figures 2.7
x
(a) An example of 30-60 day filtered zonal winds for 1986 at a reference point (90◦ E, 15◦ N). The thin horizontal lines correspond to +1 and -1 standard deviations. ’Active’ (’break’) days are defined as days for which the filtered zonal winds at the reference point are greater than +1 S.D (or less than -1 S.D). (b) 12year (1978-1989) mean precipitation difference (mm.day−1 ) between all ’active’ and ’break’ composites. Contours are ±(1, 3, 5, 7, 9, 11, 13, 15). . . . . . . . . . . 20
2.8
(a,b) Climatological mean composite vector wind anomalies (ms−1 ) at 850 hPa corresponding to ’active’ and ’break’ conditions for the 30-60 day mode and (c,d) associated relative vorticity (10−6 s−1 ). The climatological mean composite is calculated by averaging all ’active’ and ’break’ conditions occurring during the 20-year period (1978-1997). Shading in the upper panels indicates regions with anomalies significant above 90% confidence level. . . . . . . . . . . . . . . 21
2.9
Climatological mean composite vector wind anomalies (ms−1 ) corresponding to ’active’ and ’break’ conditions for the 30-60 day mode (a,b) at 500 hPa and (c,d) at 200 hPa. The climatological mean composite is calculated by averaging all ’active’ and ’break’ conditions occurring during the 20-year period (19781997). Shading indicates regions with anomalies significant above 90% confidence level. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.10 Climatological mean composite OLR anomalies (Wm−2 ) corresponding to ’active’ and ’break’ conditions. ’Active’ and ’break’ composites are constructed using unfiltered OLR anomalies and the same ’active’ and ’break’ dates defined by 30-60 day filtered zonal wind anomalies as used in Figure 2.8. OLR anomalies above 5 Wm−2 are significant above 90% confidence level. . . . . . . . . . . . . 23
2.11 Climatological mean composite pressure vertical velocity anomalies (ω) at 500 hPa (hPas−1 ). Again the same ’active’ and ’break’ dates chosen from 30-60 day filtered zonal wind anomalies for the 20-year period (1978-1997) as used in Figure 2.8 and Figure 2.10 are used. . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.12 Climatological mean composite vector wind anomalies (ms−1 ) at 850 hPa and associated relative vorticity (10−6 s−1 ) corresponding to eight phases of evolution of the 30-60 day mode for the period 1979-1989. The phase-1 corresponds to the days when the filtered zonal wind anomalies at the reference point is zero and increasing toward positive values. . . . . . . . . . . . . . . . . . . . . . . . 27
List of Figures
xi
2.13 Climatological mean composite OLR anomalies (Wm−2 ) corresponding to eight phases of evolution of the 30-60 day mode for the period 1979-1997. Eight composite phases are constructed using unfiltered OLR anomalies and the same dates defined by 30-60 day filtered zonal winds as used in Figure 2.12. . . . . . . 28
2.14 (a,b) Climatological mean composite vector wind anomalies (ms−1 ) at 850 hPa corresponding to ’active’ and ’break’ conditions for the 10-20 day mode and (c,d) associated relative vorticity (10−6 s−1 ). The climatological mean composite is calculated by averaging all ’active’ and ’break’ conditions occurring during the 20-year period (1978-1997). . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.15 Meridional bimodality of spatial structure of the dominant ISO. (a) Scatter plot of daily 30-60 day filtered vorticity at 850 hPa (10−6 s−1 ) over a northern band (70◦ E-100◦ E, 12◦ N-22◦ N) and a southern band (70◦ E-100◦ E, 5◦ S-10◦ N) during 1 June to 30 September for 19 years (1979-1997). (b) Scatter plot of 30-60 day filtered OLR anomalies (Wm−2 ) averaged over the northern TCZ (70◦ E-100◦ E, 12◦ N-22◦ N) and the southern TCZ (70◦ E-100◦ E, 0◦ -12◦ S) during 1 June to 30 September for 18 years (1979-1997, excluding 1994). . . . . . . . . . . . . . . . . 29
2.16 (a) Scatter plot of 30-60 day filtered relative vorticity at 850 hPa (10−6 s−1 ) and OLR (Wm−2 ) anomalies averaged over a box (85◦ E-95◦ E, 12◦ N-22◦ N) of the northern TCZ during 1 May to 31 October for 19 years (1979-1997). (b) same as (a) but averaged over a box (85◦ E-95◦ E, 0◦ -12◦ S) of the southern TCZ. . . . . . 31
3.1
Geographical distribution of intraseasonal and interannual activity. (a) Mean standard deviation of ISO filtered relative vorticity (10−6 s−1 ) at 850 hPa during 1 June to 30 September for 20 years (1978-1997). (b) Interannual standard deviation of seasonal mean relative vorticity (JJAS, 10−6 s−1 ) based on the same 20 years. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2
First EOF of the intraseasonal and interannual 850 hPa winds. (a) Intraseasonal EOFs are calculated with ISO filtered winds for the summer months (1 June to 30 September) for a period of 20 years (1978-1997). (b) Interannual EOFs are calculated with the seasonal mean (JJAS) winds for 40-year period (1958-1997). Units of vector loading are arbitrary. (c) Relation between IMR and interannual PC1. Filled bars indicate interannual PC1 and the unfilled bar represent IMR. Both time series are normalized by their own standard deviation. Correlation between the two time series is shown. . . . . . . . . . . . . . . . . . . . . . . . 38
List of Figures 3.3
xii
Ratio between standard deviation of interannual variation of ISO activity and interannual variation of the seasonal mean. (a) Relative vorticity at 850 hPa. (b) OLR. Contours are (0.3, 0.4, 0.6, 0.8, 1.0). . . . . . . . . . . . . . . . . . . . . . . 39
3.4
First two EOFs of the daily ISO filtered 850 hPa vorticity from 1 June to 30 September. (a) EOF1 and (b) EOF2 for seven ’strong’ years (c) EOF1 and (d) EOF2 for ten ’weak’ years (e) EOF1 and (f) EOF2 for ’all’ (20 years from 1978 to 1997) years. Arbitrary EOF loadings have been multiplied by a factor of 100. . . . 42
3.5
Evidence of change in regimes of ISOs during ’strong’ and ’weak’ monsoon years. Illustrated are two-dimensional PDFs of the ISO state vector spanned by two dominant EOFs of low-level vorticity. PDFs are calculated with principal components normalized by their own standard deviation and taking the summer days (1 June to 30 September) for (a) 7 ’strong’ monsoon years (b) 10 ’weak’ monsoon years (c) 20 combined ’strong’, ’weak’ and ’normal’ years (1978-1997). The smoothing parameter used is h=0.6 and PDFs are multiplied by a factor 100. The first two EOFs (not shown) are different in ’strong’, ’weak’ and ’all’ years but are related to ’active’ and ’break’ conditions. The origin of the plots corresponds to a very weak state representing a transition between the two states (as in the ’all’ case). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.6
Geographical patterns of the dominant regimes for low-level relative vorticity (10−6 s−1 ) shown in Figure 3.5. (a) ’strong’ monsoon years (b) ’weak monsoon years (c) ’all’ years. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.7
The monsoon trough vorticity (MTV) and the Indian Monsoon Rainfall (IMR) for a 40-year period (1958-1997). MTV is defined as the seasonal mean vorticity (JJAS) averaged in the domain 40◦ E-90◦ E and 10◦ N-30◦ N. Both time series are normalized by their own standard deviation. Correlation between the two time series is shown. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.8
First two CEOFs of the daily ISO filtered 850 hPa vorticity and OLR from 1 June to 30 September. (a) CEOF1 and (b) CEOF2 for six ’strong’ years (c) CEOF1 and (d) CEOF2 for six ’weak’ years (e) CEOF1 and (f) CEOF2 for ’all’ (20 years from 1978 to 1997) years. Arbitrary EOF loadings have been multiplied by a factor of 100. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.9
Same as Figure 3.5 but based on the state vector defined by the first two combined EOF of low-level vorticity and OLR. . . . . . . . . . . . . . . . . . . . . . 48
List of Figures
xiii
3.10 Geographical patterns of the dominant regimes shown in Figure 3.9. (a) ’strong’ monsoon years (b) ’weak monsoon years (c) ’all’ years. OLR patterns are shown as shaded contours (Wm−2 ) while the corresponding low-level vorticity are shown in contours (10−6 s−1 ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.11 (a) Time series of ISO activity index (blue) and All India Monsoon Rainfall Index (IMR, black) normalized by it’s own standard deviation for a 44-year period (1954-1997). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1
An illustration of variations of the annual cycle from year to year. The annual cycle of zonal winds (ms−1 ) at 850 hPa at a point (80◦ E, 5◦ N) are shown for 20 years. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.2
First combined EOF of mean monthly ’external’ anomalies for the period January 1979 to December 1997 (228 months). (a) Zonal winds EOF at 850 hPa, (b) OLR EOF and (c) PC1 (solid line) and Nino3 SST anomalies (dashed line). Both the time series are normalized by their own standard deviation. Units of the EOFs are arbitrary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3
Time-longitude section of mean monthly ’external’ anomalies of zonal wind at 850 hPa (ms−1 ) and OLR (Wm−2 ) averaged around equator (5◦ S-5◦ N).
4.4
. . . . . 64
Monthly variance of zonal winds (m2 s−2 ) at 850 hPa based on 396 months for the period January 1965 to December 1997. (a) Total variance (b) ’external’ variance and (c) ’internal’ variance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.5
Same as Figure 4.4 but for OLR for the period January 1980 to December 1999 (240 months). Units, (Wm−2 )2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.6
Estimates of ’F’ ratios for zonal winds at 850 hPa (a) for all northern hemisphere summer months (June-July-August) and (b) for all northern hemisphere winter months (December-January-February). . . . . . . . . . . . . . . . . . . . . . . . 69
4.7
The ’external’ variance of zonal winds at 850 hPa (m2 s−2 ) during (a) NH summer months (JJA) and (b) NH winter months (DJF). . . . . . . . . . . . . . . . . 70
4.8
The ’internal’ variance of zonal winds at 850 hPa (m2 s−2 ) during (a) NH summer months (JJA) and (b) NH winter months (DJF). . . . . . . . . . . . . . . . . . . 70
4.9
Estimates of ’F’ ratios for zonal winds at 200 hPa (a) for all northern hemisphere summer months (JJA) and (b) for all northern hemisphere winter months (DJF). . 72
4.10 Estimates of ’F’ ratios for OLR (a) for all northern hemisphere summer months (JJA) and (b) for all northern hemisphere winter months (DJF). . . . . . . . . . . 72
List of Figures
xiv
4.11 Estimates of ’F’ ratios for geopotential height at 700 hPa (a) for all northern hemisphere summer months (JJA) and (b) for all northern hemisphere winter months (DJF). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.12 The ’external’ variance of geopotential height at 700 hPa (gpm2 ) during (a) NH summer months (JJA) and (b) NH winter months (DJF). . . . . . . . . . . . . . . 74
4.13 The ’internal’ variance of geopotential height at 700 hPa (gpm2 ) during (a) NH summer months (JJA) and (b) NH winter months (DJF). . . . . . . . . . . . . . . 74
4.14 The ’internal’ variance of (a) zonal winds at 850 hPa (m2 s−2 ) and (b) OLR (Wm−2 )2 based on all months after removing the higher frequencies with period shorter than 10 days. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.15 Estimates of ’F’ ratios for zonal winds at 850 hPa for (a) NH summer season (JJA) (b) NH winter season (DJF). . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.16 Estimates of ’climate noise’ for zonal winds at 850 hPa for (a) NH summer season (JJA) (b) NH winter season (DJF). . . . . . . . . . . . . . . . . . . . . . . . 78
4.17 Estimates of ’F’ ratios for zonal winds at 200 hPa for (a) NH summer season (JJA) (b) NH winter season (DJF). . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.18 Estimates of ’F’ ratios for OLR for (a) NH summer season (JJA) (b) NH winter season (DJF). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 4.19 Estimates of ’F’ ratios for geopotential height at 700 hPa for (a) NH summer season (JJA) (b) NH winter season (DJF). . . . . . . . . . . . . . . . . . . . . . . 81
4.20 Estimates of ’climate noise’ for geopotential height at 700 hPa for (a) NH summer season (JJA) (b) NH winter season (DJF). . . . . . . . . . . . . . . . . . . . 81
5.1
Genesis dates of LPS between 1 June and 30 September of all years during 1979 to 1993 over the monsoon trough as a function of normalized departure of precipitation over the trough from the seasonal mean. . . . . . . . . . . . . . . . . 89
5.2
Leading Empirical Orthogonal Functions ( (a) EOF1 & (b) EOF2) of 10-80 day filtered wind anomalies (ms−1 ) at 850 hPa between June 1 and September 30 for the period 1964-1973. (c) Normalized time series of PC1 and PC2 for ten years (each year has 122 days). (d) Normalized Monsoon Intraseasonal Oscillation Index (MISI) for 10 years. Periods of MISI > +1 (MISI< -1) correspond to active (break) phases of the monsoon. It may be noted that positive (negative) phase of MISI represents enhancement (weakening) of the EOF1 pattern. . . . . . . . . 90
List of Figures 5.3
xv
Histogram of genesis of synoptic events (lows & depressions) for the Indian monsoon region (50◦ E-100◦ E, Eq-30◦ N) during June to September for the period 1954-1993 as a function of normalized MISI. . . . . . . . . . . . . . . . . . . . . 92
5.4
Total (climatology+composite anomaly) relative vorticity (10−6 s−1 ) at 850 hPa during the (a) ’Active’ ISO phase (MISI > +1) and (b) ’Break’ ISO phase (MISI < -1). Dark dots indicate the position of the genesis of the LPS during active and break phases. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.5
Composites based on active and break days as defined by the ISO index, MISI. (a) ’Active’ minus ’Break’ composite wind anomalies (ms−1 ) and associated relative vorticity (10−6 s−1 ) at 850 hPa during the 40 year period (1954-1993). Only vectors significant at 95% confidence level are displayed. Positive contours are shaded and negative contours are not shown. (b) ’Active’ minus ’Break’ composite precipitation anomalies (mm.day−1 ) during 1979-1993.
. . . . . . . . . . 94
Chapter 1
Introduction The seasonal mean summer monsoon precipitation over the Indian continent and neighbouring region is the lifeline of the agrarian economy of the region. A weak or late monsoon can have disastrous consequences on productivity of the crops upon which millions of people rely on for their sustenance [Swaminathan, 1987]. The monsoons, which returns with remarkable regularity each summer provides rainfall needed for 60% of World’s population. The importance of Asian summer monsoon in the global circulation and climate predictability is widely recognized. Monsoon displays substantial interannual variability, which has profound socio-economic consequences. Prediction of seasonal mean monsoon precipitation, therefore assumes great importance. Statistical prediction of some gross features of the Indian summer monsoon (e.g. all India monsoon rainfall, IMR) have been moderately successful during the decade of 80’s [Krishakumar et al., 1995], but have failed during the decade of 90’s. This is due to the fact that correlation between IMR and many predictors undergo low frequency decadal variation. The dynamical seasonal prediction of the Indian summer monsoon has thus far remained elusive [Brankovic and Palmer, 2000]. However, dynamical seasonal prediction of monsoons could be beneficial due to a variety of reasons. • Ensembles of forecast can potentially yield information on probability of ’strong’ and ’weak’ monsoon. • It can give more accurate information on regionality of anomalous rainfall and circulation compared to statistical methods. Hence, it is important to identify and understand the factors that may be limiting our current level of predictability. The predictability of the tropical climate (specially Indian summer monsoon), depends on the relative contribution of ’external’ slowly varying boundary forcing and
1 Introduction
2
’internal’ dynamics (intraseasonal oscillations) to the interannual variability [Charney and Shukla, 1981]. Following the seminal work of Charney and Shukla [1981] and Shukla [1981], during the past two decades, it was shown that climate in large part of the tropics is primarily determined by slowly varying sea surface temperature (SST) where potential for making dynamical forecasts several seasons in advance exists (e.g. Latif et al. [1998]). However, during the same period it has been also recognized that there are regions within the tropics, climate of which is not strongly governed by the anomalous boundary conditions. The Indian summer monsoon is such a system [Brankovic and Palmer, 1997; Webster et al., 1998; Goswami, 1998]. What limits the simulation and predictability of the Indian summer monsoon? Research during that past decade has identified two possible explanations. The first is that model errors in the mean monsoon simulations are still substantial enough that the signal being sought is smaller than the systematic bias. Charney and Shukla [1981] suggested that low frequency boundary forcing (e.g. sea surface temperature) predisposes the monsoon system towards a dry or wet state. In other words anomalous boundary conditions may provide potential predictability. If this is true, model simulations should be able to capture interannual variability of the Indian summer monsoon and hence could produce fairly good forecasts. But in reality, this is not the case as most models find the simulation of mean monsoon precipitation extremely difficult and have even greater difficulty in simulating the interannual variability of the Indian summer monsoon rainfall [Sperber and Palmer, 1996; Gadgil and Sajani, 1998; Goswami, 1998]. If the ’external’ slowly varying boundary forcing (e.g. sea surface temperature, soil moisture etc) determine the predictability of monsoons, there is a clear need to improve the model simulations, before any conclusive statements could be made about the dynamical seasonal predictability of the Indian summer monsoon. The second explanation involves the role of intraseasonal variability and the suggestion that it introduces a chaotic element into the prediction of seasonal mean anomalies. During the established phase of the monsoon, circulation pattern undergoes significant variations associated with a pronounced northward excursion of the tropical convergence zone (TCZ) which brings the monsoon intermittently from an ’active’ into an inactive (’break’) phase over the continent. The change in precipitation distribution between ’active’ and ’break’ phases of monsoon is substantial [Webster et al., 1998] and it is therefore quite possible that intraseasonal variability could have a significant influence on the seasonal mean monsoon precipitation.
1 Introduction
3
The Indian summer monsoon has vigorous intraseasonal oscillations in the form of ’active’ and weak (or ’break’) spells of monsoon rainfall within the summer monsoon season [Ramamurthy, 1969]. These ’active’ and ’break’ spells of the monsoon are associated with fluctuations of the tropical convergence zone (TCZ) [Yasunari, 1979, 1980, 1981; Sikka and Gadgil, 1980]. The TCZ over the Indian monsoon region represents the ascending branch of the regional Hadley circulation. Intraseasonal oscillations (ISOs) are essentially manifestation of fluctuations of the regional Hadley circulation. These fluctuations initially seen in Indian station data [Keshavamurthy, 1973; Dakshinamurthy and Keshavamurthy, 1976] were later shown to be related to coherent fluctuations of the regional Hadley circulation [Krishnamurti and Subrahmanyam, 1982; Murakami et al., 1984; Mehta and Krishnamurti, 1988; Hartmann and Michelson, 1989]. The ISOs of the Indian summer monsoon have two preferred bands of periods [Krishnamurti and Bhalme, 1976; Krishnamurti and Ardunay, 1980; Yasunari, 1980]. One band has periods between 10 and 20 days while the other band contains periods between 30 and 60 days. The 30-60 day mode has a northward and eastward propagation over the monsoon region while the 10-20 day mode has a clear westward propagation and a weak northward propagation. The seasonal summer mean monsoon precipitation (and associated circulation) is a result of the shift of the seasonal mean position of the TCZ to about 25◦ N during boreal summer from a mean position south of the equator during boreal winter. The seasonal summer mean (June-September, JJAS) precipitation distribution has a major zone of large precipitation along the monsoon trough extending to the north Bay-ofBengal. There is also a secondary zone of precipitation maximum south of the equator (between 0◦ and 10◦ S) over the warm waters of the Indian Ocean. These two maxima in seasonal mean precipitation represent two favored locations of the TCZ during the summer monsoon season [Sikka and Gadgil, 1980; Goswami, 1994]. The ISOs are fluctuations of the TCZ between the two favored locations within the monsoon season. In the intraseasonal time scales, the TCZ form repeatedly over the ocean and moves northward, persists for a while over the monsoon trough before decaying and regenerating over the ocean. The tendency of the TCZ to persist over the monsoon trough results in larger residence time over the continent leading to larger seasonal mean precipitation over the land and a weaker one over the ocean. Therefore, there is a possibility that the statistics of the ISOs influence the seasonal mean monsoon. If the ISOs indeed influence the seasonal mean significantly, the part of the seasonal mean governed by the ISOs would be unpredictable as ISOs are basically governed by internal dynamics [Webster,
1 Introduction
4
1983; Goswami and Shukla, 1984; Keshavamurthy et al., 1986] and are chaotic in nature. If the ISOs do contribute significantly to the seasonal mean, the interannual variability of the seasonal mean monsoon is expected to have a significant component arising from internal dynamics. Several recent modeling studies show that indeed a significant fraction of the interannual variability of the Indian summer monsoon may be governed by internal dynamics [Harzallah and Sadourny, 1995; Rowell et al., 1995; Stern and Miyakoda, 1995; Goswami, 1998]. Most of the studies do not provide any insight regarding the origin of the internally generated interannual variability. Based on a series of sensitivity studies with a GCM and a dynamical system model, Goswami [1997] indicates that the modulation of the energetic intraseasonal oscillations by the annual cycle could give rise to an internal quasi-biennial oscillation in the tropical atmosphere. These arguments and the modeling studies set the stage to ask the question: Do the ISOs really influence the seasonal mean monsoon? If so, how and to what extent? Unfortunately, how and to what extent the ISOs influence the seasonal mean circulation and precipitation has not been clearly established from observations. Not many studies have actually addressed this question. Mehta and Krishnamurti [1988] examined the interannual variability of the 30-50 day mode in the winds at 850 hPa and 200 hPa for the period 1980 to 1984 using European Center for Medium Range Weather Forecasts (ECMWF) operational analysis. They mainly examined the variations in the northward propagation characteristics and did not attempt to relate these to the seasonal mean. Hartmann and Michelson [1989] used 70 year (1901-1970) record of daily precipitation for 3700 stations distributed over whole India and created annual cycle of daily precipitation at 1◦ × 1◦ blocks. They find statistically significant peak in the daily precipitation around 40-50 day period over most of India south of 23◦ N and the oscillations have a northward propagation. This study did not address the interannual variability of the ISOs. Singh and Kriplani [1990] and Singh et al. [1992] used long records of daily rainfall data over the Indian continent and examined the 30-50 day oscillation. They found that ISOs has largest amplitude over the western central India around 20◦ N and can explain upto 25% of 5-day averaged rainfall. They concluded that these oscillations have large interannual variability in intensity and period and does not seem to be related with overall performance of monsoon or phases of ENSO. They, however, could not come to a clear conclusion regarding relationship between the ISOs and the interannual variability of the Indian monsoon rainfall. Rao et al. [1990] used daily IR data from INSAT-1B for the monsoon period of 1986 and 1987. After creating daily averages from three hourly
1 Introduction
5
observations, they calculated fractional cloud cover in each 2.5◦ ×2.5◦ boxes from the IR brightness temperature. The fractional cloudiness shows a periodicity of 30-50 days in both years up to 20◦ N. Based on only two years, it was not possible to conclude much on the interannual variability of this mode. Ahlquist et al. [1990] studied radiosonde observations at 12 Indian stations between 1951 and 1978. They examined ISOs with period longer than 10 days but did not try to relate the ISOs with the interannual variability of the monsoon. De and Natu [1994] examined upper wind data for six radiosonde stations at 850, 700, 500 and 300 hPa levels for the years 1979 to 1984 and 1987. They also find that 30-50 day mode has considerable interannual variability and that the mode becomes more significant during normal and excess rainfall years. Here, the sample size is small to arrive at a robust result regarding interannual variability. Kondragunta [1990] used daily OLR for the summer period form NOAA polar orbiting satellite for eight years (1975 to 1983) and studied the interannual variability of the ISO over the whole Asian region. He finds that intraseasonal oscillations occur on three time scales, 30-60 day, 10-20 day and less than 10 days. Fennessy and Shukla [1994] using GCM simulations of the Indian monsoon for 1988 and 1987 showed that the spatial structures of the interannual variability of the seasonal mean and that of the intraseasonal variability in their model simulations were quite similar. Ferranti et al. [1997] studied the relationship between intraseasonal and interannual variability over the monsoon region using data from five 10-year simulations of the ECMWF GCM differing only in their initial conditions. They examined simulated precipitation and 850 hPa relative vorticity in detail and showed that monsoon fluctuations within a season and between different years have a common mode of variability with a bi-modal meridional structure in the precipitation. While the common mode of variability is qualitatively consistent with fluctuations of the TCZ in the two favored locations, their results suffer from some systematic errors inherent in the ECMWF GCM simulation of the Indian summer monsoon. The model underestimates precipitation over the north Bay of Bengal and the monsoon trough zone. This systematic error reflects in their interannual mode having appreciable amplitude only east of 80◦ E both in precipitation and low-level vorticity. Webster et al. [1998] discusses mean circulation pattern at 850 hPa associated with ’active’ and ’break’ conditions based on ECMWF operational analysis for 14-year (1980-1993) and brings out the large scale nature of these circulation anomalies. No attempt to relate these patterns to the seasonal mean was, however, made. In another recent study Goswami et al. [1998] studied daily surface winds from National Centers for Environmental Prediction
1 Introduction
6
(NCEP)/National Center for Atmospheric Research (NCAR) reanalysis for ten years (1987-1996) and showed that the spatial structures of the intraseasonal mode and that of the dominant interannual mode are strikingly similar. Annamalai et al. [1999] examined the relationship between the intraseasonal oscillations and interannual variability using NCEP/NCAR reanalysis and ECMWF reanalysis (ERA) for the period 1979-95. While the primary objective of the study was to compare NCEP/NCAR reanalysis and ERA, they also identified a dominant mode of intraseasonal variability which captures the active/break cycles of the monsoon. However, they could not clearly identify a common dominant mode that described intraseasonal and interannual variability. They have tried to find the relationship between interannual and intraseasonal variability by using a one dimensional probability density function (PDF) of the principal component of the dominant ISO mode. Clear difference in probability of occurrence of ’active’/’break’ phases in two contrasting years also could not be identified. Statistical significance of results could not be ascertained due to small sample size. Sperber et al. [2000] investigated the relationship between the relationship between subseasonal and interannual variability of the Asian summer monsoon using 40 years of NCEP/NCAR Reanalysis. They have confirmed that a common mode of variability exists on subseasonal and interannual time scales. PDF of principal components did not show any bimodality. Further they have shown that PDF is systematically and significantly perturbed towards negative (positive) values in weak (strong) monsoon years. However, they also mention that only a subset of subseasonal modes are systematically perturbed either by ENSO or in weak/strong monsoon years, suggesting that predictability is likely to be limited by the chaotic, internal variability of the monsoon system. The PDF of the subseasonal modes are biased towards positive (negative) side during strong (weak) years only if the low frequency interannual variations of the seasonal mean are not removed. Krishnamurthy and Shukla [2000] has used gridded rainfall dataset (1901-1970) to analyze the intraseasonal and interannual variability of the summer monsoon rainfall over India. They have found that the nature of intraseasonal variability is not different during the years of major droughts or major floods. They have also found that there is considerable variability in the spatial patterns of the rainfall anomalies over India on both daily and seasonal time scales. Their results indicate that the dominant mode (leading EOF) of the daily rainfall anomalies has a spatial pattern different from the dominant
1 Introduction
7
mode of seasonal anomalies. The variances of the daily rainfall anomalies over India are about 50-100 times larger than those of the seasonal rainfall anomalies. To relate intraseasonal and interannual variability they have used a correlation analysis between the daily and the seasonal anomalies and have found that there is a signature of the seasonal anomaly pattern throughout the monsoon season. The frequency distributions of the correlations involving daily anomalies that include the seasonal anomalies clearly show a bias toward positive correlations and do not reveal any bimodality. The frequency distribution do not show any bias, if the low frequency interannual variations of the seasonal mean are removed. Recently, Lawrence and Webster [2001] have examined the interannual variations of the ISO using out-going long wave radiation (OLR) data for the period 1975-1997. By developing an index representing seasonally averaged ISO activity, they have found that summertime ISO activity exhibits an inverse relationship with Indian monsoon strength. They concluded that the ISO activity is uncorrelated with any other leading SST variability including the ENSO. A conceptual model of how the ISOs influence the seasonal mean and interannual variability of the Indian monsoon was proposed by Goswami [1994]. The conceptual model is based on the similarity between the spatial structure of the dominant ISO mode and that of the interannual variability. The seasonal summer mean (June-September, JJAS) precipitation distribution has a major zone of large precipitation along the monsoon trough extending to the north Bay-of-Bengal (see Figure 2.1(d)) and a secondary zone of precipitation maximum south of the equator (between 0◦ and 10◦ S) over the warm waters of the Indian Ocean. These two maxima in the seasonal mean precipitation represent two favored locations of the TCZ during the summer monsoon season [Sikka and Gadgil, 1980; Goswami, 1994]. The ISOs are fluctuations of the TCZ between the two locations and repeated propagation from the southern to the northern position within the monsoon season. During a typical ’active’ condition, the northern TCZ is stronger and the southern one is weaker with stronger cyclonic vorticity and enhanced convection over the northern location with stronger anticyclonic vorticity and decreased convection over the southern one. The situation reverses during a typical ’break’ condition. Higher probability of occurrence of ’active’ like (’break’ like) conditions during a monsoon season could, therefore give rise to stronger (weaker) than normal seasonal mean monsoon and precipitation. It may be noted that the ISOs are not purely sinusoidal oscillations. Due to the broadband nature of their spectrum, the intensity as well as the
1.1 Objectives
8
duration of the ’active’ phases in a season could be different from those of the ’break’ phases. Moreover, the number of ’active’ and ’break’ spells within a monsoon season (June 1 - September 30) may be different depending on the initial phase. These factors may lead to asymmetry in the probability density functions (PDF). Our conceptual model is similar, to the one proposed by Palmer [1994]. However, in contrast to Palmer [1994] who proposes that the asymmetry in the PDF is forced only by external forcing, we claim that the asymmetry could arise even without external forcing.
1.1
Objectives
The background presented above lead us to the following conclusions. Low-frequency (LF) large amplitude ISO with period around 10-20 days and 30-60 days are integral part of Indian summer monsoon. Therefore, the phase, amplitude and period of these ISOs can influence the seasonal mean monsoon. ISOs are driven by internal dynamics, involving primarily feed back between organized convection and dynamics. A conceptual picture of the variability of these LF oscillations envisages competition between the continental TCZ and oceanic TCZ. While the considerations presented above are all plausible, there has been no reliable quantitative estimate of how and to what extent ISO influence the seasonal mean and it’s variability. As a result, the present study is undertaken with the following specific objectives. • The primary objective of the present study is to use sufficiently long daily observation to bring out how and to what extent the ISOs of the Indian monsoon affect the seasonal mean and its interannual variability. The conceptual model proposed above is used as the working hypothesis. The primary objective may be achieved in two parts. Firstly, we bring out the underlying common spatial structure of the dominant ISO in all years and compare it with the spatial structure of the seasonal mean and interannual variability of the Indian monsoon. Secondly, attempt is made to relate probability of occurrence of the ISO pattern to the interannual variability of the seasonal mean. To achieve this goal a homogeneous data set for a long enough period is essential so that the statistics of the ISOs and that of the interannual variability of the seasonal mean could be reliably estimated. Many earlier studies used data only for a short period as a result of which the interannual variability of the circulation could not be reliably estimated. (Chapter 2 and
1.2 Datasets
9
Chapter 3) • Having shown that the ISOs can influence the seasonal mean and its interannual variability, the next objective is to make quantitative estimates of predictability of the monsoon climate. A measure of potential predictability of the monthly and seasonal means at a place could be obtained from the ratio of variances associated with the external to the internal components. Using long homogeneous data sets, attempts will be made to estimate the ’internal’ variability of monthly and seasonal climate. The potential predictability of the Indian monsoon region will be compared with that of other regions in the tropics. (Chapter4) • The ISOs of the monsoon lead to strengthening (weakening) of the seasonal mean monsoon in their active (break) phase. While this fact results in interannual variations of the seasonal mean monsoon at one end of the spectrum, it may modulate the statistics of the monsoon synoptic disturbances at another end. The main rain bearing system during the monsoon season are Low Pressure Systems (LPS) consisting of lows and depressions. Since the genesis of the LPS depends on the horizontal shear and low-level vorticity, it is possible that more LPS may form in active phase relative to the break phase. In other words, large scale circulation associated with the ISOs could modulate the frequency of genesis of LPS. Therefore one objective of our study will be to investigate how the synoptic events are modulated by the ISOs. (Chapter 5)
1.2
Datasets
A brief description of the different datasets used in the study is provided below.
1.2.1
Reanalysis Data
The National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) 40-year Reanalysis data is a research quality data set suitable for weather and short-term climate research. The NCEP/NCAR Reanalysis Project uses a Global Data Assimilation System (GDAS), along with the observations from 1957 to the present to produce global meteorological fields through dynamically and thermodynamically consistent interpolations to support the needs of the climate research community [Kalnay et al., 1996]. The project began in 1991 and involves the recovery and quality control of historical land surface, ship, rawinsonde, aircraft, pibal, satellite and
1.2 Datasets
10
other data. These data are then assimilated with a GDAS that is kept unchanged over the reanalysis period 1957-96, to avoid spurious climate jumps or trends. The project uses a frozen state-of-the-art global data assimilation system and a data base as complete as possible. The model used here is a T62 model (equivalent to a horizontal resolution of 210 km) with 28 vertical levels. Thus, any output variable in the reanalysis is a blend of observations and model. The fidelity of any variable to reality depends on the accuracy and density of observations as well as on the performance of the analysis scheme itself. The reliability of the parameters have been increased with the addition of delayed observations, provided by different countries and organizations. Output variables are classified into four categories; A, B, C and D; depending on the relative influence of the observations and/or the model. A variable belongs to category ‘A’ (e.g. wind, upper-air temperature) if it is strongly influenced by observed data and, hence, reliable. The designation ‘B’ (e.g. SST, air temperature at 2 m, specific humidity at 2 m, relative humidity) indicates that although the variable is directly affected by observational data, the model also has a strong influence on it. A category ‘C’ variable (e.g. cloud, precipitation, latent heat flux, sensible heat flux) indicates that there are no observations directly affecting the variable, so it is solely derived from the model. The letter ‘D’ (e.g. ice concentration, plant resistance, land sea mask) represents a field that is fixed from climatological values and does not depend on model. This study uses daily averaged zonal (u) and meridional components (v) of reanalysis winds at various pressure levels (surface, 850 hPa, 500 hPa and 200 hPa and geopotential height at 700 hPa). To accurately estimate interannual variability, we have also used monthly mean winds at the same levels for a 42-year period (1956-1997).
1.2.2
NOAA Outgoing Long wave Radiation (OLR) Dataset
One proxy for tropical rainfall measurements or deep tropical convection is the OLR data at the top of the atmosphere. Twenty four years (1974-1997) years of daily averaged OLR estimates from NOAA satellites [Salby et al., 1991; Gruber and Krueger, 1984] were used in this study. Data gap periods (January 1974 to May 1974 and March 1978 to December 1978) were avoided. Missing values were owing to satellite problems, archival problems or incomplete global coverage.These satellites uses AVHRR (Advanced Very High Resolution Radiometer) which has unique characteristics of spectral response, image geometry, frequency of spectral coverage and accessibility that makes it useful for applications in oceanography and meteorology. OLR data are available in (2.5◦ × 2.5◦ )
1.2 Datasets
11
latitude-longitude grid boxes. Daily interpolated OLR data set (2.5◦ × 2.5◦ ) for the same period (1974-1997) were also used for this study. This data set is taken from NOAA-CIRES Climate Diagnostics Center (CDC), Boulder, USA, from their website at http://www.cdc.noaa.gov/. Data gaps were filled with temporal and spatial interpolations; details of the interpolation technique can be found in Liebmann and Smith [1996].
1.2.3
Precipitation Datasets
Monthly, pentad and daily precipitation datasets were used to substantiate some of the results presented in this study. For monthly precipitation, Climate Prediction Center Merged Analysis of Precipitation (CMAP) data [Xie and Arkin, 1997] were used. CMAP is a gridded global monthly precipitation constructed on a 2.5◦ latitude-longitude grid for a 17-year period from 1979 to 1995 by merging several kinds of information sources with different characteristics. The sources include guage-observations, estimates of precipitation inferred from a variety of satellite observations. Satellite estimates come from infrared as well as microwave sensors. First, the satellite estimates are combined using a weighted average where weights are proportional to the estimated errors of the various estimates. The weighted average is then merged with the guage observations. CMAP provides very useful information for climate analysis and can be used to investigate annual and interannual variability in large scale precipitation. CMAP pentad data for a 15-year period from 1979-1993 were also used. The pentad CMAP essentially uses the same algorithm and input data as monthly CMAP [Xie and Arkin, 1997]. The version we use is based on a blend of guage data with satellite products including the GOES (Geostationary Operational Environmental Satellite) precipitation index based on geostationary infrared data, Microwave Sounding Unit, OLR based precipitation index, SSM/I (Special Sensor Microwave/Imager) scattering and SSM/I emission. A detailed description of the pentad dataset is in preparation. Daily gridded rainfall data over the Indian continent for a 12-year period from 1978 to 1989 was also utilized. The daily rainfall data was originally compiled by Singh et al. [1992] at 2.5◦ latitude-longitude grids based on daily rainfall at 365 stations uniformly distributed over the country. The original data reported in Singh et al. [1992] were later extended to 1989. The version we use was regridded to 1.25◦ × 1.25◦ latitude-longitude boxes by Mike Fennessy of the Center for Ocean-Land-Atmosphere-Studies (COLA, 1999, personal communication).
1.2 Datasets
1.2.4
12
Statistics of Low Pressure Systems
The dates and location of genesis of all lows, depressions and cyclones during AprilNovember for a 40-year period from 1954 to 1993 over the Indian monsoon region (50◦ E100◦ E, Eq-35◦ N) were collected from various sources. For the Indian monsoon region, data were based on reports from the India Meteorological Department (IMD). Data for the first 30 years were taken from Mooley and Shukla’s [Mooley and Shukla, 1987, 1989] compilation based on IMD’s daily weather reports. Data for the next 10 years (19841993) were compiled from seasonal weather summaries published by IMD in Mausam. For example, data for 1984 monsoon season can be found from [IMD, 1985]. For categorizing ’strong’ and ’weak’ monsoons we have used the All India Monsoon Rainfall index [Parthasarathy et al., 1994]. The IMR is constructed from a weighted average of 306 stations spread over the whole of the Indian subcontinent.
Chapter 2
Basic Characteristics of Monsoon Intraseasonal Oscillations In this chapter, the basic characteristics of intraseasonal oscillations of the Indian summer monsoon are examined. The characteristics of monsoon ISOs such as their horizontal and vertical structures and meridional and zonal propagation characteristics have been previously studied extensively (see references cited in the Introduction). Our objective here is not to repeat the results of the earlier studies. However, earlier studies used limited number of years. As a result, it is not well established whether different phases of the dominant ISO mode possess spatial patterns that are common to all events. Our aim here is to bring out the underlying mean feature of the dominant ISO mode that is invariant over the years. We have examined these characteristics of both ISO modes for each year of the 20-year period (1978-1997). The general characteristics of the 30-60 day mode and 10-20 day are consistent with most of the earlier studies. Therefore, some of the important features are only briefly summarized here.
2.1
Methodology
The climatological summer mean (June-September, JJAS) circulation at lower and upper atmosphere and associated mean low-level vorticity are shown in Figure 2.1(a,b,c). The seasonal mean precipitation is shown in Figure 2.1(d). It may be noted that the negative mean vorticity between the equator and 10◦ S (Figure 2.1(b)) represents a region of cyclonic vorticity in the Southern Hemisphere and is coincident with the seasonal precipitation maximum. The circulation, convection and precipitation in the monsoon region are characterized by a strong seasonal cycle. An example of zonal winds at 850 hPa at a few selected points for 1990 is shown in Figure 2.2. The annual cycle is defined by the sum of the annual and semi-annual harmonics (green solid lines in Figure 2.2).
2.1 Methodology
14
Figure 2.1: Climatological mean (JJAS) monsoon winds (ms−1 ) and precipitation (mm.day−1 ). (a) 850 hPa vector winds, (b) Relative vorticity at 850 hPa (10−6 s−1 ), (c) 200 hPa vector winds, (d) Precipitation from Xie and Arkin [1997].
The daily anomalies after removing the annual cycle are shown in the right panel. The annual cycle, which is essentially driven by external conditions, has year to year variations that manifest in the interannual variations of the seasonal mean. In many studies, daily anomalies are constructed by removing the climatological mean for each day from the daily observations. In a particular year, the annual cycle may be significantly different from the climatological mean annual cycle. This would introduce an additional bias in the daily anomalies during the monsoon season. This bias can give rise to asymmetry in the PDF of the ISOs that may not be intrinsic to the ISOs but may be related to the external forcing changes. Since we are interested in the role of intraseasonal oscillations in modifying the summer mean, we would like to avoid aliasing of any statistics of the ISOs due to possible year to year variation of the annual cycle itself. This is achieved by calculating the annual cycle for each year based on the data for that year alone and by calculating the daily anomalies after removing the annual cycle of each year. The intraseasonal oscillations are identified by estimating the spectra of zonal and
2.1 Methodology
15
Figure 2.2: Some examples of raw time series of zonal winds at 850 hPa at a few selected points during 1990. (Left panels) Daily zonal winds (ms−1 ) with the annual cycle (annual and semiannual harmonics, green lines). (Right panels) Anomalous daily zonal winds (ms−1 ).
Zonal wind 1.5 1 0.5 0
0
50
100
Period (days)
150
Power * Frequency
Power * Frequency
2 150
OLR
100
50
0
0
50
100
150
Period (days)
Figure 2.3: Examples of spectra of zonal winds and OLR for a typical year (1984) at a typical point (90◦ E, 10◦ N).
meridional winds as well as OLR anomalies. Power spectra are calculated from anomaly time series between May 1 and October 31 (184 days) using Tukey lag window method [Chatfield, 1980]. An example of spectra for zonal winds and OLR at a point in north Bay-of-Bengal for 1984 is shown in Figure 2.3. This example shows two strong peaks, one with period around 36 days and the other with period around 16 days. Similar, power spectral estimates are made for each year and at all latitudes between 30◦ S and 30◦ N along a number of longitudes (e.g. 70◦ E, 80◦ E, 90◦ E). From these estimates, the most prevalent dominant periods are chosen. It is noted that the dominant periods
2.2 Propagation Characteristics
16
Year
Mode I
Mode II
1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997
25 42 33 42 42 25 36 33 42 30 42 40 42 42 34 32 30 42 32 42
12 17 13 16 14 14 16 12 12 10 20 14 16 20 14 14 12 20 12 12
Table 2.1: Period in days corresponding to the two peaks in the spectra for different years
found from the winds agree well with those found from OLR. The dominant periods found in each year of the 20-year period (1978-1997) are listed in Table-I. The dominant periodicity in each of the two bands show considerable variation from one year to another. To study the detailed structure and characteristics of the two intraseasonal oscillations Butterworth band pass filter [Murakami, 1979] with peak response around the dominant periods are used.
2.2
Propagation Characteristics
The 30-60 day mode has a large horizontal scale (half wavelength of 70◦ -80◦ longitude) as seen from the point-correlation map of the 30-60 day filtered zonal winds with respect to those at a reference point (85◦ E, 10◦ N; Figure 2.4(a)). The mode has a first baroclinic vertical structure close to the equator and over the Indian monsoon region as seen from correlations between 30-60 day filtered zonal winds at 850 hPa and 200 hPa (Figure 2.4(b)). The horizontal scale and vertical structure of the mode shown in the example (Figure 2.4) is representative of other years. The 30-60 day mode is
2.2 Propagation Characteristics
17
Figure 2.4: An example illustrating the horizontal scale and vertical structure of the dominant ISO mode. (a) Lag-zero correlations of the 850 hPa 30-60 day filtered zonal winds with respect to a reference point (85◦ E, 10◦ N). (b) Lag-zero correlations between 30-60 day filtered zonal winds at 850 hPa and 200 hPa at each grid point. Correlations are calculated between May 1 and October 31 of 1990. Correlations exceeding 0.2 are significant at 95% confidence level. known to have a northward and eastward propagation in the Indian monsoon region [Yasunari, 1979, 1980]. This is demonstrated in Figure 2.5, where lag correlation of the 30-60 day filtered zonal winds at 850 hPa and OLR with respect to the same fields at a reference point (85◦ E, 10◦ N) for the year 1990 averaged over a longitude belt (80◦ E90◦ E) are plotted as a function of latitude. Northward propagation north of the equator and a tendency for southward propagation south of the equator are seen in zonal winds as well as in OLR. Webster et al. [1998] also referred to such northward propagation in the northern hemisphere and southward propagation in the southern hemisphere of the dominant ISO. The correlation represents an average picture of the 3-4 episodes of the oscillation during the summer monsoon season. On individual episodes, it is seen that not all episodes are associated with a clean northward propagation (not shown). The character of the northward propagation (e.g. speed of propagation) also vary from one year to another. On the average the 30-60 day mode has an eastward propagation in the monsoon region (50◦ E-110◦ E). As in the case of northward propagation, the eastward propagation may not be clear in each individual episode and has considerable year to
2.2 Propagation Characteristics
18
Figure 2.5: (a) Correlations between 30-60 day filtered zonal winds at 850 hPa with respect to that at a reference point (85◦ E, 10◦ N) at different lead/lags averaged over (80◦ E-90◦ E) for 1990. (b) Same as (a) but for 30-60 day filtered OLR. Starting contour is ±0.1 and contour interval is 0.2.
Figure 2.6: (a) Correlations between 30-60 day filtered zonal winds at 850 hPa with respect to that at a reference point (85◦ E, 10◦ N) at different lead/lags averaged over (10◦ N-20◦ N) for 1990. (b) Same as (a) but for 30-60 day filtered OLR. Starting contour is ±0.1 and contour interval is 0.2.
2.3 A Circulation Criterion for ’Active’ and ’Break’ Phases
19
year variability (Figure 2.6). The 10-20 day mode on the average has clear westward propagation in the monsoon region. It is either stationary or northward propagating in the meridional direction (figure not shown).
2.3
A Circulation Criterion for ’Active’ and ’Break’ Phases
Traditionally ’active’ and ’break’ monsoon conditions are defined based on a precipitation criterion [Ramamurthy, 1969]. Here, we propose a criterion to define ’active’ and ’break’ monsoon conditions based on a circulation index. Such a circulation based definition of ’active’ and ’break’ monsoon may be useful for various purposes. During an ’active’ phase of the Indian monsoon, typically there is more precipitation over central India and a stronger monsoon trough [Ramamurthy, 1969]. As a result we may expect westerly zonal winds south of the monsoon trough to strengthen. Opposite is expected during a ’break’ phase. With this consideration in mind, we propose a circulation based definition of ’active’ and ’break’ monsoon conditions. A reference point just south of the monsoon trough at (90◦ E, 15◦ N) is selected for this purpose and the 30-60 day filtered zonal winds at 850 hPa are plotted (Figure 2.7(a)). The days for which the filtered zonal winds at 850 hPa is greater than +1 standard deviation (as shown by the thin solid line, i.e. stronger westerly anomalies) are considered as ’active’ days, while those for which it is less than -1 standard deviation (i.e. stronger easterly anomalies) are considered as ’break’ days. The method of defining ’active’ and ’break’ conditions is somewhat similar to the one used by Webster et al. [1998] where they also used a zonal wind criterion over the north Bay-of-Bengal but used a fixed cut off anomaly (+3 ms−1 or -3 ms−1 ) to define ’active’ and ’break’ conditions. Our method of defining ’active’ and ’break’ also bears similarity with the one used by Krishnamurti and Subrahmanyam [1982] for the year 1979 where they used filtered zonal winds at a point in ’Arabian Sea’ to define ’active’ and ’break’ episodes. The identification of the ’active’ and ’break’ days is not very sensitive to small changes in the position of the reference point. It can be noted that in Figure 2.7(a) between June 1 and September 30 of this particular year, there were two ’active’ and three ’break’ episodes. The ’active’ and ’break’ days are thus identified for all years. To test whether our criterion for defining ’active’ and ’break’ monsoon conditions is related to the traditional precipitation based criterion, we calculated daily precipitation composites for all ’active’ and ’break’ days defined by the circulation criterion for
2.3 A Circulation Criterion for ’Active’ and ’Break’ Phases
20
Figure 2.7: (a) An example of 30-60 day filtered zonal winds for 1986 at a reference point (90◦ E, 15◦ N). The thin horizontal lines correspond to +1 and -1 standard deviations. ’Active’ (’break’) days are defined as days for which the filtered zonal winds at the reference point are greater than +1 S.D (or less than -1 S.D). (b) 12-year (1978-1989) mean precipitation difference (mm.day−1 ) between all ’active’ and ’break’ composites. Contours are ±(1, 3, 5, 7, 9, 11, 13, 15).
the period 1978 and 1989 during June 1 to September 30. This is the period for which gridded daily rainfall data over India was available to us. The precipitation difference between ’active’ and ’break’ composites is shown in Figure 2.7(b). It is clear that the pattern of precipitation anomalies during ’active’ (’break’) conditions is identical to the dominant empirical orthogonal function (EOF) of daily (or pentad) rainfall [Singh and Kriplani, 1990; Krishnamurthy and Shukla, 2000] with an ’active’ monsoon condition being associated with enhancement of precipitation over most of continental India except a small region in south eastern India and another in north eastern corner. Thus, the ’active’ and ’break’ monsoon conditions defined by our circulation criterion captures the
2.4 Mean Structure of ISOs
21
dominant mode of intraseasonal precipitation variability over the Indian continent and hence are essentially same as those defined by traditional precipitation criterion. As the low-level jet over Somali is also usually strengthened (weakened) during an ’active’ (’break’) condition, one could also select a reference point in Arabian Sea (e.g. 60◦ E, 10◦ N) and 850 hPa zonal wind to define the ISOs.
2.4
Mean Structure of ISOs
Figure 2.8: (a,b) Climatological mean composite vector wind anomalies (ms−1 ) at 850 hPa corresponding to ’active’ and ’break’ conditions for the 30-60 day mode and (c,d) associated relative vorticity (10−6 s−1 ). The climatological mean composite is calculated by averaging all ’active’ and ’break’ conditions occurring during the 20-year period (1978-1997). Shading in the upper panels indicates regions with anomalies significant above 90% confidence level.
In this section, the underlying common spatial patterns associated with different phases of the dominant ISO modes are isolated. To obtain the mean spatial pattern common to all episodes of the dominant ISO variability, we use 20-year data of circulation and OLR (1978-1997). The phase composite technique [Murakami and Nakazawa,
2.4 Mean Structure of ISOs
22
1985; Murakami et al., 1984] is followed to illustrate the common mode of evolution of the oscillations. Having defined the ’active’ and ’break’ days as described in section 2.3, averaged vector wind anomalies at 850 hPa associated with all the ’active’ and ’break’ phases of the 30-60 day mode are calculated within a year. A climatological mean composite ’active’ phase constructed by averaging ’active’ composites of all 20 years is shown in Figure 2.8 together with the associated composite relative vorticity pattern. The composite of all ’break’ phases is also shown in Figure 2.8. These composites (means) and similar composites to be described later are tested for statistical significance using a Student t-test by using the inter-event variability as a measure of standard error. Level of statistical significance is noted in some of these figures. The significant coherent large wind anomalies that emerge after averaging over about eighty ’active’ (’break’) episodes over a period of twenty years, shows that all ’active’ (’break’)
Figure 2.9: Climatological mean composite vector wind anomalies (ms−1 ) corresponding to ’active’ and ’break’ conditions for the 30-60 day mode (a,b) at 500 hPa and (c,d) at 200 hPa. The climatological mean composite is calculated by averaging all ’active’ and ’break’ conditions occurring during the 20-year period (1978-1997). Shading indicates regions with anomalies significant above 90% confidence level.
2.4 Mean Structure of ISOs
23
phases possess a common spatial pattern of variability. The other important feature that emerges from the composite is the large zonal scale of the circulation changes associated with ’active’ (’break’) phases of the Indian monsoon extending from about 50◦ E to 120◦ E. During the ’active’ phase, the mean monsoon circulation is strengthened and the monsoon trough cyclonic vorticity is enhanced north of 10◦ N (compare with Figure 2.1(a,b)). The anticyclonic vorticity is enhanced between equator and 10◦ N and cyclonic vorticity is weakened in the southern hemisphere. The mean composite 850 hPa wind anomalies corresponding to ’active’ and ’break’ conditions (Figure 2.8) is consistent with the pattern shown in Webster et al. [1998]. The climatological mean composite ’active’ and ’break’ phase vector wind anomalies at 500 hPa and 200 hPa are shown in Figure 2.9. At 500 hPa during ’active’ phase, the vector wind anomalies bear close resemblance with those at 850 hPa with cross-equatorial flow and enhancement of monsoon trough vorticity. At 200 hPa, the vorticity anomalies over the monsoon trough have become anticyclonic and the equatorial wind anomalies are generally out of phase with those at 850 hPa consistent with a first baroclinic mode vertical structure for this mode. The composite picture of ’active’ and ’break’ conditions described above is consistent with a seesaw between the two favorable positions of the TCZ as mentioned in the Introduction. If this scenario is correct, there should be enhanced convection in the northern position and decreased convection in the southern position during an ’active’
Figure 2.10: Climatological mean composite OLR anomalies (Wm−2 ) corresponding to ’active’ and ’break’ conditions. ’Active’ and ’break’ composites are constructed using unfiltered OLR anomalies and the same ’active’ and ’break’ dates defined by 30-60 day filtered zonal wind anomalies as used in Figure 2.8. OLR anomalies above 5 Wm−2 are significant above 90% confidence level.
2.4 Mean Structure of ISOs
24
Figure 2.11: Climatological mean composite pressure vertical velocity anomalies (ω) at 500 hPa (hPas−1 ). Again the same ’active’ and ’break’ dates chosen from 30-60 day filtered zonal wind anomalies for the 20-year period (1978-1997) as used in Figure 2.8 and Figure 2.10 are used. phase while it should be the other way round during a ’break’ phase. Figure 2.10 supports this conjecture where composite of unfiltered OLR anomalies for all ’active’ and ’break’ days are plotted. The ’active’ and ’break’ days used in the composite are exactly the same days defined by the 30-60 day filtered zonal winds at the reference point as in the circulation composite. Coherence of the OLR anomalies averaged over 20 years of ’active’ and ’break’ conditions defined by the circulation criterion shows that there is a close relationship between circulation and convection associated with ’active’ and ’break’ conditions. A notable feature of the composites is that the meridional seesaw of the convection anomalies is consistent with the low-level vorticity anomalies. It is also worth noting that even after averaging over approximately eighty active (break) episodes, fluctuations of OLR anomalies up to ±15 Wm−2 is seen over the two preferred regions. This means that, notwithstanding some variation in the intensity and mean position of the TCZ from one ’active’ (’break’) episode to another, there exists a common mean position of the TCZ during a typical ’active’ (’break’) episode. During individual years it is not unusual to see ±25 Wm−2 OLR anomalies over either of the zones. To put the dynamical link between low-level cyclonic vorticity and convection on a stronger footing, climatological mean composites of unfiltered pressure vertical velocity (ω) anomalies at 500 hPa corresponding to the same ’active’ and ’break’ days over the full 20-year period were also constructed. It is seen in Figure 2.11 that, enhanced (decreased) convection seen in Figure 2.10 are clearly associated with upward (down-
2.4 Mean Structure of ISOs
25
ward) motion in both ’active’ and ’break conditions. Location and spatial pattern of the vertical velocity anomalies correspond well with those of the convection anomalies. Thus, ’active’ (’break’) conditions are associated with a seesaw of the anomalous regional Hadley circulation. To illustrate the evolutionary character of the circulation anomalies associated with the 30-60 day mode, composite vector wind anomalies at 850 hPa and associated relative vorticity corresponding to eight phases of evolution of the oscillation is shown in Figure 2.12 for the period (1979-1989). The phase-1 corresponds to days when the filtered zonal wind anomalies at the reference point (90◦ E, 15◦ N) is zero and increasing towards positive direction. If T is the period, other phases are progressively T/8 days apart. In this way phase-3 is our ’active’ phase while phase-7 is our ’break’ phase. The transition of the vector wind anomalies from ’active’ to ’break’ phase is clear from this figure. It is also seen that while the ’active’ (’break’) phase defined by us has the largest horizontal scale, largest wind anomalies in the low-level jet region over the Arabian Sea may occur about
1 4
period prior to our ’active’ (’break’) phase. The northward prop-
agation of the TCZ is also depicted clearly from the transition of vorticity in different phases. Similarly the evolutionary character of OLR for the 30-60 day mode corresponding to the eight phases is illustrated in Figure 2.13. The ’active’ and ’break’ days used in the composite are exactly the same days defined by the 30-60 day zonal winds at the reference point as in the circulation composite. Apart from illustrating the northward propagation of the convection zones, the relationship between convection and vorticity is revealed from these two figures (Figure 2.12 and Figure 2.13). The movement of the convection anomalies seems to be in phase with the movement of the vorticity anomalies. The wind anomalies in each phase appear to arise from a linear response to heating associated with OLR anomalies. For example, the strong south westerly anomaly in the Somalijet region in phase-2 is consistent with the fact that this phase is characterized by more OLR anomalies over the continent. Similarly more zonal winds in phase-3 is due to the fact that this phase is characterized by OLR anomalies over Bay of Bengal and South China Sea. The ’active’ and ’break’ composite are constructed for the 10-20 day mode following a similar procedure. ’Active’ (’break’) conditions are now defined by the 10-20 day filtered zonal winds at 850 hPa being greater than +1 standard deviation (less that -1 S.D) at the same reference point south of the monsoon trough. Due to it’s shorter period, it is possible to have 8-10 episodes of ’active’ or ’break’ conditions for this mode during the
2.4 Mean Structure of ISOs
26
summer monsoon season. The climatological mean ’active’, ’break’ composite vector wind anomalies for the 10-20 day mode based on the entire twenty year period at 850 hPa is shown in Figure 2.14 together with the corresponding relative vorticity. Most important feature of this mode is that it has a much smaller horizontal scale, confined mainly to the Bay-of-Bengal. ’Active’ (’break’) conditions are associated with a strong cyclonic (anticyclonic) vortex at the north Bay-of-Bengal with an anticyclonic (cyclonic) vortex south of it between 10◦ N and the equator. Due to the localized character of the 10-20 day mode, it is unlikely to have a strong influence on the large scale mean circulation. However, depending on the phase relationship between the two ISOs, the strong cyclonic (anticyclonic) vorticity over the north Bay-of-Bengal associated with the ’active’ (’break) phase of the 10-20 day mode can enhance (weaken) the cyclonic vorticity over the monsoon trough zone associated with the 30-60 day mode [Goswami et al., 1998]. In this manner, it can indirectly contribute to the mean monsoon circulation.
2.4 Mean Structure of ISOs
27
Figure 2.12: Climatological mean composite vector wind anomalies (ms−1 ) at 850 hPa and associated relative vorticity (10−6 s−1 ) corresponding to eight phases of evolution of the 30-60 day mode for the period 1979-1989. The phase-1 corresponds to the days when the filtered zonal wind anomalies at the reference point is zero and increasing toward positive values.
2.4 Mean Structure of ISOs
28
Figure 2.13: Climatological mean composite OLR anomalies (Wm−2 ) corresponding to eight phases of evolution of the 30-60 day mode for the period 1979-1997. Eight composite phases are constructed using unfiltered OLR anomalies and the same dates defined by 30-60 day filtered zonal winds as used in Figure 2.12.
2.4 Mean Structure of ISOs
29
Figure 2.14: (a,b) Climatological mean composite vector wind anomalies (ms−1 ) at 850 hPa corresponding to ’active’ and ’break’ conditions for the 10-20 day mode and (c,d) associated relative vorticity (10−6 s−1 ). The climatological mean composite is calculated by averaging all ’active’ and ’break’ conditions occurring during the 20-year period (1978-1997).
Figure 2.15: Meridional bimodality of spatial structure of the dominant ISO. (a) Scatter plot of daily 30-60 day filtered vorticity at 850 hPa (10−6 s−1 ) over a northern band (70◦ E-100◦ E, 12◦ N22◦ N) and a southern band (70◦ E-100◦ E, 5◦ S-10◦ N) during 1 June to 30 September for 19 years (1979-1997). (b) Scatter plot of 30-60 day filtered OLR anomalies (Wm−2 ) averaged over the northern TCZ (70◦ E-100◦ E, 12◦ N-22◦ N) and the southern TCZ (70◦ E-100◦ E, 0◦ -12◦ S) during 1 June to 30 September for 18 years (1979-1997, excluding 1994).
2.5 Meridional Bimodality of ISO Spatial Structure
2.5
30
Meridional Bimodality of ISO Spatial Structure
As we hypothesize that the basic period and northward propagation of the monsoon ISOs result from the competition between the two favored positions of the TCZ, the existence of meridional bimodality in low-level circulation and convection would vindicate our hypothesis. In section 2.4, we showed that the peak phase of the ISO (the ’active’ and ’break’ phases) are characterized by a meridional bimodal structure. In this section, we demonstrate that the meridional bimodality is characteristic not only of the peak phases but valid through the evolution of the ISO. The robustness of the meridional bimodality of the low-level vorticity is illustrated in Figure 2.15(a). In this figure, the daily filtered vorticity over a north band (12◦ N-22◦ N) and a south band (10◦ N-5◦ S) averaged between (70◦ E-100◦ E) during the northern summer is shown as a scatter diagram for 19 years. The southern belt is part of a larger zone of opposite vorticity. It is rather striking to note that, the vorticity over the two regions tend to be out of phase on most days. The correlation between the two time series shown in this figure supports this conclusion. Based on 19 years of daily values during summer season (June 1 to September 30, i.e. 122 x 19 days), this correlation is highly significant. One may argue that the southern favored position of the TCZ is between equator and 10◦ S rather than 5◦ S and 10◦ N. It may be noted that the whole region between 12◦ S and 10◦ N fluctuates with anticyclonic (cyclonic) vorticity corresponding to cyclonic (anticyclonic) vorticity in the monsoon trough zone (Figure 2.8(c,d)). Figure 2.15(b) illustrates that the convection over the two favored locations indeed tends to fluctuate out of phase with each other. Figure 2.15(b) shows a scatter diagram of OLR anomalies averaged over the two preferred locations during the northern summer (June 1 to September 30) for the 18 years (1979-1997, excluding 1994). The correlation between convection anomalies over the two locations while not very high is highly significant. The fact that there is a bimodality of convection over the two locations was also evident in the ’active’/’break’ composites (Figure 2.10). To understand the reason of the scatter in Figure 2.15(b), we examined ’active’ and ’break’ composites of OLR for individual years. It is found that (figure not shown) that bimodality of convection is clear in every individual year. However, there is some shift (east-west or north-south) in the location of the maximum OLR anomalies over the two bands from one year to another. As we have fixed the position of the two bands in plotting Figure 2.15(b), the non-stationarity of the band from one year to another gives rise to the scatter in the plot. The year 1994 was excluded from
2.5 Meridional Bimodality of ISO Spatial Structure
31
the set (1979-1997) as it was found to be anomalous in the sense that the positions of the northern and southern TCZ were appreciably different from the mean position used in this scatter diagram. In section 2.4, we pointed out that there is a close relationship between the lowlevel vorticity and convective activity associated with the ISOs. The strength of this relationship between the low-level relative vorticity and convection is illustrated in Figure 2.16(a,b). Here, we plot a scatter diagram of relative vorticity anomalies and OLR anomalies averaged over the northern position of the TCZ (85◦ E-95◦ E, 12◦ N-22◦ N) and the southern position of the TCZ (85◦ E-95◦ E, 0◦ -12◦ S). All 19-year data between May 1 to October 31 are used in these scatter plots. The relationship between the two is significantly negative over the northern position while it is significantly positive over the southern position of the TCZ. As the southern position of the TCZ falls in the southern hemisphere, both the relationship show that cyclonic (anticyclonic) low-level vorticity is significantly correlated with increase (decrease) of convective activity in both favorable locations of the TCZ.
Figure 2.16: (a) Scatter plot of 30-60 day filtered relative vorticity at 850 hPa (10−6 s−1 ) and OLR (Wm−2 ) anomalies averaged over a box (85◦ E-95◦ E, 12◦ N-22◦ N) of the northern TCZ during 1 May to 31 October for 19 years (1979-1997). (b) same as (a) but averaged over a box (85◦ E-95◦ E, 0◦ -12◦ S) of the southern TCZ.
2.6 Discussions and Conclusions
2.6
32
Discussions and Conclusions
In this chapter, the basic characteristics of intraseasonal oscillations of the Indian summer monsoon is examined. We present a conceptual model (chapter 1) to describe how the ISOs influence the seasonal monsoon. It envisages the ISO arising out of fluctuation of the tropical convergence zone (TCZ) between two favored regions, one over the monsoon trough (northern TCZ) and other over the equatorial warm waters (southern TCZ). In one extreme of the ISOs (’active’ phase), the TCZ resides over the northern position strengthening the seasonal mean monsoon circulation, enhancing cyclonic vorticity over the northern TCZ and enhancing convection (and precipitation) over that location while suppressing convection over the southern position. In the other extreme (’break’ phase) weakened large scale monsoon flow and weakened cyclonic vorticity over the northern position keeps the northern position clear of convection and helps enhance convection over the southern position. A higher probability of occurrence of ’active’ (’break’) conditions in a monsoon season could therefore result in a stronger (weaker) than normal seasonal mean monsoon. In order to bring out the influence of the ISOs on the seasonal mean, it is desirable to separate the externally forced component of the seasonal mean from the internally forced component. We expect the slowly varying external forcing to give rise to slow and persistent changes and manifest in the interannual variation of the annual cycle. Intraseasonal anomalies are constructed in our study by removing the annual cycle of individual years (sum of annual and semiannual harmonics) from the observations. In this manner, we have been able to separate the influence of the external forcing on the ISOs. We believe that this procedure is important in bringing out the intrinsic role of the ISOs. Some studies define ISO anomalies with respect to climatological daily mean as annual cycle and hence the ISOs contain the effect of interannual variations of the annual cycle. This may be one reason why results of some previous studies have been inconclusive. Our first objective is to find the mean large scale spatial pattern associated with the ISOs and compare them with that of the seasonal mean pattern. For this purpose we have evolved a ’circulation’ criterion for defining ’active’ and ’break’ monsoon conditions. Large scale structure of the mean circulation anomalies associated with the ’active’ and ’break’ conditions of the dominant ISO modes are then obtained by constructing composite of filtered 30-60 day or 10-20 day circulation anomalies at all points for
2.6 Discussions and Conclusions
33
all ’active’ and ’break’ days. Climatological mean of all such composites for individual years is then constructed representing a spatial pattern of the ’active’ and ’break’ phases that is invariant from year to year. Such climatological mean composites corresponding to a typical ’active’ (’break’) condition of the 30-60 day mode is associated with a general strengthening (weakening) of the large scale mean monsoon flow and strengthening (weakening) of the monsoon trough. It is rather interesting that the circulation changes are not confined only over the Indian region but extended all the way to east of 120◦ E (South China Sea). The enhanced low-level cyclonic (anticyclonic) vorticity in the northern TCZ during an ’active’ (a ’break’) condition is associated with enhanced (decreased) ascending motion leading to enhanced (decreased) convection over the northern TCZ and decreased (enhanced) ascending motion and decreased (enhanced) convection over the southern TCZ. In other words, the anomalous regional Hadley circulation has ascending motion over the northern TCZ and descending motion over the southern TCZ during an ’active’ condition while the reverse is the case during a ’break’ condition. A typical evolutionary cycle of the dominant ISO based on composite of circulation and convection for 20 years (1978-1997) is also constructed that show repeated northward propagation from the southern position to the northern position (monsoon trough). On an average sense, the meridional bimodality of the spatial structure of the peak phases of the dominant ISO is evident in the composites. We also show that there is a seesaw of low-level vorticity between the two preferred locations of the TCZ on a dayto-day basis. Whether it is the northern location or the southern location of the TCZ, low-level anomalous cyclonic vorticity is associated with enhanced convection while low-level anomalous anticyclonic vorticity is associated with decreased convection establishing a link between anomalous circulation and convection.
Chapter 3
Intraseasonal Oscillations and Interannual Variability of the Indian Summer Monsoon In this chapter, the relationship between the intraseasonal oscillations and the interannual variability of the seasonal mean Indian summer monsoon is investigated. An attempt is made to arrive at some reliable conclusions through a series of detailed investigation of various aspects of the problem. Daily anomalies in a particular year is defined as departure of observations from the annual cycle of that year and a Butterworth filter is used to isolate the ISO modes (see section 2.1). In order to include the effect of both the ISO modes and keeping in mind their interannual variations in their peak period, total intraseasonal activity is defined by a band pass filtered field with peak response at 35 days and half responses at 15 days and 80 days respectively. For all the calculations described below, these ISO filtered fields are used to bring out the relationship between ISO and interannual variability.
3.1
A Common Spatial Mode of Intraseasonal and Interannual Variability
In the previous chapter (section 2.4), it was shown that the large scale structure of the wind associated with the dominant ISO mode is quite similar to that of the seasonal mean wind, strengthening and weakening the large scale flow during it’s ’active’ and ’break’ phases respectively. This similarity between the structure of the intraseasonal variability and the seasonal mean flow provides the basis for our hypothesis that the ISOs could influence the seasonal mean and it’s interannual variability. In this section, we provide further evidence that the spatial structure of the intraseasonal variability and the interannual variability are similar. The geographical distribution of the intraseasonal activity and the interannual variability are compared in Figure 3.1. In this figure,
3.1 A Common Spatial Mode of Intraseasonal and Interannual Variability
35
the standard deviation of ISO filtered 850 hPa relative vorticity averaged over the 20year period (1978-1997) and interannual standard deviation of the seasonal mean (JJAS) relative vorticity based on the same 20-year period are shown. It may be noted that mean amplitude of intraseasonal activity of this field is two to three times larger than the interannual variation of the seasonal mean in most places. The similarity of the geographical distribution of intraseasonal variability and interannual variability of 850 hPa relative vorticity is noteworthy. The correlation between the two patterns is 0.64 over the monsoon region (50◦ E-100◦ E, 20◦ S-30◦ N). Both the patterns are characterized by strong activity in the two favored positions of the TCZ namely a northern position around the monsoon trough and a southern position between the equator and 10◦ S. Regions of higher intraseasonal activity are also regions of larger interannual variability. What we have shown so far (e.g. the composite, the similarity between the S.D of ISO and interannual variability of the seasonal mean) are only indicative of a common mode of variability. To bring out the common spatial pattern of intraseasonal and interannual variability, the following procedure is adopted. An EOF analysis of the ISO filtered 850 hPa winds from June 1 to September 30 for all 20 years (1978-1997) is carried out. The first EOF explaining 14.8% of the total intraseasonal variance and representing the dominant ISO mode is shown in Figure 3.2(a). The dominant interannual mode is obtained from an EOF analysis of seasonal mean (JJAS) 850 hPa winds for 40 years (1958-1997). The first EOF explaining 16.8% variance of interannual variability of the Indian summer monsoon is shown in Figure 3.2(b). That the interannual EOF1 represents dominant variability of the Indian summer monsoon is seen from the strong correlation between PC1 and IMR (r=0.62) shown in Figure 3.2(c). Although there are some minor differences, the similarity between the spatial patterns of the dominant ISO mode and the dominant interannual mode is noteworthy. The easterlies south of the equator, the cross-equatorial flow east of 50◦ E, the convergence of air mass from north-west and south-west over the Arabian Sea around 10◦ N, the monsoon trough, the anticyclonic vortex around 75◦ E, 5◦ N are all common in both the patterns. The cross equatorial flow near African coast around equator seen in the interannual mode (Figure 3.2(b)) is not seen in the intraseasonal mode (Figure 3.2(a)). This is partly due to the fact that the dominant ISO can not be entirely represented by a single EOF due to its northward propagating character. The second dominant ISO mode is strongly correlated with the first at a lag of about 10 days and has large loadings exactly in this region∗ . Therefore, ∗
The second ISO EOF is not shown here for brevity. But the reader can refer to Chapter 5, Figure 5.2
3.1 A Common Spatial Mode of Intraseasonal and Interannual Variability
36
a common spatial pattern governs both the ISOs and the interannual variability, thus linking the ISOs with the interannual variability of the Indian monsoon. While the ISOs may have a common mode of spatial variability with the interannual variations of the seasonal mean monsoon, they may not have appreciable influence on the later unless the interannual variations of the ISO activity are significant. Therefore, we estimate the amplitude of interannual variations of the ISO activity and compare it with the amplitude of interannual variability of the seasonal mean. The standard deviation of ISO filtered vorticity at 850 hPa and OLR between June 1 and September 30 is calculated each year at each grid point. The interannual standard deviation of this intraseasonal standard deviation of each year is calculated based on 20 years (19781997). The interannual standard deviation of seasonal mean (June-September) vorticity at 850 hPa and OLR are separately calculated. The ratio between the standard deviation of interannual variations of ISO activity and interannual variation of seasonal mean is shown in Figure 3.3. It is seen that magnitude and pattern of ratio is similar for both low-level vorticity and OLR. The equatorial belt (10◦ S-10◦ N) east of 100◦ E is characterized by a low ratio, as the interannual variations are stronger in this region. In most of the Indian monsoon regions the ratio ranges from 0.4 to 0.8. This indicates that the variations of the ISO activity could account for 20% to 60% of interannual variability of the seasonal mean in the Indian monsoon region. Thus, we can expect significant modulation of the seasonal mean monsoon by the ISOs.
3.1 A Common Spatial Mode of Intraseasonal and Interannual Variability
37
Figure 3.1: Geographical distribution of intraseasonal and interannual activity. (a) Mean standard deviation of ISO filtered relative vorticity (10−6 s−1 ) at 850 hPa during 1 June to 30 September for 20 years (1978-1997). (b) Interannual standard deviation of seasonal mean relative vorticity (JJAS, 10−6 s−1 ) based on the same 20 years.
3.1 A Common Spatial Mode of Intraseasonal and Interannual Variability
38
Figure 3.2: First EOF of the intraseasonal and interannual 850 hPa winds. (a) Intraseasonal EOFs are calculated with ISO filtered winds for the summer months (1 June to 30 September) for a period of 20 years (1978-1997). (b) Interannual EOFs are calculated with the seasonal mean (JJAS) winds for 40-year period (1958-1997). Units of vector loading are arbitrary. (c) Relation between IMR and interannual PC1. Filled bars indicate interannual PC1 and the unfilled bar represent IMR. Both time series are normalized by their own standard deviation. Correlation between the two time series is shown.
3.1 A Common Spatial Mode of Intraseasonal and Interannual Variability
39
Figure 3.3: Ratio between standard deviation of interannual variation of ISO activity and interannual variation of the seasonal mean. (a) Relative vorticity at 850 hPa. (b) OLR. Contours are (0.3, 0.4, 0.6, 0.8, 1.0).
3.2 Probability of ’Active’/’Break’ Conditions and Seasonal Mean Monsoon
3.2
40
Probability of ’Active’/’Break’ Conditions and Seasonal Mean Monsoon
The fact that spatial pattern associated with the dominant ISO mode has a large spatial scale similar to the interannual variability of the seasonal mean, provides the basis to the idea that ISOs could influence the seasonal mean. For example, a higher frequency of occurrence of ’active’ (’break’) conditions within the monsoon season could lead to strengthening (weakening) of the seasonal mean resulting in a strong (weak) monsoon. This essentially is the hypothesis proposed in our conceptual model (chapter1). To test this hypothesis, we calculate probability density functions of the ISOs corresponding to ’strong’ and ’weak’ monsoon years. ’Strong’ and ’weak’ monsoon years are objectively defined based on whether IMR is greater than 1 S.D or less than -1 S.D. To have enough sample of such years we use daily data between 1956 and 1997. This period contains seven strong years (1956, 1959, 1961, 1970, 1975, 1983 and 1988) and ten ’weak’ monsoon years (1965, 1966, 1968, 1972, 1974, 1979, 1982, 1985, 1986, 1987). As mentioned earlier, the spatial pattern of the ISOs involve a northward propagating component. As a result, the evolutionary character of the ’active’ and ’break’ conditions cannot be described by a single EOF. To estimate the PDF of the ISOs, therefore, it is necessary to include more than one EOF. In the present study, we estimate the PDF of the ISOs using at least two EOFs. To obtain the PDF for the ’strong’ (’weak’) years, daily ISO filtered 850 hPa vorticity between June 1 and September 30 for all ’strong’, all ’weak’ and ’all’ (all 20 years between 1978 and 1997 taken together) years are combined and an EOF analysis is carried out in each case using singular value decomposition technique [Nigam and Shen, 1994]. The first two EOFs in each case are shown in Figure 3.4. It may be noted that the first EOF in ’strong’ and ’weak’ cases for positive projection coefficients (PC’s) represent ’active’ like and ’break’ like conditions respectively. The PDF of the PC’s in the reduced phase space defined by the first two EOFs explaining 17% of the total variance of the ’strong’ years (21% for ’weak’ and 15% for ’all’ years) are obtained using a Gaussian kernel estimator [Kimoto and Ghil, 1993; Silverman, 1986] with a smoothing parameter large enough to detect multi-modality with statistical significance. The smoothing parameter (h) is selected from the minimum obtained from the least-squares cross validation technique [Kimoto and Ghil, 1993]. In our case, h usually varies between 0.6 and 0.8. In these calculations both the PC’s are normalized by the temporal S.D of each of the PC’s.
3.2 Probability of ’Active’/’Break’ Conditions and Seasonal Mean Monsoon
41
The two-dimensional PDF corresponding to ’strong’, ’weak’ and ’all’ years are shown in Figure 3.5(a), (b) and (c) respectively. For ’strong’ and ’weak’ years the PDF of ISO activity is clearly non-Gaussian while in the case of ’all’ years, it is a Gaussian. The spatial pattern corresponding to the maxima of PDF in each case is constructed using appropriate normalization constants for the PC’s and the corresponding EOF1 and EOF2 patterns. In the ’strong’ case we note that the two maxima have almost equal probability. In the ’weak’ case there are three maxima of the PDF patterns while in the ’all’ case there is only one maximum. As the number of ’strong’ and ’weak’ monsoon years included in the PDF calculation are quite large, we expect the maxima of PDFs in Figure 3.5(a) and Figure 3.5(b) to be robust. To test the statistical significance of these maxima, we created 1000 random sets of time series having same variance and autocorrelation at 1-day lag equal to those of observed PC1 and PC2 and 2-D PDFs were calculated for each of them. In Figure 3.5(a) and Figure 3.5(b), shading represents regions where the observed PDF is significantly greater than the random ones with 90% confidence, i.e. 25 or fewer of the random PDFs were larger than those shown in Figure 3.5(a) and Figure 3.5(b). The maxima of the PDFs are found to be statistically significant in each case. In the ’strong’ and ’weak’ cases, we are primarily concerned with the statistical significance of the PDF maxima representing deviation from Gaussian distribution. Since in the ’all’ case as the PDF pattern is a Gaussian, similar significance test is not presented.
In the ’strong’
case, the maximum with normalized PC1 close to 1 and PC2 close to zero represents a strong ’active’ condition shown in Figure 3.6(a). The other maximum represents a very weak ’break’ pattern. Although the two patterns have equal probability, strong ’active’ pattern would have the dominating influence on the seasonal mean. For the ’weak’ case, the maximum with both PC1 and PC2 close to zero represents a transition pattern. Both the other maxima represent strong ’break’ conditions. One such ’break’ condition with both PC1 and PC2 making approximately equal contributions is shown in Figure 3.6(b). On the other hand, the maximum of the PDF in ’all’ year case corresponds to a transition pattern with insignificant vorticity associated with it is shown in Figure 3.6(c).
3.2 Probability of ’Active’/’Break’ Conditions and Seasonal Mean Monsoon
42
Figure 3.4: First two EOFs of the daily ISO filtered 850 hPa vorticity from 1 June to 30 September. (a) EOF1 and (b) EOF2 for seven ’strong’ years (c) EOF1 and (d) EOF2 for ten ’weak’ years (e) EOF1 and (f) EOF2 for ’all’ (20 years from 1978 to 1997) years. Arbitrary EOF loadings have been multiplied by a factor of 100.
3.2 Probability of ’Active’/’Break’ Conditions and Seasonal Mean Monsoon
43
Figure 3.5: Evidence of change in regimes of ISOs during ’strong’ and ’weak’ monsoon years. Illustrated are two-dimensional PDFs of the ISO state vector spanned by two dominant EOFs of low-level vorticity. PDFs are calculated with principal components normalized by their own standard deviation and taking the summer days (1 June to 30 September) for (a) 7 ’strong’ monsoon years (b) 10 ’weak’ monsoon years (c) 20 combined ’strong’, ’weak’ and ’normal’ years (1978-1997). The smoothing parameter used is h=0.6 and PDFs are multiplied by a factor 100. The first two EOFs (not shown) are different in ’strong’, ’weak’ and ’all’ years but are related to ’active’ and ’break’ conditions. The origin of the plots corresponds to a very weak state representing a transition between the two states (as in the ’all’ case).
3.2 Probability of ’Active’/’Break’ Conditions and Seasonal Mean Monsoon
44
Figure 3.6: Geographical patterns of the dominant regimes for low-level relative vorticity (10−6 s−1 ) shown in Figure 3.5. (a) ’strong’ monsoon years (b) ’weak monsoon years (c) ’all’ years.
3.2 Probability of ’Active’/’Break’ Conditions and Seasonal Mean Monsoon
45
We note that the seasonal mean low-level vorticity over the northern TCZ position is cyclonic (see Figure 2.1). The cumulative effect of higher frequency of occurrence of ’active’ (’break’) conditions is expected to result in stronger (weaker) than normal cyclonic vorticity in this region. Since higher frequency of ’active’ (’break’) conditions are associated with ’strong’ (’weak’) Indian monsoon, we can expect a strong relationship between seasonal mean vorticity over the monsoon trough (northern position of TCZ) and the strength of the Indian monsoon. This conjecture is tested in Figure 3.7 where we plot the seasonal mean relative vorticity averaged over the monsoon trough and IMR for the 40-year period (1958-1997). The correlation between the two time series is 0.74, strongly supporting our conjecture.
Figure 3.7: The monsoon trough vorticity (MTV) and the Indian Monsoon Rainfall (IMR) for a 40-year period (1958-1997). MTV is defined as the seasonal mean vorticity (JJAS) averaged in the domain 40◦ E-90◦ E and 10◦ N-30◦ N. Both time series are normalized by their own standard deviation. Correlation between the two time series is shown. It would be desirable to see if the conclusions derived from circulation alone will be supported if convection is also included to describe ISOs. However, OLR data as proxy for convection is available only from 1974 onwards. The period between 1974 and 1997 contains six ’weak’ monsoon years (1974, 1979, 1982, 1985, 1986, 1987) as described by the criterion used earlier. However, the same criterion indicates only three ’strong’ monsoon years in this period. To enhance the sample size of the ’strong’ monsoon years, we relaxed the objective criterion to include years for which IMR > 0.5 S.D. Based on the relaxed criterion, six ’strong’ monsoon years (1975, 1978, 1983, 1988, 1990,1994) are selected during this period. As in the previous case, a combined EOF analysis is carried out for 850 hPa vorticity and OLR for ’strong’ (’weak’) years. The first two CEOFs in each case are shown in Figure 3.8. It may be noted that the first CEOF in ’strong’ and
3.2 Probability of ’Active’/’Break’ Conditions and Seasonal Mean Monsoon
46
’weak’ cases for positive projection coefficients (PC’s) represent ’active’ like and ’break’ like conditions respectively. The PDF is then calculated on the reduced phase space defined by the first two CEOFs. Similarly, the PDF of CEOF for ’all’ years (all 20 years from 1978-1997) is also calculated. The PDFs for three different cases are shown in Figure 3.9. It is clear that the PDFs are asymmetric for both ’strong’ and ’weak’ cases while it is a Gaussian in ’all’ case. As in the earlier case, statistical significance for the observed PDFs were carried out and regions of phase space where the observed PDFs are significantly larger than the randomly generated ones with 90% confidence are shaded. Using appropriate normalization constants for the PC’s and corresponding EOF1 and EOF2 (Figure 3.8) the patterns representing maxima in PDF are calculated. In the ’strong’ case, the most probable pattern corresponds to an ’active’ condition (Figure 3.10(a)). The other maxima with much less probability represents a weak ’break’ condition (not shown). For the ’weak’ case both maxima correspond to ’break’ conditions. The one with normalized PC’s close to zero, however, represents a weak ’break’ condition (not shown) while the other maxima in PDF represents a strong ’break’ condition (Figure 3.10(b)). The most probable pattern in the ’all’ case (Figure 3.10(c)) corresponds to a very weak pattern representing a transition between ’active’ and ’break’ patterns. Thus, even if we take circulation and convection together, the strong (weak) monsoon appears to be associated with a higher probability of occurrence of active (break) like conditions.
3.2 Probability of ’Active’/’Break’ Conditions and Seasonal Mean Monsoon
47
Figure 3.8: First two CEOFs of the daily ISO filtered 850 hPa vorticity and OLR from 1 June to 30 September. (a) CEOF1 and (b) CEOF2 for six ’strong’ years (c) CEOF1 and (d) CEOF2 for six ’weak’ years (e) CEOF1 and (f) CEOF2 for ’all’ (20 years from 1978 to 1997) years. Arbitrary EOF loadings have been multiplied by a factor of 100.
3.2 Probability of ’Active’/’Break’ Conditions and Seasonal Mean Monsoon
48
Figure 3.9: Same as Figure 3.5 but based on the state vector defined by the first two combined EOF of low-level vorticity and OLR.
3.2 Probability of ’Active’/’Break’ Conditions and Seasonal Mean Monsoon
49
Figure 3.10: Geographical patterns of the dominant regimes shown in Figure 3.9. (a) ’strong’ monsoon years (b) ’weak monsoon years (c) ’all’ years. OLR patterns are shown as shaded contours (Wm−2 ) while the corresponding low-level vorticity are shown in contours (10−6 s−1 ).
3.3 Interannual Variations of ISO Activity and Seasonal Mean Monsoon
3.3
50
Interannual Variations of ISO Activity and Seasonal Mean Monsoon
Instead of trying to relate the seasonal mean monsoon to statistics of ’active’ and ’break’ conditions, in this section we ask, is there a relationship between level of ISO activity and seasonal mean Indian monsoon? Few previous studies have indicated evidence to connect the interannual variations of ISO activity and seasonal mean monsoon strength. Hendon et al. [1999] found that global ISO activity during boreal winter is inversely related to Australian monsoon strength. Recently Lawrence and Webster [2001] found that summertime ISO activity exhibits an inverse relationship with the Indian monsoon strength using OLR over a 22-year period (1975-97). However, they found that the correlation between ISO activity and monsoon over Bay of Bengal is weak. Thus the robustness of the negative relationship between ISO activity and Indian summer monsoon is not clear. We have used circulation data for a 44-year period (1954-97) from NCEP/NCAR reanalysis for this purpose. An index is defined to represent the strength of the ISO activity. In the previous chapter (section 2.5), we have shown that the monsoon ISO is characterized by a meridional bimodal structure. There is a seesaw of low-level vorticity and convection between the two preferred locations of the TCZ on a day-to-day basis. We have defined the ISO activity index as the variance associated with the low-level ISO filtered vorticity averaged over one of the center of action of ISO activity (70◦ E100◦ E, 12◦ N-22◦ N) . The all India monsoon rainfall (IMR) represents the seasonal mean monsoon strength. Figure 3.11 shows time series of the ISO activity index and IMR each normalized by it’s own standard deviation for the 44 year period. The correlation of 0.32 is significant at the 5% level. Although this relationship is not very strong, it indicates that interannual variations of ISO activity is positively related to the monsoon strength. This result is consistent with Figure 3.1. This means that strong activity of intraseasonal oscillations tends to correspond to seasons of above normal rainfall. This result is in contrast with Lawrence and Webster [2001], as they found an inverse relationship between the strength of the ISO activity and IMR using OLR data for a period (1975-1997). The correlation between the ISO activity index and IMR in this period (1975-1997) for low-level vorticity is very low (r=0.09) whereas the correlation for the period 1954-1974 is high (r=0.59). This indicates that the relationship between ISO activity and IMR is weakening in the recent decades. This demonstration of the two periods is rather arbi-
3.4 Discussions and Conclusions
51 r=0.32
2
r=0.59
r=0.09
1 0 −1 −2 1955
1960
1965
1970
1975
1980
1985
1990
1995
Figure 3.11: (a) Time series of ISO activity index (blue) and All India Monsoon Rainfall Index (IMR, black) normalized by it’s own standard deviation for a 44-year period (1954-1997). trary. To examine the changing character of this correlation, a 21-year sliding window correlation analysis was carried out between the two variables. It is seen that (figure not shown) that the correlation remained about 0.6 until mid-seventies and then decreased rapidly and remained low in eighties and nineties.
3.4
Discussions and Conclusions
The primary objective of this study is to investigate how and to what extent the monsoon ISOs influence the seasonal mean and the interannual variability of the Indian summer monsoon. The underlying hypothesis is that the seasonal mean monsoon has a component forced by internal dynamics in addition to a component forced by external conditions such as slowly varying boundary forcings. This hypothesis can be considered as an extension of the Charney and Shukla [1981] hypothesis that suggested the interannual variation of Indian monsoon to be primarily forced by boundary forcing at the earth’s surface. We propose that the part of the interannual variations of the seasonal mean that is independent of external forcing arise from the changes in the statistics of the intraseasonal oscillations of the Indian monsoon. As the ISOs are intrinsically chaotic, the predictability of the seasonal mean Indian monsoon depends on the extent to which the ISOs influence the seasonal mean relative to the externally forced component. According to our conceptual model, the intraseasonal and interannual variations of the Indian monsoon should be governed by a common mode of spatial variability. In addition, if indeed the ISOs determine the ’strong’ and ’weak’ monsoons, the PDF of the ISOs should have higher probability of occurrence of ’active’ conditions during ’strong’ monsoon years and ’break’ conditions during ’weak’ monsoon years. These two elements of our hypothesis are rigorously tested using a sufficiently long record of daily circulation and convection data.
3.4 Discussions and Conclusions
52
Our first objective was to find the mean large scale spatial pattern associated with the ISOs and compare them with that of the seasonal mean pattern. We have found that spatial pattern of the ’active’ and ’break’ that is invariant from year to year and is similar to the spatial structure of the seasonal mean (Chapter 2). The close resemblance between the spatial structure of the ’active’ and ’break’ composites and that of the seasonal mean, indicates a similarity between the spatial structure of intraseasonal and interannual variability. The spatial distribution of standard deviation of 850 hPa vorticity associated with ISO variability and that of interannual variability of the seasonal mean are shown to be closely similar (pattern correlation 0.64, Figure 3.1). That the intraseasonal and interannual variations are governed by a common spatial mode of variability (Figure 3.2) is seen from the notable similarity between the dominant EOF of intraseasonal oscillations (based on 20 years of daily ISO filtered data during the summer season) and the dominant EOF of the interannual variation of the seasonal mean (based on 40 years data of seasonal mean data). In contrast to some recent studies [Annamalai et al., 1999; Sperber et al., 2000] where it is claimed that it is not possible to describe the interannual variations of the Indian summer monsoon by a single EOF, we show that the dominant EOF indeed represents interannual variations of the Indian summer monsoon (correlation between IMR and PC1 is 0.62) if the domain is restricted between 40◦ E-100◦ E and 20◦ S-30◦ N. Sperber et al. [2000] has found a common mode of variability between intraseasonal and interannual variations in the third leading EOF. If the domain of analysis included regions east of 100◦ E, the ENSO related variation in the eastern part of the domain dominates the first EOF and the interannual monsoon variations may appear as the second or third EOF. Next, we show that the interannual variations of the summer ISO variance has the potential for significantly influencing (up to 20-60%) the interannual variations of the seasonal mean. Then, we argue that it is not the amplitude of the ISO activity but the asymmetry in the occurrence of the ’active’ and ’break’ conditions that affect the seasonal mean. We investigate whether the frequency of occurrence of ’active’ and ’break’ conditions are distinctly different during ’strong’ (flood years) and ’weak’ (drought years) monsoon years. For this purpose, a two-dimensional PDF estimation technique is employed on the ISO filtered field. Daily low-level vorticity field between 1956 and 1997 is employed to include a large number of ’strong’ (seven) and ’weak’ (ten) monsoon years. This objective technique clearly shows that the PDFs are distinctly asymmetric and different during ’strong’ and ’weak’ monsoon years and the most frequently
3.4 Discussions and Conclusions
53
occurring pattern during ’strong’ (’weak’) monsoon years is the ’active’ (’break’) pattern. On the other hand, if all years are linked together, the PDF is Gaussian with the transition between ’active’ and ’break’ pattern being the most frequently occurring pattern. Thus, the cumulative effect of the ’active’ condition during a ’strong’ monsoon season lead to stronger that normal cyclonic vorticity in the north TCZ position and stronger than normal seasonal mean. This conclusion is further supported by strong correlation between seasonal mean vorticity over the northern TCZ position and IMR (Figure 3.7). That the conclusions arrived at from the PDF of low-level vorticity are robust is supported by PDF estimate of combined low-level vorticity and convection. Using simultaneous convection and circulation data (1974-1997), combined EOF of the low-level vorticity and OLR is carried out for all ’strong’ and ’weak’ years as well as all the years taken together. This calculation also shows that the most frequent pattern during a ’strong’ (”weak’) year is the ’active’ (’break’) pattern with enhanced (decreased) cyclonic vorticity and negative (positive) OLR anomaly over the northern TCZ position. Our results are consistent with the findings of Krishnamurthy and Shukla [2000] where they examined daily rainfall over Indian continent for the period 1901-1970 and showed that ’strong’ (’weak’) monsoon years are associated with ’active’ (’break’) conditions (their Fig.12a). They define ISO anomalies with respect to a climatological mean seasonal cycle. If they remove the ’seasonal mean anomaly’ (June 1-September 30) from the anomalies they do not find a clear signal of skewness in the PDF. Their figure 13 which brings out the strengthening and weakening of the large scale monsoon flow by the ISO is also consistent with our result. They argue that the active and break phases do not change the character of the mean monsoon flow, but merely representing strengthening and weakening of the flow. However, if we look at the magnitude of change in the mean in their figure, it accounts for 20-30% change of the mean during ’active’ and ’break’ conditions. Close to the equator the change could be even 100%. This we believe represents significant change of the mean by ISO. Both Krishnamurthy and Shukla [2000] and Sperber et al. [2000] agree with each other and claim that the PDF of the ISO modes is biased towards positive or negative side during strong and weak years only if the low frequency interannual variation of the seasonal mean is not removed. Our results, differ with these studies in that we find a distinct bias of the PDF towards ’active’ (’break’) conditions during strong (weak) monsoon years even after removing the seasonal mean anomaly. Although we define anomalies as departure of observations from annual cycle of individual years, the PDF
3.4 Discussions and Conclusions
54
is calculated on the band pass filtered ISO anomaly. This procedure essentially removes the seasonal mean. The main difference between ours and earlier studies is the use of two EOFs to describe the dominant ISO mode in our study. Both the studies mentioned above used only one EOF to describe the dominant ISO mode. Since the dominant ISO mode has a prominent northward propagating character, it cannot be described by just one EOF. We believe that this is the reason why the earlier studies failed to notice any bias in the PDF after removing the seasonal mean.
Chapter 4
Estimate of Potential Predictability of Monthly and Seasonal Means in Tropics from Observations 4.1
Introduction
The predictability of weather (or the instantaneous state of the atmosphere) is limited to about two weeks [Lorenz, 1982] due to inherent instability and nonlinearity of the system. The atmosphere, however, possesses significant low frequency variability. As has been mentioned in Chapter 1, if the low frequency variations of the monthly and seasonal means were entirely governed by scale interactions of the higher frequency chaotic weather fluctuations, then the time averages will be no more predictable than the weather disturbances themselves. However, it appears that a large fraction of the low frequency variability, specially in the tropics, may be forced by slowly varying boundary conditions such as the sea surface temperature (SST), soil moisture, snow cover and sea-ice variations. Hence, the predictability of climate (e.g. space-time averages) is determined partly by chaotic internal processes and partly by slowly varying boundary forcings. This understanding that anomalous boundary conditions (ABC) may provide potential predictability has formed the scientific basis for deterministic climate predictions [Charney and Shukla, 1981; Shukla, 1981, 1998]. Research during the past decade has shown that the climate in large part of tropics is largely determined by slowly varying SST forcing where potential for making dynamical forecast several seasons in advance exists [Latif et al., 1998]. However, during the same period, we have also learnt that there are regions within the tropics, climate of which is not strongly governed by ABC. The Indian summer monsoon is such a system [Webster et al., 1998; Brankovic and Palmer, 1997; Goswami, 1998]. The intraseasonal oscillations such as the east-ward propagating Madden-Julian Oscillations (MJOs) and the north-ward propagating mon-
4.1 Introduction
56
soon ISOs with period in the range of 30 to 60 days are quite vigorous in the tropics. Both the MJOs as well as the monsoon ISOs are known to be driven by internal feedback between convection and dynamics. In addition to the scale interactions between weather disturbances, time-averaging of the chaotic ISOs can also contribute to the low frequency variability of monthly and seasonal means in the tropics. The nonlinear scale interaction associated with the weather disturbances in the tropics is likely to be weak as they are less vigorous compared to their counterpart in the extra-tropics. Therefore, we envisage that most of the internal contribution to the low frequency variations in the tropics may come from time averaged residual of the ISOs. The total low frequency variance of any variable in a given region (σ 2 ) could be written as super-position of variance due to external forcing (σe2 ) and variance due to internal processes (σi2 ). σ 2 = σe2 + σi2 Making unambiguous estimates of the ’internal’ and ’external’ components of variability from observations is rather difficult. Madden [1976, 1981], Madden and Shea [1978] and Shea and Madden [1990] attempted to estimate the two variances in some extratropical regions. They estimated synoptic scale internal variability from short time series (such as within a season) and extrapolated the power spectrum to lower frequencies by assuming a white noise extension. This approach is simple but assumes that the low frequency power spectrum would be white and it could be extrapolated from power at higher frequencies. Shukla [1983] commented at length on Madden’s [1976] approach and argued that the methodology used and assumptions made by Madden could have overestimated the natural variability or ’climate noise’ and underestimated the potential predictability. Madden [1983] while disagreeing with Shukla that his method underestimated the potential predictability agreed that there is considerable uncertainty in separating the so called ’climate noise’ from the climate signal. Shukla and Gutzler [1983] and Short and Cahalan [1983] used a more general low frequency extension of the intraseasonal variance to estimate the level of ’climate noise’. Trenberth [1984a, b] points out that these estimates depend crucially on the use of correct value of T, the effective time between independent data. He pointed out that these studies may have underestimated T by using negatively biased estimates of the lagged autocorrelations by improperly removing the annual cycle and interannual variability. Alternatively this ratio could be estimated using atmospheric general circulation models (AGCM) from a long integration with observed boundary condition and an-
4.1 Introduction
57
other long integration with fixed boundary condition [Goswami, 1998] or from an ensemble of long integrations of the AGCM with the same observed boundary conditions but differing only in the initial conditions [Stern and Miyakoda, 1995; Harzallah and Sadourny, 1995; Rowell et al., 1995]. Kumar and Hoerling [1995] estimated the ratio between the external and internal variability for the extra tropics using a large ensemble of long simulations by an AGCM. Zweirs and Kharin [1998] have examined the interannual variability and potential predictability of 850 hPa temperature, 500 hPa geopotential and 300 hPa stream function simulated by AMIP models. They have found that there is a wide variation in the ability of the AGCMs to simulate observed interannual variability, both total and weather noise component. Zheng et al. [2000] have proposed a method to estimate potential predictability of seasonal means using monthly mean time series. Using this technique they have estimated the potential predictability of surface temperature, 500 hPa geopotential height and 300 hPa winds. The potential predictability tends to be high in the tropics and low in the extratropics as per their calculations. Singh and Kriplani [1986] have estimated potential predictability of lower tropospheric monsoon circulation and rainfall over India for JJA season. Daily 700 hPa geopotential heights, mean sea level pressure and rainfall anomalies were used for the study. They have found that potential predictability of seasonal lower tropospheric fields is less over the monsoon trough, but it generally increases with decreasing latitude. For rainfall, potential predictability is about 50% over the major parts of the country. The reliability of the estimates of potential predictability in this study may be affected by insufficient data length. The method of removing the annual cycles which is important in these kind of analysis [Trenberth, 1984a] has not been outlined. Sontakke et al. [2001] have estimated potential for long-range predictability of precipitation over the Indian sub-continent using precipitation data from 1901-1970. Their study indicate that the climate noise is small compared to climate signal over the Indian monsoon region. The F-ratios of JJAS precipitation ranges from 1.5 to 2.5, with high values on the west coast of India. Although the F-ratios are not very high, it indicates certain amount of potential predictability of the seasonal mean precipitation. In all the studies of estimating potential predictability from observations mentioned earlier, the total interannual variability (i.e the climate signal) is compared to the ’climate noise’. The so called ’climate signal’ actually contains contributions from the ’external’ forcing as well as the internal ’climate noise’. To the best of our knowledge, no attempt has been made to separate the contributions from the ’external’ and the ’inter-
4.2 Estimation of Potential Predictability of Monthly Means
58
nal’ components to the observed interannual variability. Here, we propose a method of separation of interannual variances of monthly means associated with the slowly varying externally forced component and from the internally determined component. The variances associated with the ’internal’ and ’external’ components are estimated. It is also demonstrated that the ’external’ component separated by our method indeed represents the response of the tropical atmosphere to the slowly varying SST forcing. A measure of potential predictability is defined as the ratio between the ’total’ (sum of ’external’ and ’internal’) and the ’internal’ components. Primary objective of this chapter is to make a quantitative estimate of potential predictability of the Asian monsoon climate on monthly and seasonal time scales. Many studies in the past [Madden, 1976, 1981; Madden and Shea, 1978; Shea and Madden, 1990; Shukla and Gutzler, 1983; Short and Cahalan, 1983] have estimated potential predictability of the extratropical climate from observations. Following the pioneering work of Charney and Shukla [1981], some others (e.g. Singh and Kriplani 1986) also have attempted to estimate the potential predictability of the Indian summer monsoon. Due to differences in the methodology used and due to inhomogeneity of data used in different studies, it has been difficult to arrive at an universal conclusion regarding the quantitative measure of predictability over different geographical locations in general and the Indian monsoon region in particular. With the availability of long term record of homogeneous atmospheric circulation data for over 40 years (e.g. from NCEP/NCAR Reanalysis), it is now worthwhile to re-examine the quantitative measure of potential predictability in the tropics. While potential predictability over the global tropical belt will be estimated, the predictability of Asian monsoon region will be contrasted with that over the other tropical regions. In particular, we shall try to assess the contribution of the intraseasonal oscillations to the potential predictability.
4.2 4.2.1
Estimation of Potential Predictability of Monthly Means Methodology
The main data used in this study are the daily low-level zonal winds (850 hPa), upper level zonal winds (200 hPa) and 700 hPa geopotential height from NCEP/NCAR reanalysis for 33 years (1965-1997). Daily interpolated outgoing long wave radiation from the NOAA polar orbiting satellites for 20 years (1980 to 1999) were also used. Our methodology is based on the following premise. The anomalies associated with the
4.2 Estimation of Potential Predictability of Monthly Means
59
synoptic and intraseasonal oscillations may be defined as the deviations from the annual cycle. The annual cycle at any place can be defined by the sum of the first few harmonics. In the present study, the annual cycle is defined as the sum of the first three harmonics of daily data for each year. The annual cycle defined in this manner varies from year to year. An example of such interannual variations of the annual cycle of low-level zonal winds at a point over the Indian Ocean is shown in Figure 4.1. It is clear that the annual cycle has significant year to year variations. We hypothesize that the interannual variations of the annual cycle are essentially forced by the slowly varying boundary forcing. The dominant slowly varying boundary forcing in the tropics is that associated with the El Nino and Southern Oscillation (ENSO) related SST variations. Since the time scale of variations of the boundary forcing is much longer (3-4 years to decadal) than that of the annual cycle, it essentially modulates the annual cycle. Thus, the interannual variations introduced by the ’external’ (slowly varying) forcing can be estimated from the monthly means constructed from the deviations of the individual annual cycles from the climatological mean annual cycle. The annual cycle of zonal winds at 850 hPa and 200 hPa and geopotential height at 700 hPa for all years from 1965 to 1997 and those for OLR for all years from 1980 to 1999 are first calculated. From the daily annual cycles, climatological mean daily annual cycles of each field is calculated. Monthly ’external’ anomalies are estimated as monthly means of deviations of individual annual cycles from the climatological mean annual cycle. If daily anomalies in a particular year is defined as the departure of daily observations from the annual cycle of that year, they represent the ’internal’ contribution as the ’external’ component represented by the interannual variation of the annual cycle is removed in this process. Thus, the monthly means of the daily anomalies constructed in this manner represent the ’internal’ component. This definition implies that averaged over the whole year, the daily anomalies vanish. However, due to the intraseasonal oscillations, the monthly means are non-zero. Our definition of ’internal’ monthly anomaly implies that it is contributed primarily by the intraseasonal oscillations and any ’climate noise’ arising from higher frequency weather events is neglected. The ’internal’ and ’external’ monthly mean anomalies calculated in this manner are statistically independent as the temporal correlation between the two is nearly zero everywhere (figure not shown).
4.2 Estimation of Potential Predictability of Monthly Means
60
Figure 4.1: An illustration of variations of the annual cycle from year to year. The annual cycle of zonal winds (ms−1 ) at 850 hPa at a point (80◦ E, 5◦ N) are shown for 20 years.
4.2 Estimation of Potential Predictability of Monthly Means
61
Let us define total monthly anomaly of any field (say, zonal wind ) as sum of monthly anomalies associated with ’internal’ and ’external’ components. UT (x, y, t) = UE (x, y, t) + UI (x, y, t) where subscripts E and I refer to the ’external’ and the ’internal’ components. Squaring both sides and summing over all months we can write the total variance to be given by sum of variances associated with the ’internal’ and the ’external’ components, namely 2 + σ2, σT2 = σE I
as the correlation between the ’internal’ and the ’external’ components is zero. The total interannual variance may be estimated in two ways. The traditional way of calculating it is to construct monthly mean data from the raw daily data. Then construct a climatological monthly mean annual cycle. Deviations of the monthly means from this climatological monthly mean annual cycle are the total monthly mean anomalies. The total interannual variance may be calculated from these total anomalies. Alternatively, daily anomalies can be constructed with respect to the daily climatological mean annual cycle. The monthly means obtained from these daily anomalies give us the total monthly mean anomalies. Let U(m,n) represent any field for the nth day of the mth th year, where n= 1,2...365; m= 1,2...Y. The annual cycle (Ua (m, n)) is defined as the sum of the first three harmonics of daily data. To find external monthly anomalies:
Daily climatological mean of the annual cycles is defined as Uca (n) =
Y 1 X Ua (m, n) Y
(4.1)
m=1
Daily ’external’ anomaly is defined as ˜ (m, n) = Ua (m, n) − Uca (n) U
(4.2)
Monthly mean of ’external’ anomalies UE (m, k)k=1..12 =
1 30
30∗k X n=1+30∗(k−1)
˜ (m, n) U
(4.3)
4.2 Estimation of Potential Predictability of Monthly Means
62
To find ’internal’ monthly anomalies:
Daily ’internal’ anomaly is defined as ˆ (m, n) = U (m, n) − Ua (m, n) U
(4.4)
Monthly mean of ’internal’ anomalies UI (m, k)k=1..12 =
1 30
30∗k X
ˆ (m, n) U
(4.5)
n=1+30∗(k−1)
To find ’total’ monthly anomalies:
Daily climatological mean is defined as Y 1 X U (m, n) Uc (n) = Y
(4.6)
m=1
Daily ’total’ anomaly is defined as UT (m, n) = U (m, n) − Uc (n)
(4.7)
Monthly mean of daily anomalies 0
U (m, k)k=1..12 =
1 30
30∗k X
UT (m, n)
(4.8)
n=1+30∗(k−1)
To test our claim that the ’external’ anomalies as estimated by us are essentially driven by slowly varying SST changes associated with the ENSO, we carried out a combined EOF analysis of the monthly mean ’external’ anomalies of OLR and winds at 850 hPa. We have chosen the period between 1979 to 1997 for this analysis. The dominant EOF explaining about 20 percent of the total variance is shown in Figure 4.2. The spatial patterns of both OLR and low-level winds correspond well with the canonical patterns associated with ENSO [Rasmusson and Carpenter, 1982; Wallace et al., 1998]. The principal component for the dominant EOF, PC1 (normalized by its own temporal S.D) is also shown in Figure 4.2 together with normalized Nino3 SST anomalies. The correlation coefficient between PC1 and Nino3 (160◦ W-90◦ W, 5◦ S-5◦ N) SST anomalies is 0.84 indicating a strong link between the variability represented by the ’external’ component and the ENSO.
4.2 Estimation of Potential Predictability of Monthly Means
63
Figure 4.2: First combined EOF of mean monthly ’external’ anomalies for the period January 1979 to December 1997 (228 months). (a) Zonal winds EOF at 850 hPa, (b) OLR EOF and (c) PC1 (solid line) and Nino3 SST anomalies (dashed line). Both the time series are normalized by their own standard deviation. Units of the EOFs are arbitrary.
4.2 Estimation of Potential Predictability of Monthly Means
64
Figure 4.3: Time-longitude section of mean monthly ’external’ anomalies of zonal wind at 850 hPa (ms−1 ) and OLR (Wm−2 ) averaged around equator (5◦ S-5◦ N).
4.2 Estimation of Potential Predictability of Monthly Means
65
The second EOF and corresponding time coefficients (PC2) are not shown. However, PC1 and PC2 are strongly correlated at a lag of about 6 months. This lag-correlation together with the spatial patterns of the ’external’ component represent an east-ward propagation of the anomalies, again characteristic of the ENSO anomalies. Therefore, the ’external’ component separated here clearly represents the slow response of the atmosphere to the slowly varying SST forcing associated with the ENSO. Actual anomalies of low-level winds and OLR along the equator associated with the slow external forcing are shown in Figure 4.3. The magnitude of the anomalies during the warm and cold events are similar to those known to be associated with typical warm or cold phases of ENSO [Rasmusson and Carpenter, 1982] and the eastward propagation is also clearly seen.
4.2.2
Estimation of ’Internal’ and ’External’ Interannual Variances
The total variance of monthly means as well as the ’internal’ and ’external’ components of the variance of zonal winds at 850 hPa are calculated as described in the previous section based on daily data for 33 years (1965-1997). The three variances are shown in Figure 4.4. Similarly, the three variances for OLR are calculated based on available 20 years of daily data (1980-1999). The OLR variances are shown in Figure 4.5. To start with, we note that the sum of the ’external’ and ’internal’ variances almost exactly equals the total variances in all geographical locations in the tropics for both the field. Secondly, it is clear from Figure 4.4(b) and Figure 4.5(b) that the geographical distribution of the ’external’ variances of low-level zonal winds as well as OLR has the canonical pattern of the individual fields associated with the ENSO [Wallace et al., 1998; Philander, 1990; Rasmusson and Wallace, 1983]. The ’external’ variance of U850 has a major maximum centered around the dateline and a secondary maximum in the eastern equatorial Indian Ocean. Both the regions are known to be associated with large zonal wind anomalies during peak ENSO phases. The major maximum on the ’external’ variance of OLR is also centered around the dateline but has large extension to the eastern Pacific. It is also noted that most of the appreciable ’external’ variance of either OLR or U850 is confined between 10◦ N and 10◦ S, characteristic of the Walker response associated with the ENSO. On the other hand, the ’internal’ variances of U850 have large amplitude (Figure 4.4(c)) in the ’monsoon’ regions of the tropics, namely the Indian summer monsoon region, the South China Sea monsoon region and the Australian monsoon region. We note that the ’internal’ variance is generally smaller than
4.2 Estimation of Potential Predictability of Monthly Means
66
that of the ’external’ variance in the tropical Pacific. However, it could be comparable or even larger than the ’external’ variance in the monsoon regions mentioned above. The ’internal’ variance associated with the OLR (Figure 4.5(c)) also have large amplitude in same monsoon regions. In contrast to the ’external ’ variance, the ’internal’ variance is not confined to the equatorial belt but extends even up to 30◦ latitude in the Indian and Australian monsoon regions.
Figure 4.4: Monthly variance of zonal winds (m2 s−2 ) at 850 hPa based on 396 months for the period January 1965 to December 1997. (a) Total variance (b) ’external’ variance and (c) ’internal’ variance.
4.2 Estimation of Potential Predictability of Monthly Means
67
Figure 4.5: Same as Figure 4.4 but for OLR for the period January 1980 to December 1999 (240 months). Units, (Wm−2 )2 .
4.2 Estimation of Potential Predictability of Monthly Means
4.2.3
68
Potential Predictability of Monthly Means
Ideally, the potential predictability of either monthly or seasonal means climate is determined as the ratio between ’signal’ to ’noise’, the signal being the predictable ’external’ component while the ’noise’ being the ’internal’ unpredictable component. Since it is normally difficult to separate the ’external’ component from the ’internal’ component, usually potential predictability is defined as the ratio (F-ratio) between total variance (σ 2 ) and climate noise (σi2 ). In finding the potential predictability of the monthly means, since, we have separated the ’external’ and ’internal’ components, we can write F =
σe2 σ2 = + 1. σi2 σi2
Larger the value of this ratio compared to two, higher the predictability. The F-ratio of two also signifies that the signal-to-noise ratio (i.e F-1) is equal to one and that half of the observed interannual variability is potentially predictable. The monthly mean climate may be considered marginally predictable if ’F’ is greater but of the order two. If ’F’ is less than two, the climate would be unpredictable as the ’internal’ variability exercises a dominating influence on the total monthly variability. Figure 4.6 represents the geographical distribution of potential predictability for U850 . The F-ratio for zonal winds at 850 hPa for summer (JJA) months is shown in Figure 4.6(a), while for winter (DJF) months are shown in Figure 4.6(b). Potential predictability is high wherever the ENSO influence is large in the summer months (Figure 4.6(a)). These include equatorial Pacific between 10◦ S and 10◦ N, equatorial Atlantic and equatorial Indian Ocean east of 70◦ E. Parts of Africa also indicate high predictability as this region is also known to have strong influence of ENSO. It may be noted from Figure 4.6(a) and Figure 4.6(b) that during the NH summer, not only the peak values of the ’F’ are higher than those during northern winter, the area covered by ’F’ greater than two is much larger during NH summer compared to that in NH winter. Thus, during NH winter the monthly mean predictability not only decreases compared to that in NH summer, the predictable region also shrinks. Over the Indian monsoon region ’F’ ratio ranges between 2 and 3 during NH winter and goes even below 2 during NH summer. The qualitative difference in the predictability regimes during NH summer compared to NH winter is probably not very surprising if we take into account the seasonality of the ’external’ and the ’internal’ variances. As the ’external’ component of the variance arises from a slowly varying signal (with time scales longer than a year),
4.2 Estimation of Potential Predictability of Monthly Means
69
Figure 4.6: Estimates of ’F’ ratios for zonal winds at 850 hPa (a) for all northern hemisphere summer months (June-July-August) and (b) for all northern hemisphere winter months (December-January-February).
we do not expect much seasonality in the ’external’ variance. This is shown in Figure 4.7 for zonal winds at 850 hPa. Except that the maximum variance occurs in the western Pacific during NH summer compared to central Pacific during winter, the general pattern of ’external’ variance is similar in the equatorial wave-guide during both the seasons. The major difference between the ’external’ variance between the two seasons occur in the central Pacific subtropics. This is due to the ENSO induced off equatorial response being much stronger during the NH winter than in the NH summer. However, the ’internal’ variance has a pronounced seasonality (Figure 4.8). Barring Indian monsoon region and a small region in the American monsoon region, the internal variability is very week throughout the equatorial wave-guide during NH summer. This explains the larger magnitude and extension of ’F’ during NH summer (Figure 4.6(a)). On the other hand, the ’internal’ variance during NH winter are quite strong from Indian Ocean to central Pacific, the maxima being over the Australian monsoon region and the South Pacific Convergence Zone (SPCZ). The larger ’internal’ variability during NH winter is consistent with the fact the ISO activity in tropics is strong during boreal winter and spring and weak during boreal summer except over the Indian monsoon region [Madden and Julian, 1994; Wang and Rui, 1990]. Even though the ’external’ variance
4.2 Estimation of Potential Predictability of Monthly Means
70
Figure 4.7: The ’external’ variance of zonal winds at 850 hPa (m2 s−2 ) during (a) NH summer months (JJA) and (b) NH winter months (DJF).
Figure 4.8: The ’internal’ variance of zonal winds at 850 hPa (m2 s−2 ) during (a) NH summer months (JJA) and (b) NH winter months (DJF).
4.2 Estimation of Potential Predictability of Monthly Means
71
remains similar in magnitude and extent in winter compared to those in summer, the ’F’ ratio becomes smaller and the predictable region reduces to a smaller region in the far eastern Pacific due to vigorous ’internal’ activity in Indian Ocean and central and western Pacific. The estimates of ’F’ ratios for zonal winds at 200 hPa for summer (JJA) months is shown in Figure 4.9(a), while winter (DJF) months is shown in Figure 4.9(b). Potential predictability is high wherever the ENSO influence is large (see Figure 4.9(a)). These include equatorial Pacific between 10◦ S and 10◦ N, equatorial Atlantic and equatorial Indian Ocean. Parts of Africa indicate high predictability. Compared to low-level winds potential predictability is generally high. Over the Indian sub-continent the ’F’ ratio ranges between 2 and 3. For NH winter months (Figure 4.9(b)) predictable region shrinks compared to that of summer months. Maximum predictability is seen over the central equatorial Pacific and equatorial Indian ocean. The region having ’F’ ratio between 2 and 3 occupies a larger region as compared to that during JJA. Figure 4.10 represents the potential predictability for convection (or precipitation). The estimates of ’F’ ratios for OLR for NH summer (JJA) months and NH winter (DJF) months are shown in Figure 4.10(a) and Figure 4.10(b) respectively. It is seen from Figure 4.10(a) that significant predictable region (e.g.’F’ ≥ 2) for convection (or precipitation) is smaller than that for circulation. This region is mainly confined to the central and eastern equatorial Pacific coincident with the core predictable region of ENSO influence. The geographical distribution of potential predictability for OLR for NH winter months is shown in Figure 4.10(b). The predictable region gets confined to central and east equatorial Pacific. The noteworthy feature is that over the Indian monsoon region, ’F’ ratios are less than two for convection. This indicates that the internal variability in the Indian monsoon region is even stronger than the potentially predictable ’external’ component seriously limiting the potential predictability of the Indian summer monsoon. The estimates of ’F’ ratios for geopotential height at 700 hPa for NH summer (JJA) and NH winter (DJF) months are shown in Figure 4.11(a) and Figure 4.11(b) respectively. In contrast to the other fields discussed earlier such as U850 , U200 and OLR, the geopotential field at 700 hPa does not show a major maximum only over the central equatorial Pacific. The whole tropical belt (10◦ S to 10◦ N) shows high values of potential predictability and it is high in both the summer (Figure 4.11(a)) and winter (Figure 4.11(b)) months. During the summer months, southern India shows high potential pre-
4.2 Estimation of Potential Predictability of Monthly Means
72
Figure 4.9: Estimates of ’F’ ratios for zonal winds at 200 hPa (a) for all northern hemisphere summer months (JJA) and (b) for all northern hemisphere winter months (DJF).
Figure 4.10: Estimates of ’F’ ratios for OLR (a) for all northern hemisphere summer months (JJA) and (b) for all northern hemisphere winter months (DJF).
4.2 Estimation of Potential Predictability of Monthly Means
73
Figure 4.11: Estimates of ’F’ ratios for geopotential height at 700 hPa (a) for all northern hemisphere summer months (JJA) and (b) for all northern hemisphere winter months (DJF).
dictability while in the northern India and over the monsoon trough ’F’ ratio ranges between 4 and 6. In the winter months also, ’F’ ratios are high in the tropical belt. Both in summer and winter months ’F’ ratio becomes less between 20◦ and 30◦ latitudes. Since the geographical distribution of potential predictability of geopotential height is different from the other fields like zonal winds and convection, it might be interesting to look into the external and internal variances separately. In order to highlight the variance of the geopotential height in the tropics, the variances shown in Figure 4.12 and Figure 4.13 is restricted between 20◦ S and 20◦ N. This is because the variances of geopotential height in the extratropics tend to be several times larger than those in the tropics. The external variance of geopotential height is shown in Figure 4.12 for JJA and DJF months. The variance associated with the external component is quite high up to 120◦ E though some parts of Africa is showing lower variance. East equatorial Pacific also shows appreciable variance. For the winter months also, the variance up to 120◦ E is high. Over the Pacific, the peak shifts towards central Pacific. The spatial pattern of external variance of Z700 appears to have a wave number two structure. This may be associated with the externally forced interannual variations of divergent Walker circulation.
4.2 Estimation of Potential Predictability of Monthly Means
74
Figure 4.12: The ’external’ variance of geopotential height at 700 hPa (gpm2 ) during (a) NH summer months (JJA) and (b) NH winter months (DJF).
Figure 4.13: The ’internal’ variance of geopotential height at 700 hPa (gpm2 ) during (a) NH summer months (JJA) and (b) NH winter months (DJF).
4.2 Estimation of Potential Predictability of Monthly Means
75
The geographical distribution of the variance of the geopotential height associated with the internal component for the summer and winter months is shown in Figure 4.13. In the summer months, internal variance is low in the entire tropical belt. While for the winter months internal variance values are nearly two times as high as those in summer months. The seasonal variation of internal variance is consistent with the observation that the intraseasonal oscillations in the equatorial region are stronger in the boreal winter as compared to the boreal summer. Here too, the variance values are high towards the midlatitudes (not shown in Figure 4.13). The high external variance and the low internal variance in the tropics explain the high potential predictability in the tropical belt for geopotential height (Figure 4.11). What is responsible for the ’internal’ variability of the monthly means in the tropics? The synoptic disturbances in the tropics are much less energetic than their extratropical counterpart. Therefore, nonlinear interaction amongst the tropical synoptic disturbances are unlikely to result in significant energy in the low frequency regime (e.g. monthly and seasonal means). Moreover due to their higher frequency, the monthly mean residuals from them are expected to be small. Therefore, the internal variability that could influence tropical monthly means are the monsoon ISOs during NH summer and the MJO in the other parts of the tropics. To test the correctness of this conjecture, we calculate ’internal’ variance after removing the synoptic disturbances from the daily anomalies. For this purpose, a Butterworth low-pass filter that keeps all periods greater than 10 days and throws out all periods shorter than 10 days was applied on the daily anomalies of all years after removing the annual cycle of each individual years. Monthly mean anomalies, describing the ’internal’ component, are again calculated by averaging the filtered anomalies over calendar months. The ’internal’ variance calculated from the monthly means of the filtered data has no contribution from the synoptic variations and is solely contributed by the ISOs. The ’internal’ variance calculated in this manner for U850 and OLR are shown in Figure 4.14. A comparison of Figure 4.14(a) with Figure 4.4(c) and Figure 4.14(b) with Figure 4.5(c) reveals that removal of the contribution of the synoptic disturbances from the daily data had no effect on the ’internal’ variance either in magnitude or in spatial distribution. This analysis establishes that the ’internal’ variability of the monthly means is entirely governed by the tropical ISOs.
4.3 Potential Predictability of Seasonal means
76
Figure 4.14: The ’internal’ variance of (a) zonal winds at 850 hPa (m2 s−2 ) and (b) OLR (Wm−2 )2 based on all months after removing the higher frequencies with period shorter than 10 days.
4.3
Potential Predictability of Seasonal means
In this section, we define climate by seasonal mean and examine potential predictability of seasonal mean climate. The ’climatic signal’ may arise from influences truly external to the climate system or it may arise from slowly varying modes of the entire climate system. An example of the latter is the El Nino and Southern Oscillation. The day to day fluctuations or ’weather’ could give rise to variation of the seasonal mean through scale interaction. In tropics, day to day fluctuations of weather is rather weak, but the intraseasonal oscillations are strong. Hence the climate noise is mainly contributed by the scale interaction between weather disturbances and the ISOs. Since a season is significantly long compared to the typical time scale of the ISOs (30-60 days), the ’climate noise’ arising due to the ISOs cannot be estimated by simple statistical averaging (as we did in the case of monthly means) but may be estimated by some kind of low frequency extension of high frequency spectrum. The focus of this section is to find out whether there is significant difference between interannual variations of the climatic states that can be distinguished from the climate noise. Trenberth [1984a, b] has described a method to estimate the ’climate noise’ as the low frequency extension of the high frequency component. We follow this method to find
4.3 Potential Predictability of Seasonal means
77
an estimate of potential predictability of seasonal mean in the tropics, for the Northern Hemisphere summer and winter seasons. The methodology is explained in detail in the Appendix (section 4.5). The potential predictability is defined as the ratio between interannual variance of the seasonal means and the ’climate noise’. The potential predictability of NH summer and NH winter seasons for low-level zonal winds, upper level zonal winds, OLR and geopotential height have been estimated. This part of the our study is not quite new except that we make use of a long homogeneous data set and that we focus on the potential predictability of the Indian monsoon region.
Figure 4.15: Estimates of ’F’ ratios for zonal winds at 850 hPa for (a) NH summer season (JJA) (b) NH winter season (DJF).
Figure 4.15 shows the geographical distribution of potential predictability for lowlevel zonal winds (850 hPa) in NH summer and NH winter seasons. In NH summer, regions where the ENSO influence is large shows high predictability. The potential predictability is maximum in the western equatorial Pacific, and is having an eastward extension over the central and eastern Pacific and equatorial Atlantic. Parts of Africa and eastern equatorial Indian ocean also shows high potential predictability. In NH winter, the maximum shifts towards central equatorial Pacific, but the pattern remains more or less similar. It is noteworthy that the Indian monsoon region have potential predictability values of the order of 1.5 in both the seasons which means that the monsoon climate
4.3 Potential Predictability of Seasonal means
78
is marginally predictable in the summer season. The ’climate noise’ associated with U850 is shown in Figure 4.16. In the summer months, Asian monsoon region shows significant ’internal’ variance. In the winter, variance maxima shifts towards the southern equatorial Indian Ocean and the Australian monsoon region shows high ’internal’ variance. This indicate that the interannual variability of the intraseasonal oscillations in the Indian monsoon region in the NH summer and Australian monsoon region in the NH winter season is comparable to the predictable component, limiting the predictability of the Indian and Australian monsoons.
Figure 4.16: Estimates of ’climate noise’ for zonal winds at 850 hPa for (a) NH summer season (JJA) (b) NH winter season (DJF).
Figure 4.17 shows the geographical distribution of potential predictability for upper level zonal winds (200 hPa) in NH summer and NH winter seasons. Core predictable regions like the equatorial Pacific, African region and equatorial Atlantic shows high ’F’ ratios both in the summer and winter seasons. Over the Indian monsoon region ’F’ ratio ranges between 1 and 3 in the summer months, while this ratio is between 2 and 4 in the winter months. Thus, the upper level winds during the Asian summer monsoon are slightly more predictable than the low-level winds. Also south equatorial Indian Ocean shows high predictability in the winter months for the upper level zonal winds.
4.3 Potential Predictability of Seasonal means
79
Figure 4.17: Estimates of ’F’ ratios for zonal winds at 200 hPa for (a) NH summer season (JJA) (b) NH winter season (DJF).
Figure 4.18: Estimates of ’F’ ratios for OLR for (a) NH summer season (JJA) (b) NH winter season (DJF).
4.3 Potential Predictability of Seasonal means
80
Figure 4.18 shows the potential predictability distribution of convection (OLR) over the tropics in NH summer and winter seasons. Predictable regions shrinks in the case of convection compared to the large scale flow. In the summer season, western and central equatorial Pacific shows high predictability. Some parts of Africa also come under predictable regions. In the winter season, regions which have predominant ENSO influence show high predictability. Seasonal mean climate in Indian monsoon region is marginally predictable in the winter, but the ’F’ ratios are less than two in the summer season. The convection is even less predictable than low level winds during the summer monsoon season. Figure 4.19 shows the potential predictability distribution of geopotential height at 700 hPa over the tropics in NH summer and winter seasons. The ’F’ ratios in the equatorial wave-guide is quite high both in the summer and winter seasons. In the both the seasons south equatorial Indian Ocean shows maximum predictability, though the ’F’ ratios are high in the winter. ’F’ ratios are low as we move up from 10◦ latitude. Indian region shows ’F’ ratios between 3 and 6 for geopotential height. Southern India shows slightly higher ’F’ ratios. This is consistent with the earlier study done in the region for the 700 hPa geopotential height [Singh and Kriplani, 1986]. The ’climate noise’ associated with geopotential height is much less over the Indian monsoon region (Figure 4.20), compared to interannual variance in both the summer and winter months. This explains, the high predictability associated with the geopotential height over the Indian monsoon region.
4.3 Potential Predictability of Seasonal means
81
Figure 4.19: Estimates of ’F’ ratios for geopotential height at 700 hPa for (a) NH summer season (JJA) (b) NH winter season (DJF).
Figure 4.20: Estimates of ’climate noise’ for geopotential height at 700 hPa for (a) NH summer season (JJA) (b) NH winter season (DJF).
4.4 Discussions and Conclusions
4.4
82
Discussions and Conclusions
In the present study, we attempt to determine the part of monthly and seasonal mean climate variability governed by ’internal’ dynamics and that governed by ’external’ slowly varying forcing from long daily observations. Potential predictability of the climate (monthly and seasonal means) is defined as the ratio of the interannual variance of the monthly or seasonal means and the ’internal’ unpredictable component. Four different fields (low-level zonal winds (850 hPa), upper level zonal winds (200 hPa), OLR and geopotential height at 700 hPa) are used for this purpose. Daily U850 , U200 and Z700 are taken from NCEP/NCAR Reanalysis for a period of 33 years (1965-1997). Daily OLR for 20 years (1980-1999) are also used. The monthly mean climate over the monsoon regions of the world appear to be marginally predictable. But the ’F’ ratios ranges between 2 and 3 over the Indian monsoon region which is much less compared to that in other regions in the tropics. In many recent studies, the difficulty in simulating and predicting the Indian summer monsoon has been attributed to the role of the ISOs [Webster et al., 1998; Goswami, 1998, 1995]. In Goswami [1998], it was shown that the strength of the GCM simulated ENSO response decreases as we reach the Indian Ocean and Indian monsoon region and the internal variability could compete with the externally forced variability in this region. The present analysis shows, from observation that the internal variability in the Indian summer monsoon region is indeed comparable to the boundary forced variability. However the fact that the F-ratio ranges between 2 and 3 indicates that the external forced predictable signal is slightly larger than the noise in some regions. Therefore, while deterministic prediction of the monthly mean summer monsoon climate may prove to be difficult, there exists some hope of limited predictability coming from the boundary forcing. The other important result is that except over the Asian summer monsoon region, the monthly mean climate during the boreal summer is more predictable over a much larger region in the tropics than during boreal winter. As it is well known that the SST signal associated with the ENSO tends to peak during NH winter, it appeared rather strange that predictability should be weaker during this season. However, we show that the weaker and limited predictability during boreal winter is due to stronger internal variability associated with stronger ISOs during winter while the amplitude of the boundary forced variability remains similar to those in boreal summer. Thus, the
4.4 Discussions and Conclusions
83
monthly mean tropical climate seems to be more predictable in NH summer compared to NH winter over much of the tropical belt except in the Indian summer monsoon region. The predictability of the seasonal mean climate over the Indian monsoon region also appear to be marginal. The ’F’ ratio which is a measure of potential predictability is of the order of 1.5. As in the case of monthly mean climate, the Asian monsoon region is the region of lowest predictability of the seasonal climate during boreal summer. Barring the Indian monsoon region, most of the regions in the equatorial wave guide seem to be highly predictable. Equatorial Pacific are associated with higher predictability values. Not surprisingly regions that come under the influence of ENSO have high predictability. As may be expected, the geographical distribution of potential predictability of the monthly and seasonal mean climate bear similarity in all the fields. Comparison between Figure 4.6(a) and Figure 4.15(a) reveal that the core predictable regions of monthly mean climate in the summer months and that of the seasonal mean climate in the summer season is the same for low-level zonal winds. Equatorial Pacific, equatorial Atlantic, south equatorial Indian Ocean and the African region seems to be highly predictable in both the cases. Over the Indian monsoon region, ’F’ ratios are of the order of two in the monthly mean climate, while the ratios of the order of 1.5 in the seasonal mean. If we compare Figure 4.6(b) and Figure 4.15(b), it is clear that ’F’ ratios are much larger in the central equatorial Pacific for the seasonal mean winter climate compared to the monthly mean climate in the winter months. Some parts of Africa, equatorial Indian Ocean and equatorial Atlantic comes under predictable regions in both the cases. Over the Indian monsoon region, ’F’ ratio is of the order of two in the monthly mean climate, while it is of the order of 1.5 in the seasonal mean winter climate for low-level zonal winds. Thus, it appears that the seasonal mean summer monsoon may be more difficult to predict compared to the monthly means of monsoon during boreal summer. It may be noted that, of the four fields used in this study, low-level zonal winds at 850 hPa, upper level zonal winds at 200 hPa and OLR shows some what similar characteristics in both monthly and seasonal mean potential predictability. But the geographical distribution of potential predictability of geopotential height at 700 hPa shows high potential predictability over almost the whole tropical belt. Within the tropics, the Indian summer monsoon region does show relatively lower potential predictability during boreal summer compared to rest of the tropics (Figure 4.11(a) and Figure 4.19(a)).
4.4 Discussions and Conclusions
84
However, the geopotential height seem to be predictable even over the Indian monsoon region. The difference in the potential predictability of the geopotential height and the circulation and convection fields is not surprising as the geopotential field is not strongly coupled to circulation field as in the extratropics. In the tropics, the transient disturbances (that give rise to internal variability) are driven not by available potential energy associated with mean temperature gradient but by potential energy associated with convection. That is why predictability is poorest for convection (OLR) and increasingly higher for low level and upper level winds. Therefore, it is incorrect to conclude that Indian monsoon is predictable by simply looking at the geopotential height field. One need to look at the circulation, convection and precipitation fields to arrive at the correct picture of predictability of the monsoon.
4.5 Appendix : Procedure for Estimating ’Climate Noise’
4.5
85
Appendix : Procedure for Estimating ’Climate Noise’
The day to day fluctuations or ’weather’ could give rise to variation of the seasonal mean through scale interaction. This is often termed as ’climate noise’. This ’climate noise’ has to be estimated from low frequency extension of the high frequency component. We follow the method suggested by [Trenberth, 1984b] compute the ’climate noise’ of seasonal means. First step is to remove the annual cycle. Daily climatological mean annual cycle for the entire data has been found out for this purpose. The daily climatological mean is smoothed using harmonic analysis. Daily anomalies are constructed with respect to the smoothed daily climatological mean annual cycle. The daily anomalies are detrended using a least squares linear fit. Suppose that, the data base consists of N daily values (χi,j , i = 1, ..N, j = 1, ..J) that make up the season for J years in which the mean and annual cycles have been removed. The problem is to assess whether there is any significant climate variability beyond that due to climatic noise. For each year, mean χ ¯j may be computed N 1 X χ ¯j = χi,j N
(4.9)
J 1X 2 χ ¯j J
(4.10)
i=1
and 2 Sm =
j=1
is the sample interannual variance; an unbiased estimate (ˆ) of the population interannual variance which includes the effects of uncertainty in the overall mean, is therefore J
2 σ ˆm =
J 1 X 2 2 Sm = χ ¯j . J −1 J −1
(4.11)
j=1
The noise may be found as a low frequency extension of high frequency variability. The intraseasonal sample variance for the jth year is Sj2
N 1 X = (χi,j − χ ¯j )2 . N
(4.12)
i=1
Therefore, the mean intraseasonal variance is J 1X 2 S = Sj J 2
j=1
(4.13)
4.5 Appendix : Procedure for Estimating ’Climate Noise’
86
An unbiased estimate of σ 2 based solely upon the intraseasonal variances, is N S2. N − To
(4.14)
σ2 σ 2 To (N ) = Nef f N
(4.15)
σ ˆ2 = 2 is The variance due to climatic noise σN 2 σN =
where To is the time between independent values normalized by the sampling interval. The effective number of independent observations Nef f = N ∆T To where ∆T is the sampling interval. Therefore from (4.14) and (4.15) 2 σ ˆN =
To S2. N − To
(4.16)
To is not known apriori and is dependent upon autocorrelation, rL N X
To (N ) = 1 + 2
(1 −
L=1
L )rL . N
(4.17)
where rL is the autocorrelation with lag L of the data. To find rL CLj
N 1 X (χi−L,j − χ ¯j )(χi,j − χ ¯j ), = N
(4.18a)
i=L+1
Sample autocorrelation at lag L is rˆLj = CLj /Coj
(4.18b)
and the overall autocorrelation, rˆL is rˆL =
J 1X rˆLj J
(4.18c)
j=1
One way to test whether there is any signal is to form the null hypothesis that there 2 and σ 2 are both independent estimates of interannual is no signal. In that case σ ˆm ˆN
variance. The former is based upon seasonal means, while the latter is based upon intraseasonal variations. Consequently, the F ratio defined as F =
2 σ ˆm 2 σ ˆN
(4.19)
is the ratio of the two estimated interannual variances and it should follow the F distribution with J-1 and J(Nef f − 1) degrees of freedom.
Chapter 5
Clustering of Synoptic Systems During the Indian Summer Monsoon by Intraseasonal Oscillations As shown in chapter 3 and 4, the monsoon ISOs has large spatial scale and results in strengthening and weakening of the large scale monsoon flow in the extreme phases. This results in strengthening and weakening of the shear of the zonal wind and lowlevel vorticity over the monsoon trough. Since the higher frequency synoptic systems arise from instability of the zonal flow, ISOs have the potential for modulating synoptic activity during the monsoon season. In this chapter, we examine how higher frequency synoptic systems are modulated by the intraseasonal oscillations. The motivation of this study came from the fact that the slow evolution of ISOs may permit extended range prediction of the ISO phases and through it probability of occurrence of wet and dry spells of the monsoon.
5.1
Introduction
A prominent feature of the seasonal mean (June-September) Indian summer monsoon circulation is the monsoon trough (Figure 2.1a,b), an elongated semi-permanent cyclonic vortex in the lower atmosphere associated with low surface pressure extending from Pakistan in the west to Myanmar in the east [Rao, 1976]. The summer monsoon is punctuated by periods of abundant rainfall (’active’ or wet spells) and periods of scanty rain (’break’ or dry spells) in the trough region. There are three or four active and break spells each in a typical monsoon season. If long breaks occur in critical growth periods of agricultural crops, they can lead to substantially reduced yields [Gadgil and Rao, 2000; Lal et al., 1999]. Extended range prediction of the wet and dry spells of monsoon rain could therefore be of immense benefit to Indian agriculture.
5.1 Introduction
88
The main rain bearing weather systems over the monsoon trough region are synoptic scale low pressure systems with typical life time of 3-5 days and length scale of about 2000 km. Monsoon LPS are called lows if the maximum wind speed is less than 8.5 ms−1 and has one closed isobar with the central pressure in the system being lower than the surroundings by more than 2 hPa. The maximum wind speed in depressions is between 8.5 ms−1 and 17 ms−1 and there are atleast two closed isobars, with 4 hPa pressure drop associated with the system. Most depressions are born in the Bay of Bengal and give copious rain as they move inland along the monsoon trough. Monsoon lows and depressions arise as a result of dynamic instability energized by moist convection [Shukla, 1978; Goswami et al., 1980; Mak, 1987]. Large meridional shear of the eastward component of winds and high cyclonic vorticity at low levels in the monsoon trough favor the growth of these instabilities. Intraseasonal oscillations of the Indian summer monsoon have space and time scales that are distinct from those of synoptic systems. ISOs have periods of 10-70 days, zonal scale of 8,000-10,000 km, and are associated with repeated northward propagation of the tropical convergence zone from the south equatorial Indian Ocean to the monsoon trough region [Sikka and Gadgil, 1980; Yasunari, 1979; Krishnamurti and Ardunay, 1980]. As the ISOs modulate the large scale monsoon circulation, strengthening the low-level monsoon winds in one phase while weakening them in the opposite phase (see Chapter 2), they have the potential to modulate synoptic activity. Although previous studies do indicate association of synoptic activity with ISO regimes [Murakami et al., 1984, 1986; Yasunari, 1981], a comprehensive study of the influence of ISOs on LPS genesis does not exist. Here, using daily circulation data and LPS genesis data for 40 years we show that the wet and dry spells of the Indian summer monsoon arise from space-time clustering of the LPS and that the clustering is caused by modulation of the large scale monsoon circulation by ISOs. Our work implies that the predictability of the timing of wet and dry spells is strongly tied to the predictability of the slowly varying monsoon ISOs. The dates and locations of genesis of all lows and depressions during June-September of 1954-1993 over the Indian monsoon region (50◦ E-100◦ E, Eq-35◦ N) are based on reports of the India Meteorological Department (IMD). Data for the first 30 years (19541983) are taken from Mooley and Shukla’s [Mooley and Shukla, 1989, 1987] compilation based on IMD’s Daily Weather Reports; data for the next 10 years (1984-1993) are compiled from the Seasonal Weather Summaries published by IMD. For example weather summary of 1984 monsoon season is from IMD [1985]. Circulation changes associated
5.2 Wet and Dry Spells and Clustering of LPS
89
with monsoon ISOs are based on daily 850 hPa wind fields from NCEP/NCAR reanalysis for the period 1954-1993. Rainfall data are based on five-day (pentad) Climate Prediction Center Merged Analysis of Precipitation (CMAP) for fifteen years (1979-1993). Anomalies are obtained by subtracting the annual cycle (sum of the mean, annual and semiannual harmonics) from the daily (or pentad) observations for each year.
5.2
Wet and Dry Spells and Clustering of LPS
First, we demonstrate that the wet and dry spells of the monsoon rainfall arise mainly from the time clustering of LPS. Pentad rainfall anomaly spatially averaged over the two contiguous boxes (85◦ E-95◦ E, 12◦ N-17◦ N) and (70◦ E-90◦ E, 17◦ N-22◦ N) during 1979-1993 represents the rainfall over the monsoon trough, denoted by P. A zero value of P corresponds to the seasonal mean rainfall over the trough (11.5mm/day). Periods of positive (negative) P correspond to wet (dry) spells. In Figure 5.1, we mark the calendar dates of genesis of all LPS in the monsoon trough region between June and September during 1979-1993 as a function of P normalized by its standard deviation (4.5mm/day). More than two times as many LPS form during periods of positive P (111systems) compared to periods of negative P (52 systems), clearly showing the close association between the genesis of lows and depressions and timing of wet and dry spells. We propose that this clustering of LPS is caused by modulation of the large scale monsoon circulation by ISOs.
Figure 5.1: Genesis dates of LPS between 1 June and 30 September of all years during 1979 to 1993 over the monsoon trough as a function of normalized departure of precipitation over the trough from the seasonal mean.
5.2 Wet and Dry Spells and Clustering of LPS
90
Figure 5.2: Leading Empirical Orthogonal Functions ( (a) EOF1 & (b) EOF2) of 10-80 day filtered wind anomalies (ms−1 ) at 850 hPa between June 1 and September 30 for the period 19641973. (c) Normalized time series of PC1 and PC2 for ten years (each year has 122 days). (d) Normalized Monsoon Intraseasonal Oscillation Index (MISI) for 10 years. Periods of MISI > +1 (MISI< -1) correspond to active (break) phases of the monsoon. It may be noted that positive (negative) phase of MISI represents enhancement (weakening) of the EOF1 pattern.
5.3 Monsoon Intraseasonal Oscillation Index
5.3
91
Monsoon Intraseasonal Oscillation Index
To establish that ISOs influence the genesis of LPS, we define a simple index that captures intraseasonal variability of circulation in the Indian monsoon region. The space time evolution of ISOs may be described by the two leading empirical orthogonal functions (EOF1 & EOF2) of 10-80 day band-pass filtered 850 hPa summer monsoon winds (1 June to 30 September for 1954-1963) in the region 40◦ E-120◦ E, 20◦ S-30◦ N. Together they explain 25% of daily variance of the wind field; their principal components PC1 and PC2 correlate strongly with a lag of about 9 days. The sum of the two EOFs represents the northward propagating monsoon ISOs. We introduce the monsoon intraseasonal oscillation index (MISI) based on the first two principal components of the wind field, MISI(t) = PC1(t)+PC2(t-9). The spatial structure of winds associated with these EOFs and their corresponding principal components (PC1 and PC2) and MISI for 10 years (1964-1973) is shown in Figure 5.2. The spatial structure of winds associated with EOF1 (Figure 5.2(a)) bears a broad resemblance with the seasonal mean low-level circulation. The positive (negative) phase of MISI represents circulation anomalies that strengthen (weaken) the mean monsoon winds (see Figure 2.1(a)) between 5◦ N and 17◦ N by upto 30%, thereby intensifying (weakening) the monsoon trough. We say that the monsoon is in its active (break) phase in periods when normalized MISI is greater than +1 (less than -1). The index (MISI) for the remaining 30 years is constructed taking data for 10 years at a time.
5.4
Clustering of Genesis of LPS by Intraseasonal Oscillations
In order to bring out how genesis of LPS depend on the phase of the ISOs, the frequency of occurrence of the LPS corresponding to the different ISO phases are counted. ISO phases may be defined as bins of normalized MISI. Such a frequency distribution of genesis of LPS as a function of the phase of monsoon ISOs during 1954-1993 is shown in Figure 5.3. The total number of lows and depressions during this 40-year period is 503, with a seasonal average of 12.5 LPS. We note that the number of depressions during the last decade (1984-93) is lower than earlier decades [Mooley and Shukla, 1989], while the number of lows is higher leaving the average number of LPS almost unchanged. Out of the total 503 LPS, 320 occur in the positive phase of the ISO (positive MISI) and 183 in the negative phase. The enhanced low-level shear and cyclonic vorticity in the
5.4 Clustering of Genesis of LPS by Intraseasonal Oscillations
92
Frequency
100 75 50 25 0 −3
−2
−1 0 1 2 Normalised MISI
3
Figure 5.3: Histogram of genesis of synoptic events (lows & depressions) for the Indian monsoon region (50◦ E-100◦ E, Eq-30◦ N) during June to September for the period 1954-1993 as a function of normalized MISI. monsoon trough makes LPS genesis in the positive phase more probable. Figure 5.4 shows the total vorticity and locations of genesis of all LPS in the active (MISI > +1) and break (MISI < -1) phases of the monsoon. The birth of an LPS is more than twice as likely in the active phase (119 systems) than in the break phase (52 systems), with dense clustering in the monsoon trough (Figure 5.4(a)). The total vorticity in the trough region remains weakly cyclonic even during breaks (Figure 5.4(b)), and this explains why some LPS form here even in the break phase. Relatively few LPS are born in the southern Bay of Bengal in this phase although the cyclonic vorticity is high. This may be because the large vertical shear of the winds in the southern region inhibits LPS genesis [Rao, 1976] and partly because the boundary layer frictional convergence necessary for cyclogenesis is less effective in this region as compared to the northern region. The central result of the present study is that circulation changes associated with the monsoon ISOs cause lows and depressions to cluster together in both time and space (Figures 5.3, 5.4). Mechanisms similar to the one proposed here seem to be responsible for the clustering of tropical cyclones in the Gulf of Mexico [Maloney and Hartmann, 2000a], eastern Pacific [Maloney and Hartmann, 2000b] and western Pacific Liebmann et al. [1994] through modulation of large scale circulation by the Madden Julian Oscillations. Finally, we show that intraseasonal fluctuations of cyclonic vorticity in the monsoon trough are associated with coherent fluctuations in the large scale rainfall distribution. We composite the anomaly winds over all active and break days based on MISI in the period 1954-1993, and use these to create an active minus break composite of large scale monsoon circulation and vorticity (Figure 5.5(a)).
5.4 Clustering of Genesis of LPS by Intraseasonal Oscillations
93
Figure 5.4: Total (climatology+composite anomaly) relative vorticity (10−6 s−1 ) at 850 hPa during the (a) ’Active’ ISO phase (MISI > +1) and (b) ’Break’ ISO phase (MISI < -1). Dark dots indicate the position of the genesis of the LPS during active and break phases.
5.4 Clustering of Genesis of LPS by Intraseasonal Oscillations
94
Figure 5.5: Composites based on active and break days as defined by the ISO index, MISI. (a) ’Active’ minus ’Break’ composite wind anomalies (ms−1 ) and associated relative vorticity (10−6 s−1 ) at 850 hPa during the 40 year period (1954-1993). Only vectors significant at 95% confidence level are displayed. Positive contours are shaded and negative contours are not shown. (b) ’Active’ minus ’Break’ composite precipitation anomalies (mm.day−1 ) during 19791993.
5.5 Summary and Conclusions
95
This composite represents intraseasonal changes in monsoon circulation as captured by the two leading EOFs of the low-level winds. We also create the corresponding active minus break rainfall composite based on CMAP rainfall anomalies using the same active and break dates in the period 1979-1993 (Figure 5.5(b)). Enhanced cyclonic vorticity in the monsoon trough region is accompanied by enhanced rainfall. The positive rainfall anomaly is mainly due to the larger number of lows and depressions formed in the trough in the active phase of the monsoon compared to those during the break phase (Figure 5.4). Decreased precipitation over the equatorial region and the rain shadow region in southeast India are also evident in the active phase (Figure 5.5(b)). The spatial pattern of the rainfall composite is consistent with the classical pattern of intraseasonal monsoon rainfall variability seen in rain guage data over the continent [Singh and Kriplani, 1990; Krishnamurthy and Shukla, 2000].
5.5
Summary and Conclusions
The timing and duration of wet and dry spells of the summer monsoon have a strong bearing on the agricultural production and water resources in the Indian subcontinent. We show that the wet and dry spells are the result of space-time clustering of monsoon low pressure systems caused by modulation of the large scale monsoon flow by intraseasonal oscillations. The ISOs alternately enhance and reduce horizontal shear and cyclonic vorticity of low-level winds along the monsoon trough on time scales of 10-80 days. Genesis of LPS is twice as likely in periods when monsoon trough vorticity is enhanced as compared to periods when it is reduced. There is also a spatial clustering of LPS genesis, with a majority of LPS being born in north Bay of Bengal in periods of enhanced monsoon trough vorticity. Skillful statistical forecasts upto two to three weeks in advance have been demonstrated for the slow evolution of the equatorially confined, eastward propagating MJO [Lo and Hendon, 2000; Waliser et al., 1999; Mo, 2001]. We envisage that the slowly varying monsoon ISOs will turn out to have similar predictability. Work in this direction might lead to extended range prediction of the wet and dry spells of the Indian summer monsoon.
Chapter 6
Conclusions Indian summer monsoon displays substantial interannual variability, which have profound socio-economic consequences. Long range prediction of seasonal mean monsoon precipitation, therefore assumes great significance. Even though climate modelling has made great progress in simulating and predicting the climate over several tropical regions, dynamical prediction of seasonal mean monsoon precipitation however, remains as an extremely frustrating experience. Within the summer monsoon season (June-September), the timing and duration of the monsoon intraseasonal oscillations (wet and dry spells of the summer monsoon) have a strong bearing on the agricultural production and water resources in the Indian subcontinent. Monsoon studies so far, has not clearly established whether the occurrence of wet and dry spells of monsoon rainfall is due to some form of dynamical instability of the mean monsoon flow, or a mere indicator of the formation, growth and propagation of monsoon depressions, or due to low frequency chaotic intraseasonal oscillations. Research during the past decades, has indicated the possible role of intraseasonal oscillations as one of the reasons which limits the predictability of the seasonal mean monsoon. In this study, we consider the intraseasonal oscillations (ISOs) as the building block for Indian summer monsoon. We demonstrate how ISOs influence the seasonal mean and limits its predictability in one hand while enhancing potential predictability of the wet and dry spells of the monsoon by modulating the frequency of occurrence of the synoptic events on the other. Some outstanding questions regarding relationship between intraseasonal oscillations and interannual variability of the Indian summer monsoon are addressed.
6 Conclusions
97
• How could intraseasonal oscillations influence interannual variations of the Indian summer monsoon? Is there a common mode of variability between the intraseasonal and interannual variability of the Indian monsoon? • Is there a distinct difference in the probability of occurrence of ’active’ and ’break’ phases’ in the strong and weak monsoon years? • How much of the interannual variability of the Indian summer monsoon is governed by ’internal’ chaotic processes? How much of this ’internal’ low frequency variability is contributed by the monsoon ISOs? • Is there evidence from observation that occurrence of rain bearing monsoon synoptic systems (lows and depressions) are modulated by the intraseasonal oscillations? First, we show that the underlying spatial pattern of the dominant intraseasonal mode is invariant over the years and is similar to the spatial structure of the seasonal mean monsoon. The dominant ISO is characterized by a meridional bimodal structure with ascending (descending) motion and enhanced (decreased) convection over the monsoon trough and descending (ascending) motion and decreased (enhanced) convection over the oceanic TCZ in the ’active’ (’break’) phase. Thus extreme phases of the dominant ISO mode (’active’ and ’break’ phases) are associated with general strengthening (weakening) of large scale mean monsoon flow leading to strengthening (weakening) of the monsoon trough. Hence it is possible that, the statistics of ISO (phase, amplitude) affect the seasonal mean monsoon. Then, we demonstrate that the intraseasonal and interannual variations are governed by a common spatial mode of variability. Further it is shown that probability of occurrence of the intraseasonal oscillations is related to the interannual variability of the seasonal mean. The frequency of occurrence of ’active’ and ’break’ conditions are found to be distinctly different during ’strong’ and ’weak’ monsoon years. It is shown that the most frequent pattern during a ’strong’ (’weak’) monsoon year is the ’active’ (’break’) pattern with enhanced (decreased) cyclonic vorticity and convection over the monsoon trough. All these results lead to the conclusion that monsoon ISOs modulate interannual variation of the Indian monsoon in a significant way. Having shown that the ISOs can influence the seasonal mean and its interannual variability, we attempt to make quantitative estimates of potential predictability of mon-
6 Conclusions
98
soon climate. Potential predictability is defined as the ratio between the interannual variance of the the monthly or seasonal means and its internally forced ’climate noise’ component. We argue that the ISOs contribute mainly to the ’climate noise’ in the tropics as the amplitude of the synoptic disturbances is rather small and are unlikely to lead to much low frequency internal variability through nonlinear scale interactions. For monthly climate, we propose a new method to separate the ’internal’ and ’external’ contribution to the interannual variability. For the seasonal climate, the internal ’climate noise’ is estimated using a method equivalent to low frequency extension of high frequency spectrum as done in some previous studies. It is found that slowly varying boundary forcing strongly govern the predictability of monthly or seasonal climate of most of the tropical regions except the Indian monsoon region. Quantitative estimates of potential predictability of monthly and seasonal climate reveal that the potential predictability of the Indian monsoon is much lower compared to the other regions in the tropics. This is due to the fact that the influence of the internally forced component of the seasonal mean is comparable to its externally forced counterpart in the Indian monsoon region. These estimates reveal that the Indian monsoon climate may be considered only marginally predictable. We also find that the internally forced component of the monthly/seasonal climate in the Indian monsoon region is due to the intraseasonal oscillations. The monsoon ISO results in strengthening and weakening of the mean monsoon flow in the extreme phases. The main rain bearing system during the monsoon season are the Low Pressure Systems (LPS) consisting of lows and depressions. Since the genesis of the LPS depends on the horizontal shear and low-level vorticity, it is possible that more LPS form in the active phase relative to the break phase. In other words, large scale circulation associated with the ISOs could modulate the frequency of genesis of LPS. Using LPS genesis data for more that 40 years and corresponding circulation data to describe the ISOs, we show that the dry and wet spells of the Indian monsoon are caused by clustering of low pressures systems in space and time which is caused by the modulation of the large scale monsoon flow by the intraseasonal oscillations. The slow evolution of the ISO may permit extended range prediction of the ISO phases and through them dry and wet spells of the Indian summer monsoon. In this study, we have used long homogeneous data sets (30-40 years) to examine the statistics of the ISO. Hence we hope that the results are reliable. Above results have important implication on the seasonal mean monsoon prediction. While the monsoon
6 Conclusions
99
ISOs seem to result in limiting the predictability of monthly or seasonal mean monsoon climate, it is possible that the same ISOs lead to enhancing extended range prediction of spells of synoptic activity. As demonstrated in the case of equatorial MJO, extended range prediction of monsoon ISOs may be possible due to it’s slow evolution. Since ISOs modulate the main rain bearing systems in the monsoon region, the prediction ISO phase may lead to predicting the dry and wet spells of the Indian summer monsoon. Studies in this direction will help in increasing the predictability of the Indian summer monsoon. Thus, ISOs appear to play a crucial role in determining predictability of monsoon in different time scales. Hence, the success in predicting the Indian summer monsoon rainfall would depend on the precise representation of ISOs in a dynamical model.
Bibliography Ahlquist, J., V. Mehta, A. Devanas, and T. Condo, Intraseasonal monsoon fluctuations seen through 25 years of Indian radiosonde observations, Mausam, 41, 273–278, 1990. Annamalai, H., J. M. Slingo, K. R. Sperber, and K. Hodges, The mean evolution and variability of the Asian summer monsoon: Comparison of ECMWF and NCEP-NCAR Reanalyses, Mon. Wea. Rev., 127, 1157–1186, 1999. Brankovic, C., and T. N. Palmer, Atmospheric seasonal predictability and estimates of ensemble size, Mon. Wea. Rev., 125, 859–874, 1997. Brankovic, C., and T. N. Palmer, Seasonal skill and predictability of ECMWF PROVOST ensembles, Quart. J. Roy. Meteor. Soc., 126, 2035–2067, 2000. Charney, J. G., and J. Shukla, Predictability of Monsoons, in Monsoon Dynamics, edited by J. Lighthill and R. P. Pearce, pp. 99–108, Cambridge University Press, Cambridge, 1981. Chatfield, C., The analysis of time series: An introduction, Chapman and Hall, London, 1980. Dakshinamurthy, J., and R. N. Keshavamurthy, On oscillations of period around one month in the Indian summer monsoon, Ind. J. Meteorol. Geophys., 27, 201–203, 1976. De, U. S., and J. C. Natu, Low frequency oscillations in tropospheric winds during contrasting summer monsoon over India, Mausam, 45, 261–266, 1994. Fennessy, M., and J. Shukla, Simulation and predictability of monsoons, in Proceedings of the international conference on monsoon variability and prediction, WMO/TD 619, pp. 567–575, Trieste, 1994. Ferranti, L., J. M. Slingo, T. N. Palmer, and B. J. Hoskins, Relations between interannual
Bibliography
101
and intraseasonal monsoon variability as diagnosed from AMIP integrations, Quart. J. Roy. Meteor. Soc., 123, 1323–1357, 1997. Gadgil, S., and P. R. S. Rao, Famine strategies for a variable climate - A challenge., Curr. Sci., 78, 1203–1215, 2000. Gadgil, S., and S. Sajani, Monsoon precipitation in the AMIP runs, Climate Dyn., 14, 659–689, 1998. Goswami, B. N., Dynamical predictability of seasonal monsoon rainfall: Problems and prospects, Proc. Ind. Nat. Sci. Acad., 60A, 101–120, 1994. Goswami, B. N., A multiscale interaction model for the origin of the tropospheric QBO, J. Climate, 8, 524–534, 1995. Goswami, B. N., Chaos and Predictability of the Indian summer monsoon, PramanaJournal of Physics, 48, 719–736, 1997. Goswami, B. N., Interannual variations of Indian summer monsoon in a GCM: External conditions versus internal feedbacks, J. Climate, 11, 501–522, 1998. Goswami, B. N., and J. Shukla, Quasi-periodic oscillations in a symmetric general circulation model, J. Atmos. Sci., 41, 20–37, 1984. Goswami, B. N., R. N. Keshavamurthy, and V. Satyan, Role of barotropic-baroclinic instability for the growth of monsoon depressions and mid-tropospheric cyclones, Proc. Ind. Acad. Sci. (Earth & Planetary Sciences), 89, 79–97, 1980. Goswami, B. N., D. Sengupta, and G. Sureshkumar, Intraseasonal oscillations and Interannual variability of surface winds over the Indian monsoon region, Proc. Ind. Acad. Sci. (Earth & Planetary Sciences), 107, 45–64, 1998. Gruber, A., and A. F. Krueger, The status of the NOAA outgoing long-wave radiation data set, Bull. Amer. Meteor. Soc., 65, 958–962, 1984. Hartmann, D. L., and M. L. Michelson, Intraseasonal periodicities in Indian rainfall, J. Atmos. Sci., 46, 2838–2862, 1989. Harzallah, A., and R. Sadourny, Internal versus SST-forced atmospheric variability simulated by an atmospheric general circulation model, J. Climate, 8, 475–495, 1995.
Bibliography
102
Hendon, H. H., C. Zhang, and J. D. Glick, Interannual variation of the Madden-Julian oscillation during austral summer, J. Climate, 12, 2538–2550, 1999. IMD, Monsoon Season (June-September, 1984), in Mausam, vol. 36, pp. 393–402, India Meteorological Department, New Delhi, 1985. Kalnay, E., et al., The NCEP/NCAR 40-year reanalysis project, Bull. Amer. Meteor. Soc., 77, 437–471, 1996. Keshavamurthy, R. N., Power spectra of large scale disturbances of the Indian southwest monsoon, Ind. J. Meteorol. Geophys., 24, 117–124, 1973. Keshavamurthy, R. N., S. V. Kasture, and V. Krishnakumar, 30-50 day oscillation of the monsoon: a new theory, Beitr. Phys. Atmo., 59, 443–454, 1986. Kimoto, M., and M. Ghil, Multiple flow regimes in the northern hemisphere winter. Part I: methodology and hemispheric regimes, J. Atmos. Sci., 50, 2625–2643, 1993. Kondragunta, C. R., On the intraseasonal variations of the Asiatic summer monsoon, Mausam, 41, 11–20, 1990. Krishakumar, K., M. K. Soman, and K. Rupakumar, Seasonal forecasting of Indian summer monsoon rainfall, Weather, 50, 449–467, 1995. Krishnamurthy, V., and J. Shukla, Intraseasonal and interannual variability of rainfall over India, J. Climate, 13, 4366–4377, 2000. Krishnamurti, T. N., and P. Ardunay, The 10 to 20 day westward propagating mode and ’breaks’ in the monsoons, Tellus, 32, 15–26, 1980. Krishnamurti, T. N., and H. N. Bhalme, Oscillations of monsoon system. Part I: Observational aspects, J. Atmos. Sci., 45, 1937–1954, 1976. Krishnamurti, T. N., and D. Subrahmanyam, The 30-50 day mode at 850mb during MONEX, J. Atmos. Sci., 39, 2088–2095, 1982. Kumar, A., and M. P. Hoerling, Prospects and limitations of atmospheric GCM climate predictions, Bull. Amer. Meteor. Soc., 76, 335–345, 1995. Lal, M., K. K. Singh, G. Srinivasan, L. S. Rathore, D. Naidu, and C. N. Tripathy, Growth and yield responses of soybean in Madhyapradesh, India to climate variability and change, Agric. Forest Meterol., 93, 53–, 1999.
Bibliography
103
Latif, M., D. Anderson, T. P. Barnett, M. Cane, R. Kleeman, A. Leetma, J. J. O’Brien, A. Rosati, and E. K. Schneider, A Review of predictability and prediction of ENSO, J. Geophys. Res., 103(C7), 14,375–14,393, 1998. Lawrence, D. M., and P. J. Webster, Interannual variations of the intraseasonal oscillation in the south Asian summer monsoon region, J. Climate, 14, 2910–2922, 2001. Liebmann, B., and C. A. Smith, Description of a complete (interpolated) outgoing longwave radiation dataset, Bull. Amer. Meteor. Soc., 77, 1275–1277, 1996. Liebmann, B., H. H. Hendon, and J. D. Glick, The relationship between tropical cyclones of the Western Pacific and Indian Oceans and the Madden-Julian Oscillation, J. Meteor. Soc. Japan, 72, 401–412, 1994. Lo, F., and H. H. Hendon, Empirical extended-range prediction of the Madden-Julian Oscillation, Mon. Wea. Rev., 128, 2528–2543, 2000. Lorenz, E. N., Atmospheric predictability experiments with a large numerical model, Tellus, 43, 505–513, 1982. Madden, R. A., Estimates of natural variability of time averaged sea level pressure, Mon. Wea. Rev., 104, 942–952, 1976. Madden, R. A., A quantitative approach to long range prediction, J. Geophys. Res., 86, 9817–9825, 1981. Madden, R. A., Reply, Mon. Wea. Rev., 111, 586–589, 1983. Madden, R. A., and P. R. Julian, Observations of the 40-50 day tropical oscillation: A Review, Mon. Wea. Rev., 122, 813–837, 1994. Madden, R. A., and D. J. Shea, Estimates of natural variability of time averaged temperature over the United States, Mon. Wea. Rev., 106, 1695–1703, 1978. Mak, M., Synoptic-scale disturbances in the summer monsoon, in Monsoon Meteorology, edited by C. P. Chang and T. N. Krishnamurti, pp. 435–460, Oxford University Press, New York, 1987. Maloney, E. D., and D. L. Hartmann, Modulation of hurricane activity in the Gulf of Mexico by the Madden-Julian Oscillation, Science, 287, 2002–2004, 2000a.
Bibliography
104
Maloney, E. D., and D. L. Hartmann, Modulation of Eastern North Pacific hurricanes by the Madden-Julian Oscillation, J. Climate, 13, 1451–1460, 2000b. Mehta, V. M., and T. N. Krishnamurti, Interannual Variability of 30-50 day wave motion, J. Meteor. Soc. Japan, 66, 535–548, 1988. Mo, C. K., Adoptive filtering and prediction of intraseasonal oscillations, Mon. Wea. Rev., 129, 802–817, 2001. Mooley, D. A., and J. Shukla, Tracks of low pressure systems which formed over India, adjoining countries, the Bay of Bengal and Arabian Sea in summer monsoon season during the period 1888-1983, Center for Ocean Land Atmosphere Studies, Calverton, MD 20705, USA, 1987, avaialble from J. Shukla, COLA, USA. Mooley, D. A., and J. Shukla, Main features of the westward-moving low pressure systems which form over the Indian region during the summer monsoon season and their relation to the monsoon rainfall, Mausam, 40, 137–152, 1989. Murakami, M., Large scale aspects of deep convective activity over the GATE area, Mon. Wea. Rev., 107, 994–1013, 1979. Murakami, T., and T. Nakazawa, Tropical 45 day oscillations during 1979 northern hemisphere summer, J. Atmos. Sci., 42, 1107–1122, 1985. Murakami, T., T. Nakazawa, and J. He, On the 40-50 day oscillation during 1979 northern hemisphere summer. Part I: Phase propagation, J. Meteor. Soc. Japan, 62, 440–468, 1984. Murakami, T., L. X. Chen, and A. Xie, Relationship among seasonal cycles, low frequency oscillations and transient disturbances as revealed from outgoing long wave radiation, Mon. Wea. Rev., 114, 1456–1465, 1986. Nigam, S., and H. S. Shen, Structure of Oceanic and Atmospheric low frequency variability over the tropical Pacific and Indian Oceans. Part I: COADS Observations, J. Climate, 6, 657–676, 1994. Palmer, T. N., Chaos and the predictability in forecasting monsoons, Proc. Ind. Nat. Sci. Acad., 60A, 57–66, 1994. Parthasarathy, B., A. A. Munot, and D. R. Kothawale, All India monthly and seasonal rainfall series: 1871-1993, Theor. Appl. Climatol., 49, 217–224, 1994.
Bibliography
105
Philander, S. G., Elnino, Lanina and the Southern Oscillation, Academic Press, New York, 1990. Ramamurthy, K., Monsoon of India: Some aspects of ’break’ in the Indian South west monsoon during July and August, India Meteorological Department, New Delhi, 1969, forecasting Manual, Part IV.18.3. Rao, A. V. R. K., A. K. Bohra, and V. Rajeswararao, On the 30-40 day oscillations in the southwest monsoon. A satellite study, Mausam, 41, 51–58, 1990. Rao, Y. P., Southwest Monsoon, India Meteorological Department, New Delhi, 1976, meteorological Monograph. Rasmusson, E. M., and T. H. Carpenter, Variations in the tropical sea surface temperature and surface wind fields associated with Southern Oscillation/El Nino, Mon. Wea. Rev., 110, 354–384, 1982. Rasmusson, E. M., and J. M. Wallace, Meteorological aspects of the ElNino/Southern Oscillation, Science, 222, 1195–1202, 1983. Rowell, D., C. K. Folland, K. Maskell, and M. N. Ward, Variability of summer rainfall over tropical North Africa (1906-1992): Observations and modeling, Quart. J. Roy. Meteor. Soc., 121, 669–704, 1995. Salby, M., H. Hudson, K. Woodberry, and K. Tavaka, Analysis of global cloud imagery from multiple satellites, Bull. Amer. Meteor. Soc., 4, 467–479, 1991. Shea, D. J., and R. A. Madden, Potential for long range prediction of monthly mean surface temperature over north America, J. Climate, 3, 1444–1451, 1990. Short, D. A., and R. F. Cahalan, Interannual variability and climate noise in satellite observed long-wave radiation, Mon. Wea. Rev., 111, 572–577, 1983. Shukla, J., CISK-barotropic-baroclinic instability and growth of monsoon depressions, J. Atmos. Sci., 35, 495–508, 1978. Shukla, J., Dynamical predictability of monthly means, J. Atmos. Sci., 38, 2547–2572, 1981. Shukla, J., Comments on ’Natural variability and predictability’, Mon. Wea. Rev., 111, 581–585, 1983.
Bibliography
106
Shukla, J., Predictability in the midst of chaos: a scientific basis for climate forecasting, Science, 282, 728–731, 1998. Shukla, J., and D. S. Gutzler, Interannual variability and predictability of 500 mb geopotential heights over the northern hemisphere, Mon. Wea. Rev., 111, 1273–1279, 1983. Sikka, D. R., and S. Gadgil, On the maximum cloud zone and the ITCZ over Indian longitude during southwest monsoon, Mon. Wea. Rev., 108, 1840–1853, 1980. Silverman, B. W., Density Estimation for Statistics and Data Analysis, Chapman and Hall, London, 1986. Singh, S. V., and R. H. Kriplani, Potential Predictability of lower-tropospheric monsoon circulation and rainfall over India, Mon. Wea. Rev., 114, 758–763, 1986. Singh, S. V., and R. H. Kriplani, Low frequency intraseasonal oscillations in Indian rainfall and outgoing long wave radiation, Mausam, 41, 217–222, 1990. Singh, S. V., R. H. Kriplani, and D. R. Sikka, Interannual variability of the Madden-Julian Oscillations in Indian summer monsoon rainfall, J. Climate, 5, 973–979, 1992. Sontakke, N. A., D. J. Shea, R. A. Madden, and R. W. Katz, Potential for long-range regional precipitation over India, Mausam, 52, 47–56, 2001. Sperber, K. R., and T. N. Palmer, Interannual tropical rainfall variability in general circulation model simulations associated with Atmospheric Model Intercomparison Project, J. Climate, 9, 2727–2750, 1996. Sperber, K. R., J. M. Slingo, and H. Annamalai, Predictability and the relationship between subseasonal and interannual variability during the Asian summer monsoons, Quart. J. Roy. Meteor. Soc., 126, 2545–2574, 2000. Stern, W., and K. Miyakoda, The feasibility of seasonal forecasts inferred from multiple GCM simulations, J. Climate, 8, 1071–1085, 1995. Swaminathan, M. S., Abnormal monsoons and economic consquences: the Indian experiment, in Monsoons, edited by J. S. Fein and P. L. Stephens, pp. 121–134, Wiley and Sons, New York, 1987. Trenberth, K. E., Some effects of finite sample size and persistence on meteorological statistics. Part I: Autocorrelations, Mon. Wea. Rev., 112, 2359–2368, 1984a.
Bibliography
107
Trenberth, K. E., Some effects of finite sample size and persistence on meteorological statistics. Part II: Potential predictability, Mon. Wea. Rev., 112, 2369–2379, 1984b. Waliser, D. E., C. Jones, C. Schemm, and N. E. Graham, A statistical extended-range tropical forecast model based on the slow evolution of the Madden-Julian Oscillation, J. Climate, 12, 1918–1939, 1999. Wallace, J. M., E. M. Rasmusson, T. P. Mitchell, V. E. Kousky, E. S. Sarachik, and H. von Storch, On the structure and evolution of ENSO-related climate variability in the tropical Pacific, J. Geophys. Res., 103(C7), 14,241–14,259, 1998. Wang, B., and H. Rui, Synoptic Climatology of Transient Tropical Intraseasonal Convection Anomalies: 1975-1985, Met. Atmos. Phys., 44, 43–61, 1990. Webster, P. J., Mechanism of monsoon low-frequency variability: surface hydrological effects, J. Atmos. Sci., 40, 2110–2124, 1983. Webster, P. J., V. O. Magana, T. N. Palmer, J. Shuka, R. T. Tomas, M. Yanai, and T. Yasunari, Monsoons: Processes, predictability and the prospects of prediction, J. Geophys. Res., 103(C7), 14,451–14,510, 1998. Xie, P., and P. A. Arkin, Global precipitation: A 17-year monthly analysis based on guage observations, satellite estimates and numerical model outputs, Bull. Amer. Meteor. Soc., 78, 2539–2558, 1997. Yasunari, T., Cloudiness fluctuation associated with the northern hemisphere summer monsoon, J. Meteor. Soc. Japan, 57, 227–242, 1979. Yasunari, T., A Quasi-stationary appearance of 30-40 day period in the cloudiness fluctuation during summer monsoon over India, J. Meteor. Soc. Japan, 58, 225–229, 1980. Yasunari, T., Structure of an Indian summer monsoon system with around 40-day period, J. Meteor. Soc. Japan, 59, 336–354, 1981. Zheng, X., H. Nakamura, and J. A. Renwick, Potential predictability of seasonal means based on monthly time series of meteorological variables, J. Climate, 13, 2591–2604, 2000. Zweirs, F., and V. V. Kharin, Intercomparison of interannual variability and predictability in : an AMIP diagnostic subproject, Climate Dyn., 14, 517–528, 1998.
