Summer UK Temperature and Its Links to Preceding ... - AMS Journals

0 downloads 0 Views 2MB Size Report
Dec 15, 2003 - snow cover to upcoming climate having been published .... nificant negative correlations are observed from late ..... A scatterplot for London.
4108

JOURNAL OF CLIMATE

VOLUME 16

Summer U.K. Temperature and Its Links to Preceding Eurasian Snow Cover, North Atlantic SSTs, and the NAO BUDONG QIAN

AND

MARK A. SAUNDERS

Department of Space and Climate Physics, University College London, Holmbury St. Mary, United Kingdom (Manuscript received 28 January 2003, in final form 12 May 2003) ABSTRACT Motivated by an attempt to predict summer (June–August) U.K. temperatures, the time-lagged correlations between summer U.K. and European temperatures and prior snow cover, North Atlantic sea surface temperatures (SSTs), and the North Atlantic Oscillation (NAO) are examined. The analysis centers on the 30-yr period 1972– 2001 corresponding to the interval of reliable satellite-derived land snow cover data. A significant association is found between late winter Eurasian snow cover and upcoming summer temperatures over the British Isles and adjacent areas, this link being strongest with January–March snow cover. Significant links are also observed between summer temperatures and the preceding late winter NAO index and with a leading principal component of North Atlantic SST variability. The physical mechanisms underlying these time-lagged correlations are investigated by studying the associated variability in large-scale atmospheric circulation over the Euro–Atlantic sector. Seasonal expansion in the Azores high pressure system may play an important role in the time-lagged relationships. The potential seasonal predictability of summer U.K. temperatures during the period 1972–2001 is assessed by cross-validated hindcasts and usable predictive skill is found. However, the presence and cause of temporal instability in the time-lagged relationships over longer periods of time requires further investigation.

1. Introduction The predictability of climate and weather at leads exceeding ;10 days (the predictability limit of day-today synoptic changes) is linked to surface forcings and their feedback on the atmosphere (e.g., Palmer and Anderson 1994; Goddard et al. 2001). Surface forcings may be either local and/or remote (teleconnected) in origin. Despite considerable recent progress, our understanding of climate forcings and how they feed back to influence climate variability and predictability at different leads and over different predictand time scales remains far from complete. Anomalous sea surface temperature (SST) is the surface forcing most widely related to regional and global climate variability, with the El Nin˜o–Southern Oscillation (ENSO) phenomenon forming the best documented example. Many studies have examined ENSO, its physical mechanism, impacts, and predictability (e.g., see Trenberth et al. 1998; Kumar and Hoerling 1998; Barnston et al. 1999; Diaz et al. 2001, for recent reports). Compared to ENSO and tropical climate variability, the seasonal forcing and predictability of extratropical climate is neither as compelling nor as well

Corresponding author address: Dr. Mark A. Saunders, Department of Space and Climate Physics, University College London, Holmbury St. Mary, Dorking, Surrey RH5 6NT, United Kingdom. E-mail: [email protected]

q 2003 American Meteorological Society

understood. Evidence for predictive skill has been reported, for instance, by Colman (1997) and Colman and Davey (1999) who claim some sound skill for summer temperatures in the United Kingdom and Europe from preceding winter North Atlantic SSTs; by Czaja and Frankignoul (1999, 2002) who showed the observed impact of Atlantic SST anomalies on the North Atlantic Oscillation (NAO) and potential predictability for early winter NAO; and by Saunders and Qian (2002) who evaluated the predictive skill of the winter NAO from preceding North Atlantic SSTs for the previous 51 yr. In contrast, Johansson et al. (1998) concluded that, at leads greater than 1 month, North Atlantic SSTs are not useful for European seasonal predictions, although local SSTs may still be helpful. Compared to anomalous SST forcings and teleconnections, the influence of anomalous snow cover on seasonal climate variability has received relatively little investigation. This is despite the first reports linking snow cover to upcoming climate having been published 100 yr ago with Blanford (1884) and Walker (1910) relating Indian summer monsoon rainfall to prior winter snow cover extent on the northwest Himalayas. Snow cover may force atmospheric circulation and climate in different ways: for example, through the snow–albedo– temperature feedback mechanism because of snow’s high albedo; by causing anomalous temperature gradients; by insulating heat exchanges between the surface and the underlying atmosphere; and by consuming latent

15 DECEMBER 2003

QIAN AND SAUNDERS

heat when melting (Cohen 1994). The lack of an extended time series of accurate hemispheric or global snow cover observations, has been the major factor to limit snow cover–climate studies in the past. However, from the late 1960s, satellite-derived snow cover data started becoming available for observational study (Robinson et al. 1993). Taking advantage of the availability of satellite snow cover data, Hahn and Shukla (1976) followed by Dey and Bhanu Kumar (1982) examined the association between Eurasian winter–spring snow cover and the timing and strength of the Indian summer monsoon. This link has been confirmed and new associations identified between the Asian summer monsoon and prior Himalayan snow cover with the availability of longer time series of quality snow cover data (e.g., Yang and Xu 1994; Bamzai and Shukla 1999; Kripalani and Kulkarni 1999; Ye and Bao 2001), and with numerical model simulations (e.g., Barnett et al. 1988; Yasunari et al. 1991; Douville and Royer 1996). Other documented climate links to snow cover forcings include the influence of autumn Eurasian snow cover on the Northern Hemisphere winter climate (Cohen and Entekhabi 1999; Cohen et al. 2001; Saito et al. 2001), the significant ( p , 0.01) link between summer snow cover over northern North America and northern Eurasia and the upcoming winter North Atlantic Oscillation 1972/73–2001/02 (Saunders et al. 2003), and the feedback from snow cover extent to explain the rise in spring continental surface air temperature since the mid-1970s (Groisman et al. 1994). This paper is not aimed, ambitiously, at discussing general relationships between snow cover, SSTs, and European climate. Rather it examines the period 1972– 2001, using the most updated and reliable data, to determine whether summer European (in particular United Kingdom) temperatures are linked significantly to snow cover and North Atlantic SSTs in preceding seasons and, if so, the physical reasons for their association. Our study also examines the potential application of significant time-lagged relationships for the long-range seasonal prediction of U.K. summer temperatures. The paper is structured as follows. The analyzed datasets are described in section 2. Possible teleconnections between European summer temperatures and Eurasian snow cover, North Atlantic SSTs, and the NAO in preceding seasons are investigated in section 3. The physical basis for significant teleconnections is examined in section 4. The predictive skill for U.K. summer city temperatures 1972–2001 is assessed in section 5, and conclusions are given in section 6. 2. Data a. U.K. temperatures The Central England Temperature (CET) time series comprises monthly surface air temperatures from 1659

4109

to the present. It was assembled first by Manley (1974) and updated by Parker et al. (1992). The CET is a weighted temperature average from up to four stations representative of central England. It is archived and provided by the Hadley Centre of the Met Office. The summer [June–July–August (JJA)] 3-month mean CET is used mainly in this study. We also employ JJA air temperatures for seven U.K. cities 1950–2001 computed from daily temperature time series obtained from the British Atmospheric Data Centre. These stations are, respectively, from north to south, Aberdeen (Dyce), Glasgow (Bishopton), Newcastle (Weather Center), Manchester (Ringway), Birmingham (Coleshill), London (Heathrow), and Bristol (Weather Center). b. Gridded NCEP–NCAR reanalysis data The National Centers for Environmental Prediction– National Center for Atmospheric Research (NCEP– NCAR) reanalysis dataset (Kalnay et al. 1996) is a merger of numerical model forecasts and observations, and— despite the presence of model systematic errors in regions with sparse observations—it forms the most reliable historical global climate dataset available to date. The gridded reanalysis data are available on a grid of 2.58 latitude 3 2.58 longitude. We employ monthly NCEP–NCAR SSTs for the period 1950–2001. The SST field is derived from the optimal interpolation of SST reanalyses (Reynolds and Smith 1994) from 1982, and from the Met Office Global Sea Ice and Sea Surface Temperature (GISST) data prior to 1982. We analyze these data for the North Atlantic sector 08–658N and 08–1008W. Grid cells having sea ice present in any month of the historical record are discarded to minimize any bias in the SSTs due to sea ice. This leaves 621 grid cells to employ in principal component analysis (PCA). Monthly mean sea level pressure (MSLP) fields are also obtained from the NCEP–NCAR reanalysis data for the North Atlantic and European sector for the period 1950–2001. These MSLP data are examined to help understand the physical basis for the links observed between summer (JJA) temperature and Eurasian snow cover extent–SSTs–NAO in preceding seasons. Pan-European 2-m surface air temperatures are also analyzed to confirm the extent to which the teleconnected relationships identified for the U.K. sector are valid elsewhere across Europe. c. Satellite-derived snow cover data We employ monthly Northern Hemisphere land-snow extent records for the 30-yr period from January 1972 to December 2001. These data are provided by Rutgers University (New Brunswick, New Jersey) and computed from weekly snow cover charts produced from visible satellite images by the National Oceanic and Atmospheric Administration (Robinson et al. 1993). Charting

4110

JOURNAL OF CLIMATE

VOLUME 16

improved considerably in 1972 with the satellite deployment of the Advanced Very High Resolution Radiometer having a spatial resolution of 1.0 km. Three separate monthly snow cover indices are available pertaining respectively to the landmass of the Northern Hemisphere, the Eurasian continent, and North America (either with or without the inclusion of Greenland). The snow cover data analyzed comprise the most up-to-date records currently available following corrections made in recent years. d. North Atlantic Oscillation index We use the monthly NAO index provided by the Climatic Research Unit (CRU) at the University of East Anglia. The CRU NAO index is the difference in normalized sea level pressure between southwest Iceland and Gibraltar (Jones et al. 1997). It is a useful index to show the strength of the NAO, especially during winter (Osborn et al. 1999). We examine the NAO index for its potential to predict summer temperature. 3. Time-lagged correlations The motivation for our study is to examine the predictability of summer (JJA) U.K. temperatures from prior snow cover, North Atlantic SSTs, and the NAO. We investigate this by first using a time-lagged correlation analysis to determine the strength and significance of the links between JJA U.K. temperature and each of the above potential predictors for the 30-yr period 1972– 2001. Predictor periods (i.e., the time length of the predictors used) of 1-, 2-, 3-, and 5-months are considered and predictor lead times out to ;6 months are examined. The recent studies by Saunders and Qian (2002) and Saunders et al. (2003) provide a precedent to this approach. Our analysis uses linear detrended time series, unless indicated. Detrending minimizes the influence of time series trends on the strength and significance of the deduced correlations. For example, the presence of trends in both time series may produce a spurious significant correlation even though there is no physical linkage between interannual variability in the two climate parameters. The use of raw (not detrended) time series gives, in all cases, links of similar or stronger magnitude and significance to those described herein. The correlation significances displayed are not corrected for the influence of multiyear-to-decadal signal variability (Davis 1976; Chen 1982). Correction for the latter (by computing the effective number of degrees of freedom in the estimation of the cross correlation by including autocorrelation coefficients in both input time series out to lags of N/2 yr, where N is the time series length 230 in this case), shows that all correlations presented as significant at the 0.001 level remain significant at the 0.01 level after correction for serial correlation.

FIG. 1. The strength, significance, and temporal nature of the link between late winter Eurasian snow extent and the coming summer (JJA) CET during 1972–2001. (a) Correlation between lagged bimonthly Eurasian snow cover extent and summer CET. The negative correlations from raw (solid line) and detrended (dashed line) time series are plotted. Horizontal lines display the confidence levels of nonzero correlation assessed using a two-tailed Student’s t test. (b) Detrended time series of JFM Eurasian snow cover and following JJA CET.

a. Eurasian snow cover Time-lagged correlations are examined between JJA CET and the monthly, bimonthly, and trimonthly extent of snow cover, over respectively Eurasia, North America, and the whole Northern Hemisphere in preceding seasons from the prior September to the contemporaneous August. Significant negative correlations are found with Eurasian snow cover in the late winter months [January–February–March; (JFM); Fig. 1a]. The link to Eurasian snow cover in other prior months is either weak or negligible, as also is the contemporaneous summer link. Time-lagged correlations with snow cover extent in North America are not significant. Correlations with snow cover extent for the whole Northern Hemisphere are significant for late winter snow cover but weaker than those observed for Eurasian late winter snow cover. Since it is clear that the significant correlations between JJA CET and prior Northern Hemisphere snow cover are due to Eurasian snow cover, only Eurasian snow cover will be considered further.

15 DECEMBER 2003

QIAN AND SAUNDERS

Figure 1a shows the temporal evolution of correlation coefficients between JJA CET and preceding consecutive bimonthly Eurasian snow cover extents from April– May (AM) back to September–October (SO). Correlations are computed for the 30-yr period 1972–2001 and are shown using raw time series (solid line) and after linear detrending of each time series (dashed line). Significant negative correlations are observed from late winter to early spring with a peak of 20.55 (20.51 in detrended data) for February–March (FM) snow extent. The latter is significant to 0.005 in both raw and detrended analyses. It was further discovered that the 3month mean (JFM) extent of Eurasian snow cover has the highest correlation of 20.59 (20.55 after detrending) with JJA CET. Detrended time series of JJA CET versus JFM Eurasian snow cover are displayed in Fig. 1b. These show an inverse relation apart from years 1974, 1981, 1983, 1991, and 1993. We investigate how city summer temperatures across the United Kingdom are correlated to prior JFM Eurasian snow cover in Fig. 2a. The analysis is made for seven stations. Significant correlations (at significance level 0.005) are found for all stations based on raw and detrended time series. Little spatial difference in correlation strength is observed with values ranging from 20.52 to 20.59. Detrending makes the correlation at Bristol drop from 20.61 to 20.53. It appears that trends account for a greater percentage of temperature variance in England than in Scotland. The spatial correlation pattern between gridded JJA 2-m air temperatures over Europe from the NCEP– NCAR reanalysis and prior JFM Eurasian snow cover 1972–2001 is shown in Fig. 2b. We find high correlations centered over the United Kingdom with significant negative correlations also covering western France, Belgium, Netherlands, and the Baltic Sea. Significant positive correlations are observed in the Middle East. Tracking the temporal evolution of the correlation patterns between JFM Eurasian snow cover and running 3-month mean temperatures for FMA, MAM, . . . , through to JJA, reveals higher correlations over the U.K. sector with MJJ mean temperature as shown in Fig. 2c. This implies that late winter Eurasian snow cover may influence early summer U.K. temperatures more strongly than late summer U.K. temperatures. This finding may be traced to atmospheric circulation changes as discussed in section 4. b. Atlantic SSTs Atlantic SSTs have been used to predict summer CET and U.K. air temperatures (Colman 1997; Colman and Davey 1999). Reasonable predictive skill was found for July–August (JA) temperatures, especially for the late summer period 14 July–3 September. The predictor was the leading principal component (PC) of January–February (JF) SST variability over the North Atlantic. The SST data grid resolution employed was, however, rather

4111

FIG. 2. Correlation between JFM Eurasian snow cover and summer temperatures during 1972–2001 from detrended time series. (a) JJA U.K. city temperatures. (b) JJA temperatures across Europe and the North Atlantic. (c) MJJ temperatures across Europe and the North Atlantic. The correlation sign is indicated by dashed and solid lines and the correlation significance by shading. Significance levels of 0.10, 0.05, and 0.01 are plotted corresponding, respectively, to correlation values of 0.306, 0.361, and 0.463 (the same in other figures).

coarse, at 108 3 108 latitude–longitude grid in Colman (1997) and 58 3 58 latitude–longitude grid in Colman and Davey (1999). Bearing in mind the potential importance of tropical SSTs for seasonal prediction, it may

4112

JOURNAL OF CLIMATE

VOLUME 16

FIG. 3. The link between summer temperatures and prior North Atlantic SSTs. (a) EOF2 of SONDJ SSTs (explaining 10.7% of the total SST variance by its corresponding PC2). (b) Detrended time series of SONDJ SST PC2 and following JJA CET. (c) Correlations between SONDJ SST PC2 and upcoming JJA U.K. city temperatures during 1972–2001 from detrended time series. (d) Same as (c) but across Europe and the North Atlantic.

be interesting to extend the ocean domain farther southward to the equator. Colman (1997) used the North Atlantic region between 408 and 708N while Colman and Davey (1999) extended to the domain to 208–808N. The SST domain (section 2b) employed in this study is the same as in Saunders and Qian (2002). High latitudes are excluded because the sea surface is covered by sea ice, especially in wintertime. A PCA was performed on normalized 1-, 3-, and 5-month mean SST anomalies in the domain from preceding seasons back to the previous summer for the period 1950–2001. Normalization of SST time series is performed by subtracting the mean and dividing by the standard deviation. This ensures that all grid cells have equal importance in the covariance matrix. In each case the 10 leading PCs of lagged SST variability were correlated with JJA CET for the period 1972–2001 to identify the spatial modes of SST anomalies associated with JJA CET. The PC2 of the September–January (SONDJ) 5month mean SST anomalies was found to have the highest correlation (20.51) with JJA CET. After detrending,

this correlation drops to 20.47, but remains significant to 0.01. The corresponding spatial mode EOF2 of SONDJ SST anomalies is shown in Fig. 3a. It has a tripole pattern, with warm SST anomalies to the southeast of Newfoundland and within 108 of the equator, and cold SST anomalies over the southwest North Atlantic and to the northwest of the United Kingdom being linked to warmer than average upcoming JJA CET. Clearly this SST pattern is related to the wintertime NAO. The correlation between the SONDJ SST PC2 and DJF CRU NAO index is 20.50 (20.49 after detrending). We note that the SST EOF predictor pattern north of 208N in Fig. 3a resembles the EOF pattern from JF SSTs given in Colman and Davey (1999), although the latter’s domain does not extend south of 208N. PCA was also applied on normalized JF SST anomalies but the latter did not show stronger or more significant correlations with JJA CET than obtained with the SONDJ SST PC2. This difference with the results of Colman and Davey (1999) who employ JF SST PC1 to predict summer U.K. and European temperatures may arise

15 DECEMBER 2003

QIAN AND SAUNDERS

4113

from the use of a different SST domain: SST data of different resolution and a different record period. However, it may also point to the instability in the link between lagged North Atlantic SST anomalies and U.K. summer temperatures. Detrended time series of JJA CET versus prior SONDJ SST PC2 are displayed in Fig. 3b. These show a general inverse relation. Correlation coefficients (after detrending) between JJA surface air temperatures at U.K. cities and the prior SONDJ SST PC2 are displayed in Fig. 3c. These show stronger correlations in England (about 20.46) and lower correlations in Scotland. The highest correlation is observed in western France (Fig. 3d). We also note that the summer temperature correlations with the prior SONDJ SST PC2 are generally weaker than with the preceding JFM Eurasian snow cover extent. However, the SONDJ SST PC2 mode is linked more significantly to summer surface air temperature variability over parts of the North Atlantic than is prior JFM Eurasian snow cover. c. North Atlantic Oscillation Since the NAO is the dominant mode of winter atmospheric variability over the North Atlantic and European sector it may be linked to variability in Eurasian snow cover extent and SST anomalies in the North Atlantic. As mentioned already, the SONDJ SST PC2 has a significant correlation with the DJF CRU NAO index. The correlation between JFM Eurasian snow cover extent and the DJF CRU NAO index is also significant to 0.01 with a correlation coefficient of 20.54 (20.53 after detrending). Thus, it may be expected that the winter NAO may be correlated to U.K. summer temperatures. Correlating JJA CET with prior monthly, bimonthly, and trimonthly CRU NAO index values in preceding seasons, we find that the JF CRU NAO index has the highest correlation of 0.57 (0.54 after detrending) with JJA CET. This link is significant to 0.01 even after correction for time series autocorrelation. Figure 4a shows the JJA CET and JF CRU NAO index detrended time series, with consistent fluctuations being observed. Correlations of JJA temperatures in U.K. cities with the prior JF CRU NAO index are shown in Fig. 4b. Significant values are observed for all stations. The lowest correlation of 0.43 (0.05 significance level) is observed in Aberdeen while the highest value of 0.56 is seen in Birmingham and Bristol. The spatial correlation pattern between gridded JJA air temperatures over Europe and the prior JF CRU NAO index 1972–2001 is shown in Fig. 4c. This exhibits a similar pattern but of opposite sign to Fig. 2b having a high correlation center over the south of the British Isles. The correlation link between the JF CRU NAO index and MJJ temperatures (not shown) is even stronger than with JJA temperatures; a result mirroring that found for JFM Eurasian snow

FIG. 4. The link between the JF CRU NAO index and coming summer temperatures during 1972–2001. (a) Detrended time series of the JF CRU NAO index and JJA CET. (b) The correlation from detrended time series with JJA U.K. temperatures. (c) The correlation pattern and significance with JJA temperatures across Europe and the North Atlantic from detrended time series.

cover. Finally, we note that the correlations of JJA temperatures in the seven U.K. cities are higher with the JFM Eurasian snow cover than with the JF CRU NAO index, except in Bristol.

4114

JOURNAL OF CLIMATE

VOLUME 16

4. Some explanations for the time-lagged correlations As demonstrated in section 3, JJA CET and temperatures in seven U.K. cities are correlated significantly to the JF CRU NAO index, JFM Eurasian snow cover extent, and North Atlantic SONDJ SST anomalies for the period 1972–2001. It is important to obtain a thorough understanding of the physical mechanisms behind these correlations, both for understanding the physical behavior of the climate system and for applying results to the statistical seasonal forecasting of summer temperatures. This physical understanding becomes even more important since only a 30-yr record is available for the snow cover data. Surface air temperature variability must be associated with corresponding variability in atmospheric circulation. For example, above-normal summer temperatures can be expected if an anticyclone prevails more than normal over a region in summer. In western Europe, the strength and position of the Azores subtropical high pressure system may play an important role in summer temperature variability. Any delay in the seasonal northeast expansion of the Azores high from spring to summer, can result in below-normal summer temperatures in the region of its northeastern boundary. We will show evidence below, using correlation analysis with MSLP fields, to support this notion. a. JFM Eurasian snow cover We examined spatial correlation patterns between gridded three-month mean MSLP fields (from JFM, FMA, . . . , through to OND) over the North Atlantic and European sector and JFM Eurasian snow cover extent 1972–2001. A clear NAO signature exists in the contemporaneous correlation pattern between JFM MSLP and JFM Eurasian snow cover (Fig. 5a). This reflects the dominant NAO forcing of late winter Eurasian snow cover extent (Bamzai 2003). The correlation coefficient between the JF CRU NAO index and JFM Eurasian snow cover is 20.74 (20.72 after detrending). A negative (positive) NAO index is associated with more (less) Eurasian snow cover due to colder (warmer) conditions prevailing over high latitudes in western Eurasia. With the arrival of spring, the NAO pattern in Fig. 5a decays quickly and a new correlation pattern strengthens. Figure 5b shows the spatial correlation pattern between AMJ MSLP fields and lagged JFM Eurasian snow cover. The new pattern exhibits a positive correlation center over the subtropical Atlantic and a negative center over the North Sea. It reaches its strongest phase for MJJ MSLP fields (Fig. 5c). The pattern is consistent with more late winter Eurasian snow cover being associated with a more southwestward-located Azores high pressure in spring and early summer. Above-average late winter Eurasian snow cover may

FIG. 5. The correlation pattern and significance after detrending between JFM Eurasian snow cover and upcoming 3-month mean MSLP fields in the Euro–Atlantic sector during 1972–2001 for (a) JFM, (b) AMJ, and (c) MJJ.

delay the seasonal expansion of the Azores high since more snow cover on the continent can slow the warming up of the landmass during spring and early summer due to snow’s high albedo and consumption of latent heat when melting. A colder European landmass would be unfavorable for the seasonal northeast extension of the Azores high. Further spatial correlation results between JFM Eurasian snow cover and seasonal MSLP fields (not shown)

15 DECEMBER 2003

QIAN AND SAUNDERS

4115

indicate that no significant correlations exist after JAS. This implies that the climate influence of late winter snow cover lasts about 6 months. b. Atlantic SST anomalies The spatial correlation pattern between JFM MSLP fields and the prior SONDJ SST PC2 1972–2001 is shown in Fig. 6a. A weak positive correlation region occurs south of Greenland and a weak negative area off the northwest African coast. This pattern indicates that the SST PC is related inversely to the NAO with positive values of the PC corresponding to a negative NAO. The correlation coefficients are significant to 0.05. The correlation coefficient between SONDJ SST PC2 and the CRU NAO index is 20.49 for DJF NAO and 20.33 for JF NAO. Figure 6b shows the spatial correlation pattern between MJJ MSLP fields and the prior SONDJ SST PC2. Areas of positive (negative) correlation exist over the eastern Atlantic (North Sea) forming a pattern similar to that in Fig. 5c between MJJ MSLP fields and JFM Eurasian snow cover. The negative correlations are weaker than those in Fig. 5c, and this may explain why the correlation between JJA CET and this SST PC is weaker than the corresponding correlation with JFM Eurasian snow cover. Stronger positive correlations occur over the eastern Atlantic in JJA (Fig. 6c) than in MJJ, which may imply a stronger but less expanded Azores high. Further spatial correlation results (not shown) indicate a significant negative correlation center with late summer MSLP fields to the south of Newfoundland while the significant correlations over the eastern Atlantic and Europe have disappeared. Significant correlations are not observed in the region from SON. c. North Atlantic Oscillation Let us consider the physical basis for how the winter NAO can influence summer temperatures over the British Isles by considering first the temporal stability of the association. The wintertime (DJF) CRU NAO index is highly correlated to the DJF CET with a correlation coefficient of 0.74 for the period 1824–2001. This correlation is very stable with time, the running 30-yr correlation coefficients ranging between 0.66 and 0.86 over this period. In contrast, the correlation between the JF CRU NAO index and JJA CET is not stable through time. Figure 7 displays the temporal variation in the running 30-yr correlation coefficients between the JF CRU NAO index and JJA CET during 1824–2001. Significant positive correlations are only observed for the recent 35 yr or so. Positive correlations (but barely significant at the 0.10 level) are also found from 1824 to the late nineteenth century. However, weak negative correlations are seen from the late nineteenth century to the mid twentieth century. It is worth reiterating that

FIG. 6. The correlation pattern and significance after detrending between the SONDJ SST PC2 and upcoming 3-month mean MSLP fields in the Euro–Atlantic sector during 1972–2001 for (a) JFM, (b) MJJ, and (c) JJA.

the significant positive correlation since 1970 does not arise from temporal trends in recent decades. Spatial correlation patterns between forthcoming 3month mean MSLP fields and the JF CRU NAO index are displayed in Fig. 8. The plots correspond to MAM, MJJ, and JJA MSLP fields for Figs. 8a,b,c respectively. Figure 8a shows a positive correlation center over the British Isles, implying a stronger than normal northeast expansion of the Azores high in springs following late

4116

JOURNAL OF CLIMATE

VOLUME 16

FIG. 7. Running 30-yr correlation between the JF CRU NAO index and JJA CET for the period 1824–2001. Each 30-yr correlation coefficient is plotted at the time of its 16th year so, e.g., the value for 1824–53 is plotted at 1839.

winters with a positive NAO phase. By MJJ (Fig. 8b), the negative correlation center south of the positive center has moved westward and intensified while the positive center has moved slightly eastward over the North Sea. This pattern remains consistent with a northeast placement of the Azores high following a late winter of NAO positive phase. A similar pattern in also observed with JJA MSLP fields (Fig. 8c). Comparing Fig. 8 to Fig. 5 shows similar MSLP correlation structures for both the JF CRU NAO index and for JFM Eurasian snow cover. However the correlation center over the British Isles and North Sea is weaker for NAO than for snow cover. Because the JF NAO is strongly correlated to JFM snow cover it is possible that the correlation patterns between late spring–summer MSLP fields and the JF CRU NAO index are merely reflecting the influence of late winter Eurasian snow cover on the atmospheric circulation in forthcoming seasons. If this is the case it suggests that feedbacks from snow cover take an important role in the ‘‘persistence’’ of the temperature–winter NAO relationship from winter to summer. However, this cannot explain why the relationship between summer temperatures and the winter NAO is strong only in recent decades. It is possible that with a longer data series of reliable snow cover data, the JFM snow cover to JJA temperature link may also become unstable. In the absence of such data, the physical explanation offered herein for linking late winter Eurasian snow cover to U.K. summer temperatures seems reasonable. The temporal instability of the relationship between summer temperature and winter NAO remains an open question requiring further investigation. d. Summer temperature variability in western Europe As demonstrated in section 3, the significant correlations 1972–2001 between JJA temperatures in western Europe and prior Eurasian snow cover, North Atlantic

FIG. 8. The correlation pattern and significance after detrending between the JF CRU NAO index and upcoming Euro–Atlantic MSLP fields 1972–2001 for (a) MAM, (b) MJJ, and (c) JJA.

SSTs, and the NAO are centered on the British Isles and its surrounding regions. Why is this area preferred? From section 4a, changes in the seasonal northeast expansion of the Azores high may preferentially affect this region thereby making it susceptible to higher summer temperature variability than other European regions. To examine this further we performed a PCA on standardized JJA gridded temperatures from the NCEP–NCAR reanalysis for the period 1972–2001 over the European region 358–708N, 208E–158W. The leading PC of tem-

15 DECEMBER 2003

QIAN AND SAUNDERS

4117

perature variability (TPC1) explains 30% of the total variance. Its EOF pattern is displayed in Fig. 9a and shows maximum strength over the British Isles and adjacent regions of Europe, thus confirming that these European areas do indeed witness the highest variability in summer temperatures. Consistent with this we note that the correlation between TPC1 and JJA CET is 0.83. TPC1 is also significantly linked to JFM Eurasian snow cover, the JF CRU NAO index, and to the SONDJ SST PC2, with correlation coefficients of 20.64, 0.57, and 20.42, respectively. The correlation pattern between TPC1 and JJA MSLP is shown in Fig. 9b. This indicates a significant link between TPC1 and variability in the Azores high pressure. In particular, high summer temperatures over the British Isles and the near-European continent appear related to a northeast expanded Azores high and to a northwest-shifted Icelandic low. A similar correlation pattern is observed between TPC1 and MJJ MSLP fields (Fig. 9c), although the negative correlation center is not seen over the Greenland sector implying that movement of the Icelandic low may be a large-scale atmospheric circulation response to the northeast expansion of the Azores high. In summary, the British Isles and the near-European continent may be the ‘‘preferred’’ area for European summer temperature predictability because it corresponds to the leading mode of JJA temperature variability across Europe. 5. Potential predictability of summer U.K. temperatures a. Cross validation The significant correlations between summer temperatures and prior-season Eurasian snow cover, SSTs and the NAO, imply the potential for summer seasonal temperature prediction. In this section we assess the potential predictability of JJA CET and JJA mean temperature for seven U.K. cities from standard cross-validated hindcasts (Michaelsen 1987; Barnston 1994; Wilks 1995) for the recent 30-yr period (1972–2001) of snow cover data availability. Independent replicated real-time forecasts (Saunders and Qian 2002) are not performed due to lack of sufficient data for calibrating prediction models. Potential summer temperature predictability may also be expected in other (non-U.K.) regions of high correlation as described in section 3. b. Predictors, detrending, and skill measures

FIG. 9. Summer temperature variability over western Europe and its links to Euro–Atlantic MSLP variability. (a) EOF1 of JJA 2-m air temperatures over western Europe 1972–2001 (its corresponding PC, TPC1, explaining 30% of the total temperature variance). The correlation pattern and significance after detrending between JJA TPC1 and (b) JJA MSLP and (c) MJJ MSLP over the Euro–Atlantic sector.

We assess summer temperature predictability using two models: 1) with one predictor (JFM Eurasian snow cover) and 2) with two predictors (JFM Eurasian snow cover and SONDJ SST PC2). We do not use the JF CRU NAO index since it is highly correlated with JFM Eurasian snow cover, and furthermore we suspect that the

high correlations between summer temperatures and the winter NAO may be mirroring the correlations between summer temperatures and JFM snow cover. We include the SONDJ SST PC2 as a predictor in the second model

4118

JOURNAL OF CLIMATE

VOLUME 16

TABLE 1. Cross-validated predictive skill during 1972–2001 for JJA CET and U.K. city temperatures. Numbers with and without parentheses are the skill estimated from regression models with raw and detrended time series, respectively. U.K. cities CET Aberdeen Glasgow Newcastle Manchester Birmingham London Bristol 0.93 0.67 0.74 0.82 0.94 0.99 1.08 1.12

s (8C) Snow cover prediction model

r rmse (8C)

Snow cover and SST prediction model

r rmse (8C)

0.53 (0.53) 0.79 (0.79)

0.52 (0.53) 0.58 (0.57)

0.45 (0.49) 0.66 (0.64)

0.58 (0.57) 0.67 (0.67)

0.48 (0.48) 0.82 (0.83)

0.58 (0.58) 0.81 (0.81)

0.58 (0.58) 0.88 (0.88)

0.56 (0.59) 0.94 (0.91)

0.61 (0.63) 0.74 (0.72)

0.52 (0.57) 0.58 (0.56)

0.41 (0.49) 0.68 (0.65)

0.63 (0.62) 0.64 (0.64)

0.56 (0.60) 0.78 (0.75)

0.65 (0.67) 0.75 (0.74)

0.66 (0.66) 0.81 (0.82)

0.68 (0.67) 0.83 (0.84)

because its time series is not correlated with JFM Eurasian snow cover (0.12 after detrending). The presence of significant linear trends in the predictand time series (e.g., JJA CET has a linear trend of 0.268C decade 21 during 1972–2001, while at Bristol the trend is 0.548C decade 21 ) suggests predictive skill may be anticipated merely from predicting the trend. To remove trends in the predictor and predictand time series and their effect on the deduced JJA temperature predictability we computed cross-validated skills using linear regression models calibrated with both raw and detrended and predictand time series. It turns out that raw and detrended models have very similar hindcast skill. We compute the predictive deterministic skill for JJA temperature using two skill measures: the correlation (r) between the forecast and observed JJA temperature values, and the root-mean-square error (rmse) between forecast and actual (e.g., Wilks 1995). A model having a high correlation and a low rmse (relative to the corresponding standard deviation, s, in the observations) implies a good prediction. c. Cross-validated predictive skills for summer temperatures Cross-validated predictive skills for JJA CET and summer temperatures in seven U.K. cities are listed in Table 1. These are computed using the two models and two skill measures described above. Values are shown separately for models calibrated using raw time series data and detrended time series data; however, it is found that both approaches produce similar results. Table 1 also includes the standard deviation, s, in each predictand temperature time series during 1972–2001 for reference. The JFM Eurasian snow cover model anticipates, in early April, the upcoming JJA temperature with a correlation between 0.45 and 0.58 (0.48 and 0.59 using detrended time series) with higher values in southern England and the lowest in Scotland. The JFM Eurasian snow cover and SONDJ SST PC2 model anticipates, also from early April, the coming summer temperatures with a correlation between 0.41 and 0.68 (0.49 and 0.67 using detrended time series) again with the lowest values

in Scotland and highest in southern England. These regional differences agree with expectations based on the findings in Figs. 2a and 3a. Using the interpretation for the value of a forecast based on its correlation skill (Barnston and Ropelewski 1992) both models show JJA temperature forecast skill that is ‘‘marginal but useable.’’ Including the SONDJ SST PC2 improves the predictive skill, except in Scotland where the SONDJ SST PC2 does not have a significant correlation with summer temperatures. In England, the combined snow cover and SST prediction model gives correlations during 1972–2001 exceeding 0.6. A scatterplot for London Heathrow of observed JJA temperatures against the values predicted from cross validation is shown in Fig. 10. This shows that both cool and warm summers are predicted with some success, although a few warm years are underestimated. Inspection of Table 1 shows that a higher correlation is accompanied by a larger decrease in rmse relative to s, indicating similarity of different skill measures. The rmses of predictions from the JFM Eurasian snow cover and SONDJ SST PC2 model reflect an improvement of around 20% over a climatological forecast using a 1971–2000 climate norm. This model correctly anticipates the correct temperature anomaly sign in around 70% of the summers. Clearly a degree of predictive skill is contributed from trends, especially in southern Britain where the trend in summer temperatures is stronger than in the north. However, the prediction skill from the trend component of summer temperatures is rather low, even in the south. 6. Conclusions We have identified statistically significant links to summer European (in particular United Kingdom) temperatures from prior JFM Eurasian snow cover, SONDJ North Atlantic SSTs, and the JF North Atlantic Oscillation for the period 1972–2001. These links are present in both raw and linear detrended time series. The snow cover and NAO links to JJA temperatures are slightly stronger than the SST link. The associations offer useful predictive skill from early April; a lead of 2 months

15 DECEMBER 2003

4119

QIAN AND SAUNDERS

the opposite effect. It will be interesting to run numerical climate models with prescribed snow cover conditions to verify whether late winter Eurasian snow cover influences the early summer northeast expansion of the Azores high pressure system as suggested. Acknowledgments. Budong Qian was supported by the U.K. Natural Environment Research Council Grant GR3/R9925. We thank David Robinson and Thomas Estilow of the Snow Data Resource Center at Rutgers University (New Jersey) for providing the most updated snow cover data. We also thank Judah Cohen for helpful comments. We acknowledge NOAA–CIRES, Climate Diagnostics Center, Boulder, Colorado, for the NCEP– NCAR Global Reanalysis Project Data. REFERENCES FIG. 10. Scatterplot of predicted JJA temperatures for Heathrow, London using cross validation against observed JJA temperature values for 1972–2001. The predictions are made using data through the end of Mar.

before the start of summer. Cross-validated 30-yr hindcasts for U.K. summer temperatures from preceding snow cover and North Atlantic SSTs have a correlation skill for English cities of over 0.60 and an rmse improvement over a climatology of 20%. We have focused on the 1972–2001 period because this corresponds to the interval of reliable satellite-derived snow cover data. A longer data series would provide greater confidence in the temporal stability of the identified links. Examining the association between the JF CRU NAO index and JJA CET 1824–2001 shows that the link is not temporally stable and that it is strongest during our study period. The SONDJ SST PC2 link to JJA CET is also stronger post-1970 than between 1950 and 1970. The presence and cause of temporal instability in the time-lagged relationships over longer periods of time requires investigation. Are the stronger links since 1970 related, for example, to global warming or to epochs of stronger North Atlantic decadal variability? Understanding the reason(s) for instability will be crucial to gain sufficient confidence to apply these associations in the seasonal prediction of summer U.K. and European temperatures. We suggest the physical basis for the teleconnections from late winter to spring to early summer may be associated with the seasonal northeast expansion of the Azores high pressure system. When snow cover over the European continent in late winter is excessive, spring and early summer warming may be delayed due to the effect of snow’s high albedo and consumption of latent heat when melting. The resulting cooler temperatures may delay the seasonal expansion of the Azores high leading, in turn, to lower than normal summer temperatures over the British Isles and surrounding areas. Small Eurasian snow cover in late winter would have

Bamzai, A. S., 2003: Relationship of snow cover variability and Arctic Oscillation index on a hierarchy of time scales. Int. J. Climatol., 23, 131–142. ——, and J. Shukla, 1999: Relation between Eurasian snow cover, snow depth, and the Indian summer monsoon: An observational study. J. Climate, 12, 3117–3132. Barnett, T. P., L. Duemenil, U. Schlese, and E. Roeckner, 1988: The effect of Eurasian snow cover on global climate. Science, 239, 504–507. Barnston, A. G., 1994: Linear statistical short-term climate predictive skill in the Northern Hemisphere. J. Climate, 7, 1513–1564. ——, and C. F. Ropelewski, 1992: Prediction of ENSO episodes using canonical correlation analysis. J. Climate, 5, 1316–1345. ——, Y. He, and M. H. Glantz, 1999: Predictive skill of statistical and dynamical climate models in SST forecasts during the 1997/ 98 El Nin˜o episode and the 1998 La Nin˜a onset. Bull. Amer. Meteor. Soc., 80, 217–244. Blanford, H. F., 1884: On the connexion of the Himalayan snowfall with dry winds and seasons of drought in India. Proc. Roy. Soc. London, 37, 3–22. Chen, W. Y., 1982: Fluctuations in Northern Hemisphere 700-mb height field associated with the Southern Oscillation. Mon. Wea. Rev., 110, 808–823. Cohen, J., 1994: Snow cover and climate. Weather, 49, 150–156. ——, and D. Entekhabi, 1999: Eurasian snow cover variability and Northern Hemisphere climate predictability. Geophys. Res. Lett., 26, 345–348. ——, K. Saito, and D. Entekhabi, 2001: The role of the Siberian high in the Northern Hemisphere climate variability. Geophys. Res. Lett., 28, 299–302. Colman, A., 1997: Prediction of summer central England temperature from preceding North Atlantic winter sea surface temperature. Int. J. Climatol., 17, 1285–1300. ——, and M. K. Davey, 1999: Prediction of summer temperature, rainfall and pressure in Europe from preceding winter North Atlantic Ocean temperature. Int. J. Climatol., 19, 513–536. Czaja, A., and C. Frankignoul, 1999: Influence of the North Atlantic SST anomalies on the atmospheric circulation. Geophys. Res. Lett., 26, 2969–2972. ——, and ——, 2002: Observed impact of Atlantic SST anomalies on the North Atlantic Oscillation. J. Climate, 15, 606–613. Davis, R. E., 1976: Predictability of sea surface temperature and sea level pressure anomalies over the North Atlantic Ocean. J. Phys. Oceanogr., 6, 249–266. Dey, B., and O. S. R. U. Bhanu Kumar, 1982: An apparent relationship between Eurasian spring snow cover and the advance period of the Indian summer monsoon. J. Appl. Meteor., 21, 1929–1932. Diaz, H. F., M. P. Hoerling, and J. K. Eischeid, 2001: ENSO vari-

4120

JOURNAL OF CLIMATE

ability, teleconnections and climate change. Int. J. Climatol., 21, 1845–1862. Douville, H., and J.-F. Royer, 1996: Sensitivity of the Asian summer monsoon to an anomalous Eurasian snow cover within the Meteo-France GCM. Climate Dyn., 12, 449–466. Goddard, L., S. J. Mason, S. E. Zebiak, C. F. Ropelewski, R. Basher, and M. A. Cane, 2001: Current approaches to seasonal-to-interannual climate predictions. Int. J. Climatol., 21, 1111–1152. Groisman, P. Ya, T. R. Karl, and R. W. Knight, 1994: Observed impact of snow cover on the heat balance and the rise of continental spring temperatures. Science, 263, 198–200. Hahn, D. G., and J. Shukla, 1976: An apparent relationship between Eurasian snow cover and Indian monsoon rainfall. J. Atmos. Sci., 33, 2461–2462. Johansson, A., A. Barnston, S. Saha, and H. van den Dool, 1998: On the level and origin of seasonal forecast skill in northern Europe. J. Atmos. Sci., 55, 103–127. Jones, P. D., T. Jo´nsson, and D. Wheeler, 1997: Extension to the North Atlantic Oscillation using early instrumental pressure observations from Gibraltar and South-West Iceland. Int. J. Climatol., 17, 1433–1450. Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471. Kripalani, R. H., and A. Kuikarni, 1999: Climatology and variability of historical Soviet snow depth data: Some new perspectives in snow–Indian monsoon teleconnections. Climate Dyn., 15, 475– 489. Kumar, A., and M. Hoerling, 1998: Annual cycle of Pacific–North American predictability associated with different phases of ENSO. J. Climate, 11, 3295–3308. Manley, G., 1974: Central England temperatures, monthly means 1659 to 1973. Quart. J. Roy. Meteor. Soc., 100, 389–405. Michaelsen, J., 1987: Cross-validation in statistical climate forecast models. J. Climate Appl. Meteor., 26, 1589–1600. Osborn, T. J., K. R. Briffa, S. F. B. Telt, P. D. Jones, and R. M. Trigo, 1999: Evaluation of the North Atlantic Oscillation as simulated by a coupled climate model. Climate Dyn., 15, 685–702. Palmer, T. N., and D. L. T. Anderson, 1994: The prospects for seasonal

VOLUME 16

forecasting—A review paper. Quart. J. Roy. Meteor. Soc., 120, 755–793. Parker, D. E., T. P. Legg, and C. K. Folland, 1992: A new daily central England temperature series, 1772–1991. Int. J. Climatol., 12, 317–342. Reynolds, R. W., and T. M. Smith, 1994: Improved global sea surface temperature analyses using optimum interpolation. J. Climate, 7, 929–948. Robinson, D. A., K. F. Dewey, and R. R. Heim Jr., 1993: Global snow cover monitoring: An update. Bull. Amer. Meteor. Soc., 74, 1689–1696. Saito, K., J. Cohen, and D. Entekhabi, 2001: Evolution of atmospheric response to early-season Eurasian snow cover anomalies. Mon. Wea. Rev., 129, 2746–2760. Saunders, M. A., and B. Qian, 2002: Seasonal predictability of the winter NAO from North Atlantic sea surface temperatures. Geophys. Res. Lett., 29, 2049, doi:10.1029/2002GL014952. ——, ——, and B. Lloyd-Hughes, 2003: Summer snow extent heralding of the winter North Atlantic Oscillation. Geophys. Res. Lett., 30, 1378, doi:10.1029/2002GL016832. Trenberth, K. E., G. W. Branstator, D. Karoly, A. Kumar, N.-C. Lau, and C. Ropelewski, 1998: Progress during TOGA in understanding and modelling global teleconnections associated with tropical sea surface temperatures. J. Geophys. Res., 103, 14 291–14 324. Walker, G. R., 1910: Correlations in seasonal variations of weather. Mem. India Meteor. Dept., 21 (II), 22–45. Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences. Academic Press, 467 pp. Yang, S., and L. Xu, 1994: Linkage between Eurasian winter snow cover and regional Chinese summer rainfall. Int. J. Climatol., 14, 739–750. Yasunari, T., A. Kitoh, and T. Tokioka, 1991: Local and remote responses to excessive snow mass over Eurasia appearing in the northern spring and summer climate—A study with MRI GCM. J. Meteor. Soc. Japan, 69, 473–487. Ye, H., and Z. Bao, 2001: Lagged teleconnections between snow depth in northern Eurasia, rainfall in southeast Asia and seasurface temperatures over the tropical Pacific Ocean. Int. J. Climatol., 21, 1607–1621.