The characteristics of spatial and temporal variations of land surface temperature in the Yangtze River Delta a, b
, Zhihao Qin b, c, Hongxiu Wanb, d School of Remote Sensing, Nanjing University of Information Science and Technology, Nanjing, P. b R. China 210044; International Institute of Earth System Science, Nanjing University, c Nanjing, P. R. China 210093; Institute of Natural Resources and Regional Planning, Chinese d Academy of Agricultural Sciences, Beijing, P. R. China 100081; Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences, Nanjing, P. R. China 210008
a
Yongming Xu *
ABSTRACT Land s urface t emperature (L ST) is one of the key pa rameters in the at mosphere-land ene rgy a nd water tra nsfers. An understanding of the s patial and tem poral variations of l and s urface te mperature is i mportant to broad re search fields, including cl imate, ve getation, hy drology, e tc. In t his paper, t he cl oud contamination of M ODIS L ST p roduct was analyzed f irst, and sho wed that th ere ar e numerous data gaps in MOD IS 8-d ay co mposite LST pr oduct, ind icating the necessity of data interpolation. Then the Harmonic Analysis of Time-Series (HANTS) algorithm was applied to the LST time-series to rebuild cloud-free images and to distill harmonic components. According to the harmonic characters and reconstruct LST, the spatial and temporal variations of land surface temperature in the Yangtze River Delta were studied. Keywords: LST time series; MODIS; HANTS; Temporal and spatial variations
1. INTRODUCTION Land surface temperature (LST) is one of the key factors in the energy and water transfers between the atmosphere and the ground [1-3]. The temp-spatial patterns of LST not only reflect the variations of climatic factors, such as solar radiance, precipitation, but also reveal the characters of land surface [4]. An understanding of the spatial and temporal variations of land s urface tem perature is im portant to broad research fields, includi ng clim ate, veget ation, hydrology, etc. In m any instances, satellite re mote sensing is t he only m eans to comprehensively assess the land surface te mperature at l arge scale. In r ecent years, r emote sensing h as been w idely u sed in th e study o f land surface temperature. Mo st r esearches focused on the LST retrieval alg orithms an d urb an h eat island, howev er few pro cedures ex ist th at use satellit e observations for LST spatial and temporal variations [5-8]. The Yang tze River Delta is located in the eastern China, which is one of the most economically developed regions in China. In recent years the delta experienced rapid urbanization and industrialization. Intervention of human activities on the natural environment has significant influence on climate, environmental and ecological changes, making this delta to become a suitable area for the study of human-land relationship. The object of this paper is to understanding the seasonal variation and spatial pattern of land surface temperature in the study area.
2. DATA AND PROCESSING The m ain satellite d ata u sed in th is st udy were t he 8-day co mposite land surface t emperature data d erived from AQUA/MODIS. The 1km LST d ata (MYD11A2) were calculated by a co mmon split window approach from the linear difference of the bri ghtness t emperature between ba nd 31 and 3 2. L ST dat a were collected fr om January 2005 to December 2005, thus, 46 time s cenes. In addition, a 9-category land cove r map of the study area was a lso employed to illustrate the surface conditions. *
[email protected]; phone 8625-58699865; fax 8625-58731191
MIPPR 2009: Remote Sensing and GIS Data Processing and Other Applications, edited by Henri Maître, Hong Sun, Bangjun Lei, Jufu Feng, Proc. of SPIE Vol. 7498, 749809 · © 2009 SPIE · CCC code: 0277-786X/09/$18 · doi: 10.1117/12.831995 Proc. of SPIE Vol. 7498 749809-1
Due to the cloud cover and other interfering factors, there were abundant data gaps in remote sensing observations. First, the data gaps of LST dataset from MODIS were a nalyzed spatially an d te mporally. Th e nu mber of invalid pixels was displayed in space a nd ti me. The tem poral curve gave the proportional data ava ilability in each ti me scene (figure 1). From this figure we could find that there were more than 60% pixels were invalid sometime, which showed the large spatial data gaps in these scenes. The spatial display depicted the proportion of invalid values throughout the year p er pixel (Figure 2), from which we could easily find that the proportions of invalid values of most pixels were more than 16%. There were great deals of cl oud-contaminated ob servations in bo th tem poral and sp atial scales, sug gesting th e necessity of rebuilding cloud-free LST time series.
Fig. 1. Temporal quality analysis of MODIS LST product
Fig. 2. Spatial quality analysis of MODIS LST product
A ha rmonic al gorithm nam ed H ANTS ( Harmonic Anal ysis of N umerical Tim e-Series) was a dopted t o rec onstruct gapless the LST time-series. It considers only the most significant frequencies expected to be present in the time profiles,
Proc. of SPIE Vol. 7498 749809-2
and applies a least squ ares curve fitting procedure based on harmonic components [9, 10]. In this way cloudy observations have been removed and then replaced by the HANTS fitting values. HANTS was developed with the application to vegetation time-series such as NDVI and EVI at high temporal resolution. LST dataset also could be analyzed by HANTS program because its periodical characters on a yearly basis. In this paper, the num ber of frequencies u sed by HA NTS was set at 2: the annual harm onic (f requency= 1) and the hal f-yearly harmonic ( frequency= 2). Eac h ha rmonic ha s t wo parameters: t he am plitude a nd p hase, rep resented i n u nits o f K a nd days o f y ear. The i nvalid data reject ion t hreshold was s et at 25 5K, m eaning t hat L ST val ue below 255K s hould be rejected. The fit error t olerance was set at 5K, m eaning that th e absolute error of th e remaining observations should be smaller th an 5K with resp ect to th e curren t fitted curv e in HANTS iteratio n. Figure 3a and 3b show th e orig inal an d HANTS reconstructed LST t ime series fo r a p ixel with few inv alid d ata (12 1.55ºE, 31 .60ºN) an d a pi xel wi th m ore invalid dat a (121.47ºE, 3 1.27ºN) respectively. Through HANT analysis, we got not o nly t he rec onstructed cl oud-free LST series, by also am plitude an d phase of t he 1st and 2nd harmonics. Harm onic a mplitude d enotes th e periodic fluctuating range of LST time-series, and harmonic phase illustrates the timing of the LST peak.
Fig. 3. The original and the HANTS reconstructed LST time-series (a) a pixel with few invalid data (121.55ºE, 31.60ºN) (b) a pixel with more invalid data (121.47ºE, 31.27ºN)
3. RESULTS To a nalysis the LST dynamic cha racters of differe nt land cove r t ypes, 9 cl asses of t he ori ginal l and co ver m ap w ere merged to 4 basic classes: water, urban, cropla nd and forest. Then the HANTS reconstructed land s urface temperatures for the four classes were ev aluated. General statistics results of HANTS harmonic characteristics for the four land cover types are provided in table 1. From this table we find that urban has the highest man LST ( 297.85K) and water has the lowest (291.03K), which means that the annual mean LST of urban is abo ut 7K h igher than water. Urban also has the highest amplitude of 1st harmonic, expressing remarkable seasonality, but the lowest amplitude is not belonging to water, but to forest. The highest amplitude of urban can be attributed to its lower evaporation and heat capacity. Low amplitude of forest may be t he result of canopy shielding effect. As for the 1st harmonic phase, urban, cropland and forest show similar values near 188 day, indicating their maximum LST in early July. Water shows t he lowest mean LST wit h its peak occurrence later about 10 days than other land cover types. The lag effect of water surface temperature results from its high heat capacity. Table 1. General statistics of LST characteristics for the four land cover types
Land cover type
Mean LST (K)
1st harmonic amplitude (K)
1st harmonic phase (Day of Year)
Water 29
1.03
11.37
196.93
Urban 29
7.85
13.50
188.19
Cropland 2
96.18
12.02
187.44
Proc. of SPIE Vol. 7498 749809-3
Forest 29 5.21 11.03 189.05 Figure 4 gives the mean LST time series of the four land cover types. The LST of water is obviously lower than other three land covers throughout the year. Urban shows higher temperature in summer but is similar to cropland and forest in winter. There is little difference between the temperature of cropland and forest in most of the year, except for the period from May to July, the crop harvest season of study area. The dec rease of vegetatio n cover leads to the rise of cropland temperature, contributing to the difference between cropland and forest.
Fig. 4. LST time series of the four land cover types
The amplitude and phase of the HANTS first harmonics were mapped to study the spatial variations of LST across the Yangtze R iver Del ta (Fi gure 5a an d 5b). The re gions with the greates t a mplitude are high urbaniz ed areas , s uch as Shanghai and Suxichang area. Northwest, southwest and southeast areas have weakest seasonal variability. From the 1st harmonic phase i mage, t he s patial pat tern of LS T peak c an be ap preciated. In a general m anner, p hase over t he delta ranges from 180 to 201, which means the LST peak time occurs in early to middle July. Regions far from the sea without large water bodies have early p eak time, particularly evident in Nanjing and surrounding areas. Large water bodies and costal areas see their LST peak later than other regions, including the Taihu Lake, Qiandaohu Lake, Yangtze River and Zhoushan Archipelago.
Fig. 5. The spatial distributions of 1st harmonic (a) Amplitude (b) Phase
Proc. of SPIE Vol. 7498 749809-4
Seasonal average images of LST also have been calculated. In spring, summer and autu mn, urban areas ex hibit higher temperature, indicating significant Urban Heat Island effect in th e Yan gtze Riv er Delta. In winter, th ere is sm all difference between the average LST of urban and other land cover types. As for the Z-shaped city strap composed by Nanjing, Sux ichang, Sh anghai, Hanghzou, Ning bo and o ther cities, th e urban h eat islan d effect are more d istinct. In addition, it sh ould be noted that some small regions show high LST throu gh out the year (a bove 300K even in winter). Consulted with maps, we find that these regions are downtown areas in major cities. Their higher temperature and lower seasonal variability may correspond to the dense population and thriving commercial activities.
Fig. 6. The day time se asonal a veraged la nd surfa ce te mperature of the Ya ngtze Ri ver De lta (a ) Spri ng (b) Summe r (c ) Autumn (d) Winter
Proc. of SPIE Vol. 7498 749809-5
4. CONCLUSIONS The con clusions ar e as follows: 1) Th ere ar e nu merous d ata g aps i n M ODIS 8-day co mposite LST product and interpolation is n eeded before ap plications. 2) HANTS i s a powerful to ol t o reconstruct LST cloud- free tim e-series, although primarily designed for vegetation applications such as NDVI and EVI. 3) Urban area has the highest mean LST of the four main land cover types and remarkable seasonal fluctuation. Water shows the lowest mean LST with its peak occurrence later about 10 days than other land cover types. 4) In spring, summer and autumn, there is significant Urban Heat Island effect in the Ya ngtze Rive r Delta, esp ecially in th e Z-sh aped city strap . While in winter, th ere is small difference between the average LST of urban and other land cover types. 5) The HANTS algorithm has proven to be an effective method not only for LST time-series reproduction, but also for analyze LST temp-spatial variations.
ACKNOWLEDGEMENTS This r esearch was supported b y the National Natural Science Foundation of China (G rant N umber: 3 0571078, 40801040), t he Scien tific R esearch Fo undation of Nanjing Un iversity o f i nformation Scien ce and Technology (Grant Number: Y 653), an d t he Open Fund o f Key Lab oratory of M eteorlogical Di saster of M inistry of Educat ion (Grant Number: KLME0703; KLME0805). Th e au thors would also lik e to th ank USGS/LP DAAC for th eir pro vision of MODIS data.
REFERENCES 1. S. W. R unning, C . J ustice, V. Salomonson, et al , "Ter restrial rem ote sensi ng s cience an d al gorithms pl anned f or EOS/MODIS", International Journal of Remote Sensing, 15(17), 3587-3620 (1994). 2. L. Man zo-Delgado, R. Agu irre-Gomez, R. Al varez, "M ultitemporal an alysis of land surface temp erature u sing NOAA–AVHRR: Preli minary relatio nships b etween cl imatic anom alies an d f orest fires", International Journal of Remote Sensing, 25(20), 4417-4423 (2004). 3. Y. Jia, Z. Li, " Progress in Land Surface Te mperature Retrieval from Passive Microwa ve Remotely Sensed Data", Progress in Geography, 25(3), 96-105 (2006). 4. W. Wan, D. Lv, "Diurnal and seasonal variation of clear-sky land surface temperature of several representative land surface types in china retrieved by GMS 5", Acta Meteorologica Sinica, 63(6), 957-968 (2005). 5. Z. Qi n, A. Karnieli, "Progre ss in the remote sensi ng of land s urface temperature and ground em issivity using NOAA-AVHRR", International Journal of Remote Sensing, 20(12), 2367-2393 (1999). 6. J. A. Voogt, T. R. Oke, "Thermal remote sensing of urban climates", Remote Sensing of Environment, 86(3), 370384 (2003). 7. M. Stathop oulou, C . Car talis, "D aytime u rban heat is lands from Lan dsat ETM + a nd Corine l and c over dat a: An application to major cities in Greece", Solar Energy, 81(3), 358-368 (2007). 8. Y. Julien, J. Sobrino, W. Ve rhoef, "Changes in land s urface temperatures and NDVI values over Europe between 1982 and 1999. Remote Sensing of Environment", Acta Geographica Sinica, 103, 43 – 55 (2006). 9. W. Verhoef, "A pplication of H armonic Analysis o f NDVI Time Ser ies ( HANTS) ", Fourier analysis of temporal NDVI in the Southern African and American Continents, Wageningen, The Netherlands, 1996. 10. G. J. R oerink, M. Menenti and W. Verhoef. "Reconstructing cloudfree NDVI composites using Fourier analysis of time series". International Journal of Remote Sensing, 21(9), 1911-1917 (2000).
Proc. of SPIE Vol. 7498 749809-6