关键词: 地表温度, 多尺度地理加权回归, 时空演化, MODIS, 西藏

Abstract:

Land surface temperature （LST） is one of the important indicators reflecting the ecological environment and has a strong correlation with air temperature. Against the backdrop of rising global air temperature， Tibet， a sensitive area for climate change， is worthy of being monitored the spatiotemporal change for LST by remote sensing technology on a large scale， which is helpful for gaining insight into the evolutionary process of the Tibetan thermal environment and for long-term monitoring of basic ecological change in plateau areas. However， remote sensing images in plateau areas generally face the problem of cloud shading， with the annual average cloud amount exceeding 50%， which greatly reduces the availability of remote sensing data and makes it difficult to effectively monitor the LST in plateau areas over a large range and long-time span. In addition， the research on LST and surface parameters is mainly based on classical linear regression and geographically weighted regression， and the scale effects of different surface parameters on the spatial differentiation pattern of LST are not fully considered， which results in an unstable regression.In this paper， with the help of Google Earth Engine （GEE）， the MODIS LST data of Tibet without cloud cover from 2000 to 2020 were obtained by the methods of cloud identification， could masking， image overlay and mean value composite to address the problem of cloud cover in the plateau region. LST was classified into 5 classes by normalized classification method， and then the spatiotemporal distribution characteristics of Tibet’s LST is explored using trend analysis， landscape pattern indexes and movement analysis of the center of gravity. Moreover， the surface parameters which effect LST such as normalized difference vegetation index （NDVI）， bare soil index （BI）， perpendicular impervious surface index （PISI）， humidity （WET） are retrieved and digital elevation model （DEM） is attained. In view of the fact that traditional regression methods cannot take into account the spatial scale variations， the paper use multi-scale geographically weighted regression which can address this problem and the effect on LST and the spatial scale variations of LST influencing factors are analyzed. The following conclusions are drawn from the research analysis： The mean value of LST increased from 18.72 ℃ to 20.28 ℃ over the past two decades （2000—2020） with an annual growth of 0.09 ℃， showing a weak increased trend. The interannual variation law of LST in Tibet is as follows： fluctuating increase—continuous decrease—rapid increase—large decrease—slight increase. From the perspective of ecological regions， the interannual changes of LST in each ecological region are significantly different. LST and its increased trend are spatially uneven in Tibet， with a higher value in northwest than in southeast. The spatial structure of the LST is dominated by the medium LST zone and above， which accounts for about 70% of the study area. Low LST zone and high LST zone is aggregated with simple and regular shapes while sub-low LST， medium LST， and sub-high LST zones are fragmented with complex shapes. The distribution of the center of gravity for each LST zones during 2000—2020 has obvious directionality and the movements of that have significant differences. In particular， the movements of the center of gravity in the low LST zone and high LST zone show a shift from toward each other to moving in opposite directions， reflecting that the regional LST gap between the east and west of study area has experienced a process from narrowing to widening. DEM and WET have negative effects on LST， BI， PISI and NDVI have positive effects， and the intercept have different influence in different ecological zones. Among of them， DEM has fairly small scale and the strongest effect； the intercept has the smallest scale and fairly strong effect.

Key words: land surface temperature （LST）, multi-scale geographically weighted regression, spatiotemporal evolution, MODIS, Tibet