ENHANCED MODELING OF REMOTELY SENSED ANNUAL LAND SURFACE TEMPERATURE CYCLE

Abstract

Satellite thermal remote sensing provides access to acquire large-scale Land surface temperature (LST) data, but also generates missing and abnormal values resulting from non-clear-sky conditions. Given this limitation, Annual Temperature Cycle (ATC) model was employed to reconstruct the continuous daily LST data over a year. The original model ATCO used harmonic functions, but the dramatic changes of the real LST caused by the weather changes remained unclear due to the smooth sine curve. Using Aqua/MODIS LST products, NDVI and meteorological data, we proposed enhanced model ATCE based on ATCO to describe the fluctuation and compared their performances for the Yangtze River Delta region of China. The results demonstrated that, the overall root mean square errors (RMSEs) of the ATCE was lower than ATCO, and the improved accuracy of daytime was better than that of night, with the errors decreased by 0.64 K and 0.36 K, respectively. The improvements of accuracies varied with different land cover types: the forest, grassland and built-up areas improved larger than water. And the spatial heterogeneity was observed for performance of ATC model: the RMSEs of built-up area, forest and grassland were around 3.0 K in the daytime, while the water attained 2.27 K; at night, the accuracies of all types significantly increased to similar RMSEs level about 2 K. By comparing the differences between LSTs simulated by two models in different seasons, it was found that the differences were smaller in the spring and autumn, while larger in the summer and winter.