Abstract:
The limitation of thermal satellite images at finer spatial resolution (FR) led to increased demand for developing various downscaling techniques for the generation of FR image of land surface temperature (LST) to enhance the information content. Thus, the major concern of the analysis is the thermal sharpening of MODIS-LST in various urban regions by establishing the correlation of LST with various spectral indices (SI). Various regression techniques using combination of SI were applied for the thermal sharpening of LST from MODIS data over four different Indian cities with different climate zones i.e. Bikaner, Vadodara, Hyderabad and Varanasi. The LST image from MODIS sensors at spatial resolution of 930 m was disaggregated to 100 m and accuracy was determined by comparing with the LST image of Landsat-8-TIRS at 100 m. The best combination of indices found to include both vegetation and built-up/soil indices for thermal sharpening of MODIS-LST. Further, the variation of the best combination for different cities indicates its dependence on the present land cover. The correlation coefficients (R) between the downscaled MODIS-LST image and the reference-Landsat image were found to be 0.84, 0.76, 0.83 and 0.92, whereas the RMSE values were found to be 1.27, 1.02, 0.73 and 0.62 for Bikaner, Vadodara, Hyderabad and Varanasi, respectively. The RMSE values were found remarkably below the standard deviation of each reference LST image. Therefore, the downscaling approach adopted in this study showed high potential for accurate LST mapping at FR in various urban areas. © 2020, Springer-Verlag GmbH Austria, part of Springer Nature.
