A Deep Neural Network-Based Approach for Studying the Relationship Between Land Surface Temperature and Surface Air Temperature

Air temperature is one of the most important parameters for assessing and monitoring the changing weather and climate patterns. Measurement of air temperature is done from a limited number of automatic weather stations distributed in certain parts of the country, with a lot of discontinuities in space and time because of various operational constraints. Contrary to it, satellites provide seamless observations of Land Surface Temperature in space and time globally, only obscured by the cloud cover. Though Land Surface Temperature and Air Temperature have different physical interpretations, nevertheless some previous studies have shown some correlation between them, which varies according to elevation, land cover type, time of observation (day or night). Hence, land surface temperature can be one of the means to derive air temperature in the regions where the availability of automatic weather stations is limited. The present study attempts to develop a deep neural network-based model to estimate air temperature from land surface temperature based on elevation and land cover. The model results were evaluated by comparing them with ground observations using statistical indices viz., Nash and Sutcliffe’s coefficient of efficiency, Legates and McCabe coefficient of efficiency, Coefficient of determination, and Index of agreement. The values obtained for these indices are 0.74, 0.85, 0.85, and 0.97 respectively, which reflect the predictive capability of the model.