Calibration of Separate Window Model Factors to Calculate Land Surface Temperature using MODIS Images

Abstract

Land surface temperature (LST) is one of the most important parameters influencing physical processes of energy on the land surface and in high seas, both in local and global scales. Satellite infrared temperature data (TIR) is linked directly to LST using radiation transmission models. However, direct estimation of LST from radiation in TIR spectrum will be of low accuracy. Since the radiation measured by satellites depends not only on land surface parameters (temperature and irradiance power) but also on atmospheric influences. LST calculation suggests different methods for decreasing atmospheric influences, which can be classified in three major classes: single band methods, multiple band methods, and multiple angle methods. The present article investigates multi-temporal data of MODIS images in 12 different dates with quite uniform temporal distribution during 2014 using five useful multiple band methods of calculating LST including, Price Model (1994), Becker and Li Model (1990), Platt and Prata Model (1991), Ulivieri et al. model (1994), Coll et al. model (1994). Then, coefficients of investigated models were calibrated using the least repetitive squares model. During the calibration, main coefficients of the models were used as the initial value and optimal coefficients were calculated using a series of data. Afterward, the accuracy of the modified models was evaluated using LST from MODIS and the Iranian weather stations data. Results illustrate the modified Price Model by an average of RMSE 0.41 Centigrade degree as the most accurate model. Moreover, the variance of RMSE is 0.08 for mentioned dates which confirm generalizability of the outcomes. The maximum and minimum of RMSE equals 0.26 and 0.50 respectively (February 19th and June 27th respectively) for modified Price model. Finally, the linear relation was investigated, between LST calculated using modified Price Model and data measured by Iranian weather stations.

The linear regression factor of these two series of data was 0.9978 which indicates a significant linear relation between calculated LST data and reference temperatures of the Iranian weather stations.