Accuracy Improvement of Land Surface Temperature Extracted from Thermal Bands of Landsat Satellite using Linear Regression and Ground Observations

Monitoring the changes in land surface temperature (LST) caused by seasonal and non-seasonal fluctuations is essential due to its profound impact on the human and natural environment. The use of remote sensing satellite imagery has been extensively considered for the continuous monitoring of LST with its low cost and high speed. In this research, first, three conventional temperature extraction approaches based on Plank’s law were used to extract LST of surrounded areas of Lake Urmia. Then, using a linear regression and observations of the temperatutre in meteorological stations, LST maps extracted from conventional methods were modified. The results showed that conventional temperature extraction approaches for TM and TIRS sensors of Landsat Satellite had an accuracy of about 4 °C and 8 °C, respectively. After modifying LST maps using the linear regression, this accuracy reduced to 1 °C and 0.5 °C for TM and TIRS sensors, respectively. This indicated the proper performance of the regression approach presented in this study for temperature modification.