Downscaling of the Land Surface Temperature Data Obtained at four Different Dates in a Year Using the GWR Model: A Case Study in Antakya, Turkey

Land surface temperature (LST) is a major factor that affects many biophysical processes in the land-atmosphere relationship. This factor is obtained from satellite images having different temporal and spatial resolutions. This study applied the geographically weighted regression (GWR) model for four different dates representing each season a year to improve the LST images obtained in coarse resolution. In this study, MODIS LST images that are available having fine temporal but coarse spatial resolution were modeled using NDBI and NDVI indices, and their spatial resolution is improved. In addition, LANDSAT 8 images were used as reference images to evaluate the accuracy of the images obtained from the models. Results of the GWR model have been evaluated by comparing it statistically with TsHARP and DisTradother commonly used methods. As a result of the comparison by using the average of four dates outputs, the GWR model (R-2 = 0.73, RMSE = 0.78) was more successful than the TsHARP (R-2 = 0.56, RMSE = 1.00) and DisTrad (R-2 = 0.49, RMSE = 1.09) methods. The most successful downscaling performance in the GWR model was obtained in the spring season (RSR = 0.48). According to these findings, the GWR model can be used for downscaling LST images in urban areas. However, before applying this algorithm to scenarios outside of urban areas, it is recommended to use the required parameters and optimize their combinations.