Producing High-resolution Land Surface Temperature Maps Using Sentinel-2 Satellite Data in Isfahan

Monitoring land surface temperature (LST) using remote sensing data and examining its temporal and spatial variation is important in various studies such as environmental research, land use change, water resources management, and drought monitoring, especially at local scales. Remote sensing data can provide a long-term and continuous spatial coverage of LST. However, LST data retrieved from thermal infrared (TIR) band imagery have a coarser spatial resolution than surface reflectance (SR) data collected from shortwave bands on the same instrument. LST products with high spatial and temporal resolutions are not yet available. Therefore, several downscaling algorithms to produce high-resolution LST maps from sensors with appropriate temporal resolution have been developed recently. In this study, thermal sharpening of land surface temperature obtained from Sentinel-3 satellite images with a spatial resolution of 1 km and temporal resolution of less than 1 day was carried out using the PyDMS model and Sentinel-2 images to produce LST maps with a spatial resolution of 20 meters for Isfahan, Iran. PyDMS is a machine learning algorithm based on decision tree regression that relates the reflectance of high-resolution bands to the LST of the corresponding low-resolution image. The results of this model have been compared against the LST measurements at a depth of 5 cm in three meteorological stations including Isfahan Airport, Isfahan, and Kaboutarbad. Moreover, LST products of MODIS and Landsat-8 have been used to assess the consistency of the sharpened LSTs. The results show that the correlation coefficient is higher than 0.74 in all 3 stations and the RMSE error is equal to 6.7, 4.0 and 15.5 °C in Isfahan, Kabutrabad and the airport, respectively. Moreover, the spatial pattern of the sharpened LST is compatible with the spatial pattern of the LST products of Landsat 8, Sentinel 3 and MODIS. The findings of this study indicate the promising application of the PyDMS algorithm for producing LST maps with a spatial resolution of 20 meters and temporal resolution of fewer than 7 days, though bias correction using in situ LST can improve results. Production of LST maps with both high spatial and temporal resolutions is extremely useful for many practical applications such as soil and crop management practices, evapotranspiration estimation, and irrigation water management.