Downscaling Methodology for Satellite Land Surface Temperatures over Urban Environments

Abstract:

A linear regression method was developed and tested to derive high spatial and high temporal resolution Land Surface Temperature (LST) products using a combination of Landsat 8 Thermal Infrared Sensor and the Geostationary Operational Environmental Satellite-R Series (GOES-R). Landsat 8 provides high spatial resolution (30 m) estimates of skin temperature every 16 days, and GOES-R, which has lower spatial resolution (2 km), evaluates the skin temperature at a much higher temporal resolution (5 minutes). There is a systematic bias between Landsat 8 and GOES-R in temperature measurements even at the same time and spatial scales due to sensor configurations, footprint size, radiometric and spectral differences and retrieval algorithm. The linear regression model used to downscale GOES-R to Landsat 8 spatial resolution every 5 minutes accounted for the systematic differences between the two satellites by removing the average of Landsat 8 temperature from the GOES-R LST observations. Additionally, the temporal differences between the land cover types at finer resolution were also accounted for. The disaggregated results were comparable to the observed Landsat 8 LST for the same measurement time. The downscaled estimates showed reasonable agreement (0.63 K error and RMSE of 1.61K) when they were validated against independent Landsat 8 images. Preliminary results also revealed that the proposed model can reasonably estimate LST in urban regions. High resolution LST for cities may improve prediction of heat indices and the effects of urban heat islands. Improving indices and understanding heat islands are crucial for sustainable and resilient urban environments.