Validation of NOAA Land Surface Temperature Retrievals Across Successive Satellite Generations Using SURFRAD Datasets

Abstract

Land Surface Temperature (LST) is an essential parameter for climate monitoring, hydrological studies, and research on land–atmosphere interactions. In this study, LST retrievals from NOAA satellites over several decades, from NOAA-7 to NOAA-21, are evaluated using ground observations from the NOAA Surface Radiation Budget Network (SURFRAD). The validation dataset comprises observations from seven SURFRAD stations distributed across climatologically diverse regions of the United States: Bondville (Illinois), Fort Peck (Montana), Goodwin Creek (Mississippi), Table Mountain (Colorado), Penn State (Pennsylvania), Desert Rock (Nevada), and Sioux Falls (South Dakota). These stations represent contrasting environmental conditions, including agricultural and grassland surfaces, humid and temperate environments, semi-arid regions, and sparsely vegetated desert surfaces. Reference surface temperatures were derived from measurements of upwelling and downwelling longwave radiation using the Stefan–Boltzmann equation. Satellite and ground-based measurements were matched through rigorous spatial and temporal collocation procedures, together with cloud screening and quality-control filtering. In this study, the Split-Window (SW) algorithm was used to estimate LST from thermal infrared data.

The validation accounts for both spatial and temporal variability in surface conditions, including differences in land-cover characteristics, vegetation density, soil exposure, surface moisture, and seasonal variations in vegetation and atmospheric conditions. The performance of the LST products was evaluated using the Mean Bias Error (MBE), Root Mean Square Error (RMSE), and correlation coefficient. The results reveal a noticeable enhancement in retrieval accuracy over time, with the most recent VIIRS-based products outperforming those derived from the earlier AVHRR sensors. In particular, LST estimates from JPSS-1/NOAA-20 and JPSS-2/NOAA-21 typically exhibit biases within ±1.5 K and RMSE values lower than 2 K. The study also shows that validation statistics vary according to surface characteristics and season, emphasizing the role of land-cover heterogeneity and temporal variability in LST retrieval performance. Overall, the results confirm the usefulness of SURFRAD measurements as an independent reference for satellite product assessment and contribute to the establishment of a reliable, long-term NOAA land surface temperature climate data record.