Validation of land surface temperature products from MODIS, ECOSTRESS, Landsat, GOES-R, VIIRS and Sentinel-3 benchmarked on in situ measurements in the U.S. Corn Belt

Abstract:

Land surface temperature (LST) is an essential variable to closely connect ecosystem energy balance and water dynamics, and it is a strong indicator of plant water stress, and has been used as critical inputs in models for calculating crop water use and stress. In the past few years, several new LST products have become available to the public, including ECOSTRESS, Sentinel-3, GOES-16and Landsat Provisional LST products, and these LST datasets provide new options in terms of spatial and temporal resolution for users. Given the availability of all these operational satellite LST products and an increasing demand of reliable LST for agricultural applications in particular, a systematic and thorough study of validation of different satellite LST data for croplands is missing. This study provides a comprehensive study on validating both the new satellite LST data including ECOSTRESS, Sentinel-3, GOES-16 and Landsat Provisional LST, and several mainstream LST products including MODIS Land Surface Temperature/Emissivity product (MOD11 and MYD11), MODIS/Aqua Land Surface Temperature/3-Band Emissivity product (MYD21), GOES-R and VIIRS LST, for agricultural landscapes in the U.S. Corn Belt, where ~⅓ of global corn and soybean are produced. Results show that the nighttime and daytime biases of all LST products on different sites were generally within ± 2 K and ± 3 K, respectively. Regarding the daytime LST, the highest agreement with ground observations was achieved by ECOSTRESS with an overall bias of -0.9 K and RMSE of 2.2 K. MOD11 and MYD11 products slightly underestimated daytime LST with an overall absolute bias (< 0.8 K) and RMSE (< 2.8 K). MYD21, Landsat Provisional LST, GOES-R, VIIRS and Sentinel-3 LST achieved an overall absolute bias (< 2.1 K) and RMSE (< 4.5 K) in terms of daytime LST. Regarding nighttime LST, all LST products reached a low overall absolute bias (< 0.5 K) and RMSE (< 1.7 K) except ECOSTRESS and MOD11 (absolute bias < 1.9 K and RMSE < 3.4 K). The fact that there is a huge accuracy difference between daytime and nighttime LST products, indicates that the algorithms of satellite LST products need further improvement under spatial thermal heterogeneous conditions.