Angular normalization of GOES-16 and GOES-17 land surface temperature over overlapping region using an extended time-evolving kernel-driven model
Auteurs
Qin B., Chen S., Cao B., Yu Y., Yu P., Na Q., Hou E., Li D., Jia K., Yang Y., Hu T., Bian Z., Li H., Xiao Q., Liu Q.
Référence
Remote Sensing of Environment, vol. 318, art. no. 114532, 2025
Description
Land surface temperature (LST) is an important parameter that critically contributes to Earth’ s climate. Thermal anisotropy is a major challenge that must be addressed while generating long-term LST products from satellites. For instance, the differences between GOES-16 and GOES-17 LST products caused by thermal anisotropy have not yet been resolved, which impacts the high-frequency monitoring of the land surface. The coupled contributions of the gap fraction and hotspot effects in the thermal infrared domain result in the existence of thermal anisotropy effect. The time-evolving kernel-driven model (TEKDM) is a recently proposed practical tool for conducting LST angular normalization for geostationary satellites. However, the existing six-parameter TEKDM considers only the hotspot effect and ignores the gap fraction effect, which may limit the TEKDM-based angular normalization method. In this study, we proposed an extended seven-parameter TEKDM considering both the gap fraction and hotspot effects and normalized the angular effect of GOES-16 and GOES-17 LST products over the overlapping region using this model. The accuracy of this seven-parameter TEKDM was evaluated using a physically based discrete anisotropic radiative transfer (DART) simulation dataset. Subsequently, the seven-parameter TEKDM-based angular normalization method was evaluated using the GOES-16 and GOES-17 LST products of the overlapping region for one year against ten AmeriFlux sites. The results showed that the seven-parameter TEKDM had a RMSE (MBE) of 0.36 K (0.0019 K). Compared with the RMSE of the NOAA-released GOES LST products, the seven-parameter TEKDM-based normalization method could reduce the RMSE of GOES-16 and GOES-17 LST products from 2.2 K and 2.6 K to 1.7 K, respectively, with a reduction of 0.5 K (22.7 %) and 0.9 K (34.6 %), respectively. Furthermore, the RMSE/MBE of GOES-17 LST exhibited a different diurnal variation pattern than that of GOES-16 LST, which could be explained by the different illumination-viewing geometries of the two satellites. This emphasizes the necessity of conducting angular normalization of current geostationary satellite LST products. The seven-parameter TEKDM provides a feasible method for generating long-term high-quality LST datasets for remote sensing communities.
