Radiance-based validation of land surface temperature products derived from Collection 6 MODIS thermal infrared data

摘要

Land surface temperature (LST) is an important climate variable related to surface energy and water balance. The LST products retrieved from Terra-/Aqua-MODIS thermal infrared data have been widely used in climatological, hydrological, meteorological, and ecological applications. The newest collection (C6) MODIS LST products have been freely available for the user community since 2016. It is necessary to evaluate the accuracy of the C6 MODIS LST products prior to their application. In this study, the C6 MODIS LST products were validated using a radiance-based (R-based) method. We selected 38 validation sites worldwide over eight land cover types, i.e., in-land water, forest, shrubland, grassland, cropland, urban, snow, and bare soil surfaces. Except for the results for grassland sites during the daytime and night-time and the bare soil sites during the daytime, the C6 MODIS LST products are in good agreement with the R-based LST values, with bias and root mean square error (RMSE) values of less than 1 K for all sites during the daytime and night-time. Large LST discrepancies over grassland sites are caused by misclassification in the MODIS land cover type products, which can lead to large uncertainties in the determination of surface emissivity. We also compared the accuracies of the C5 and C6 MODIS LST products. Results indicate that there are no significant accuracy discrepancies between the C5 and C6 MODIS LST products for all sites, with the exception of bare soil sites. Compared with the C5 MODIS LST products, significant accuracy improvement in the C6 MODIS LST products over most bare soil sites can be found. The absolute bias values are reduced from approximately 1.4-3.0 K for C5 to approximately 0.3-0.8 K for C6, and the RMSE values from approximately 1.4-3.1 K for C5 to approximately 0.5-0.8 K for C6. However, further overestimation in the C6 MODIS LST products can be found over two bare soil sites due to the poor emissivity adjustment provided by the emissivity adjustment model incorporated into the C6 split window algorithm.