Evaluation of six satellite- and model-based surface soil temperature datasets using global ground-based observations

Abstract The comprehensive evaluation of model- and satellite-based surface soil temperature (ST) products is a prerequisite for applications of these datasets in hydrology, ecology, and climate change, as well as in passive microwave soil moisture retrieval algorithms. Distinguished from existing regional validations, this study used ground soil temperature observations of approximately 800 stations from 5 sparse and 15 dense networks worldwide to fully assess six model- and satellite-based surface ST products from April 2015 to December 2017. The products consist of five model-based ST from the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), the Goddard Earth Observing System Model version 5 Forward Processing (GEOS-5 FP), the Global Land Data Assimilation System (GLDAS) Noah, the ERA-Interim and the newly released ERA5 produced by the European Centre for Medium Range Weather Forecasts (ECMWF), and one satellite-based ST retrieved from the Advanced Microwave Scanning Radiometer 2 (AMSR2) by using the Land Parameter Retrieval Model (LPRM). The accuracy of these products was comprehensively assessed per availability of ground networks, soil temperature interval, soil moisture interval, land cover, climate zone, and elevation. The results show that the GEOS-5 exhibits the smallest averaged unbiased root mean square difference (ubRMSD) of 1.84 K among all the surface ST products. All model-based products show a high skill in capturing the temporal trend of ground observations with an averaged correlation coefficient larger than 0.97. The ERA5 surface ST obtains visible improvements compared to its predecessor ERA-Interim by showing smaller ubRMSD and absolute bias values. All model-based surface ST products generally show lower values than ground ST, while their bias tends to be warmer as soil temperature and soil moisture increase except for the highest temperature and moisture conditions. Moreover, unstable performance of most model-based products in shrubland and grassland, tropical, arid and cold climate, and high elevation regions is also demonstrated by larger ubRMSD values. The averaged ubRMSD of the satellite-based LPRM ST is 3.04 K, and more attentions should be paid to the impacts of elevations and underlying surfaces on improving this product. These new findings will be valuable for future refinement of ST datasets, algorithms used to estimate soil moisture from satellite data, and applications in various disciplines.