ABSTRACT The land surface‐air temperature difference ( T s − T a ) is a key parameter in land‐atmosphere energy exchange. As the core driver of sensible heat flux, it regulates turbulent motion and vertical thermal structure, directly impacting near‐surface stability, boundary layer development, and extreme climate events. This study evaluates the performance of reanalysis products in simulating T s − T a over the Contiguous United States (CONUS), using observations from more than 100 U.S. Climate Reference Network (USCRN) stations during 2002–2024. Results show that ERA5 and ERA5‐Land reproduce the spatial pattern of T s − T a reasonably well, but exhibit systematic biases: both reanalyses datasets significantly underestimate the daily maximum T s − T a (occurring during daytime), with mean biases of −3.1°C and −3.4°C averaged over CONUS, and the largest underestimation found in the arid southwestern United States during summer. Terrain analysis further reveals that these biases are topography‐dependent: models consistently underestimate nocturnal T s − T a in valleys, where cold‐air pooling occurs, while ERA5‐Land overestimates nighttime T s − T a over flat and elevated terrain in summer and autumn. Benefiting from its higher‐resolution land surface model, ERA5‐Land shows some improvement over ERA5, reducing the daily mean bias from −0.9°C to −0.6°C and the daily minimum bias from −0.8°C to +0.4°C. However, both reanalysis products fail to capture the observed significant decreasing trend in minimum T s − T a (−0.52°C ± 0.005°C/decade in observations versus +0.02°C/decade in ERA5 and +0.01°C/decade in ERA5‐Land). This discrepancy is attributed to the assumption of synchronous trends in surface temperature and air temperature within the reanalysis systems, revealing limitations of current reanalysis products in representing the long‐term dynamics of land‐atmosphere interaction. This study provides a scientific basis for optimising climate models and informing reanalysis data applications.
Yu et al. (Mon,) studied this question.