The ERA5 reanalysis dataset plays a pivotal role in global and regional climate change research and has witnessed increasingly extensive adoption by the scientific community in recent years. However, assessments regarding its regional representativeness remain preliminary. Employing the in-house developed high-resolution temperature dataset, C-LSAT HR (1900–2024), as a benchmark, we evaluated the performance of ERA5 over China. Our analysis yields the following noteworthy insights:

1. ERA5 demonstrates superior performance compared to a multitude of other reanalysis datasets;

2. Specifically during the 1965–1966 period, ERA5 exhibits significant systematic and random biases over China, attributed to the deficiency of data assimilation in the region.These findings provide a critical reference for future applications of ERA5 data.

   
   

The abstract of the article:
Reanalyses are essential for climate change research and operational forecasting. Among the widely used reanalyses, ERA5 stands out as one of the most popular, with its latest version, extending back to 1940, released in 2023. However, a systematic comparative evaluation of ERA5's full-period dataset in China is lacking. This study compares the surface air temperature (SAT) data from ERA5 with those from four other extended reanalysis products (20CRv3, 20CRv3-ERA5, CERA-20C and ERA5-Land), using the recently updated C-LSAT HR as the benchmark dataset. The comparison focuses on climatology, long-term trends, and SAT variations over China. Our findings reveal that all three centennial-scale reanalyses significantly underestimate the warming trend in China since the early 20th century. Overall, ERA5 performs the best across all above evaluation aspects. However, the system and random errors in the years 1965 and 1966 are considerably larger than usual, primarily due to the lack of observations as input data. Further analysis using station observations as the “reference truth” shows that these anomalous errors persist throughout both years, with an overall cold bias of -0.74°C (1965) and -0.67°C (1966). The magnitude of the errors varies by season and region, with the most pronounced increase observed during the summer months. The MAE (MBE) in the summers of 1965 and 1966 increase by 0.44°C (-0.59°C) and 0.50°C (-0.65°C), respectively. Spatially, ERA5 records the largest system biases in the summer for southern Xinjiang and in the autumn for the Tibetan Plateau, with the MAE increase exceeding 3°C in these regions. The most significant contribution ratio of RMSE for abnormal years (IRMSE) occurs in the summer in South China and southern Xinjiang, where it exceeds 15% at most stations, although the average MAE increase for South China is only 1~2°C.

Lin J., Zhang H., Wei S., Xu Q., Li Z., Li Q.*, 2025, Outstanding performance of ERA5 reanalysis temperature in China since 1950s and quantification of its abnormal error in 1965-1966, Quart. J. Roy. Meteor. Soc., e5042:1-21，DOI:10.1002/qj.5042