Evaluation of the performance of WRF9km in simulating climate over the upper Yellow River Basin

Abstract

Understanding the current climate in the Yellow River Basin is essential for accurately predicting future climate change and assessing its impacts on water resources and ecosystems; however, existing models exhibit notable biases in this region, primarily due to low resolution and errors in driving data and model domains. Using in-situ station observation data, CN05.1 gridded meteorological observation dataset, along with the ERA5 and MERRA2 reanalysis datasets, the performance of the WRF9km in simulating temperature and precipitation from 1980 to 2016 was comprehensively evaluated. Results indicate that the WRF9km model effectively captures the spatial pattern of air temperature, with a spatial correlation exceeding 0.86 (at the 95% confidence level) and a cold bias of −2.8 °C compared to CN05.1. This bias is primarily due to the underestimation of downward radiation and the overestimation of surface albedo. However, the WRF9km model fails to reproduce the observed warming trend across the entire region, especially during the summer. For precipitation, the WRF9km model generally reproduces the observed spatial pattern, with spatial correlation coefficients above 0.80 for all seasons except winter (at the 95% confidence level). However, the model overestimates precipitation relative to CN05.1 and underestimates it when compared to MERRA2. The precipitation bias is mainly attributed to the misrepresentation of wind fields and moisture by the WRF9km model. Regarding precipitation trends, different datasets yield divergent results, indicating substantial inter-annual variability that is difficult for the WRF9km to capture. Compared to the driving ERA5 data, the WRF9km model reduces cold biases between November and December, as well as wet biases across all seasons. The model also better simulates the winter warming trend in the western part of the UYRB and the summer wetting trend in the northern part. The evaluation of the WRF9km model provides valuable insights for the development of dynamical downscaling in terrain complex regions, especially for improving the surface albedo scheme and input driving data.