Dominant Role of Clouds in Regulating Surface Incident Solar Radiation from Twentieth-Century Reanalyses over Mainland China

Abstract Long-term historical surface incident solar radiation ( R s ) is essential for investigating climate change and evaluating climate model performance. Yet, the reliability of the available century-long reanalysis R s products over mainland China remains uncertain. In this study, we systematically compare the performance of five twentieth-century reanalyses in simulating R s over mainland China, using sunshine-duration-derived R s from approximately 2400 meteorological stations (1960–2010). The results indicate that only Coupled European Reanalysis of the Twentieth Century (CERA-20C) and 20CRv3 provide better simulations of the long-term average R s over mainland China, while none of the reanalyses adequately reproduce the observed long-term trends. We restrict the attribution analysis to the recent decade of 2000–10 for CERA-20C and 20CRv3, during which reliable clear-sky solar radiation ( R c ) data, which mainly reflect the aerosols’ impact on R s , are available. Using an extreme gradient boosting (XGBoost) regression model interpreted with the Shapley additive explanation (SHAP) framework, we quantify the relative contributions of biases in R c and total cloud cover (TCC), together with auxiliary spatial and temporal predictors, to both the monthly biases and the trend biases of R s . The attribution results consistently show that TCC biases exert larger relative contributions than R c biases in both reanalysis datasets. Specifically, the fractional contribution of TCC biases is approximately 1.8–2.3 times that of R c biases for monthly R s biases and 2.4–3.8 times for trend biases. Together, these findings highlight the dominant role of cloud-related biases in shaping both the average state and long-term trends of R s in twentieth-century reanalyses, underscoring the need for improved representations of cloud processes and their interactions with radiation in future reanalysis products. Significance Statement The twentieth-century reanalysis products offer century-scale climate data, making them essential for studying decadal climate variations. The objective of this study is to evaluate the performance of five mainstream twentieth-century reanalysis datasets in simulating surface incident solar radiation over mainland China. Although Coupled European Reanalysis of the Twentieth Century (CERA-20C) and 20CRv3 show a more accurate representation of the climate state, substantial uncertainties remain, particularly in their representation of long-term trends. Our results demonstrate that total cloud cover biases play a dominant role in shaping both the average level and long-term trends of surface incident solar radiation in twentieth-century reanalyses, whereas aerosol-related effects are comparatively weaker. Therefore, improving cloud–radiation interactions in future reanalysis products is of paramount importance.