CHDSR：基于REST2_v9.1和集成机器学习技术的中国日均表面太阳辐射散射数据集（1980–2022）

散射太阳辐射是地表太阳辐射的重要组成部分，对光伏发电和碳循环具有重要贡献。基于 REST2_v9.1 模型和集成机器学习模型，我们利用 ERA5 与 MERRA2 再分析数据，更新并构建了全国范围的 43 年（1980–2022）日均散射太阳辐射数据集（CHDSR-updated），空间分辨率为 10 km。与以往的散射太阳辐射数据集相比，该改进数据集通过引入物理模型、优化机器学习模型和改进模型输入，有效提升了散射太阳辐射的估算精度。与地面观测数据验证结果表明，CHDSR-updated 数据集在样本交叉验证中表现出高且稳定的性能，相关系数（R）、均方根误差（RMSE）和平均绝对误差（MAE）分别为 0.943、13.901 W/m² 和 10.466 W/m²，相比之前提高了 0.063，RMSE 和 MAE 分别下降了 5.639 W/m² 和 4.404 W/m²。

Scattered solar radiation is a critical component of surface solar radiation, making significant contributions to photovoltaic power generation and the carbon cycle. Based on the REST2_v9.1 model and ensemble machine learning models, we utilized ERA5 and MERRA2 reanalysis data to update and construct a national-scale 43-year (1980–2022) daily mean scattered solar radiation dataset (CHDSR-updated) with a spatial resolution of 10 km. Compared with existing scattered solar radiation datasets, this improved dataset effectively enhances the estimation accuracy of scattered solar radiation by introducing physical models, optimizing machine learning models, and refining model inputs. Validation results against ground-based observation data demonstrate that the CHDSR-updated dataset exhibits high and stable performance in sample cross-validation, with correlation coefficient (R), root mean square error (RMSE), and mean absolute error (MAE) of 0.943, 13.901 W/m², and 10.466 W/m², respectively. Compared with previous datasets, the correlation coefficient has increased by 0.063, while RMSE and MAE have decreased by 5.639 W/m² and 4.404 W/m², respectively.