CHDNI: Daily Surface Direct Normal Irradiance Dataset in China (1980-2022, 10 km) based on REST2_v9.1 Model and Stacking Machine Learning Techniques

Evaluating, mapping, and monitoring the high-quality direct normal irradiance (DNI) is of vital importance for the proper design, financing, and operation of solar power plants using concentrating technologies. We constructed a 43-year (1980–2022) daily gridded DNI dataset (CHDNI), using ERA5 and MERRA_2 reanalysis data, with a spatial resolution of 10 km through a developed hybrid model, which integrated REST2_v9.1 radiation transfer model and stacking machine learning model. The hybrid model effectively enhances the accuracy of DNI estimation by introducing physical models, improving machine learning models, and optimizing model inputs. The validation results, with ground-based measurements, showed that hybrid model had a high and stable performance with the correlation coefficient (R), root-mean-square error (RMSE), and mean absolute error (MAE) for the sample-based cross-validations of 0.92, 32.04Wm−2, and 23.68Wm−2, respectively.  Considering the large volume of this dataset, we only uploaded the CHDNI data during 2008/1/1-2008/12/31. CHDNI had good consistency with observations in 2008, where R, RMSE and MAE were 0.92, 29.13Wm−2, and 21.62Wm−2, respectively. For complete data series during 1980-2022, please contact us via email ([e-mail]).

高质量直接法向辐照度（Direct Normal Irradiance, DNI）的评估、制图与监测，对于聚光型太阳能电站的合理设计、融资与运维至关重要。本研究依托开发的混合模型——整合了REST2_v9.1辐射传输模型与堆叠机器学习模型，结合ERA5与MERRA_2再分析数据，构建了时空分辨率为10 km的1980—2022年共43年逐日格网化直接法向辐照度数据集（CHDNI）。该混合模型通过引入物理模型、优化机器学习模型结构与输入特征，有效提升了直接法向辐照度的估算精度。基于地面实测数据的验证结果表明，在样本交叉验证中，该混合模型表现出优异且稳定的性能，相关系数（R）、均方根误差（RMSE）与平均绝对误差（MAE）分别为0.92、32.04 W·m⁻²与23.68 W·m⁻²。考虑到该数据集体量庞大，本研究仅上传了2008年1月1日至2008年12月31日的CHDNI数据。2008年的CHDNI数据与实测值具备良好的一致性，其相关系数、均方根误差与平均绝对误差分别为0.92、29.13 W·m⁻²与21.62 W·m⁻²。若需获取1980—2022年完整的数据集序列，请通过邮箱[e-mail]与我们联系。