Accurate surface downward shortwave radiation (SDSR) is essential for solar resource assessment, photovoltaic applications, and land–atmosphere studies. Although ERA5 is widely used in radiation-related research, its SDSR estimates over China still show considerable uncertainties under complex topographic and climatic conditions. Using hourly observations from the 162-station China Meteorological Administration (CMA) radiation network during April 2024–March 2025, of which 160 stations were retained after quality control, this study systematically evaluated ERA5 SDSR and developed a physics-guided Light Gradient Boosting Machine (LightGBM) correction framework. Raw ERA5 exhibits a strong systematic positive bias (PBIAS = 57.40%, ME = 124.2 W/m²) together with a pronounced nonlinear structural bias, characterized by overestimation under low-radiation conditions and underestimation under high-radiation conditions. The largest errors occur in the Southern Monsoon region in summer and the Northwest Arid region in spring, indicating the combined effects of cloud extinction, aerosol attenuation, and terrain-related representativeness differences. To address these mechanisms, the correction model incorporates physically relevant predictors from ERA5 and Copernicus Atmosphere Monitoring Service (CAMS), including cloud microphysical variables, aerosol optical depth, solar geometry, and elevation. SHapley Additive exPlanations (SHAP) analysis shows that the learned correction behavior is broadly consistent with known radiative-transfer processes. On the independent station hold-out test set, the correction increases the Pearson correlation coefficient from 0.8680 to 0.8967 and reduces RMSE from 173.1 to 100.8 W/m², while substantially suppressing the strong positive bias of raw ERA5. Additional robustness tests, including season-blocked validation, interpolation-sensitivity analysis, ablation experiments, and multi-model comparison, further support the stability of the framework. External benchmarking against FY-4B and Himawari also shows that the corrected ERA5 substantially narrows the gap relative to independent geostationary satellite products. Overall, the proposed framework provides an effective and physically interpretable approach for improving ERA5 SDSR over China.

Mountain photovoltaic (PV) stations are increasingly deployed in complex terrain, where generation is jointly controlled by solar-resource variability, near-surface meteorology, and local topography. However, the quantitative contribution of topographic factors to regional-scale PV generation remains insufficiently evaluated, and many prediction studies rely on single-station or short-term records. In this study, monthly measured generation from 118 standardized village-level mountain PV stations in Badong County, western Hubei Province, China (2019–2021), was integrated with Solargis Global Horizontal Irradiance (GHI)-related solar-resource data, high-resolution gridded meteorological data, a 25 m digital elevation model, seasonal-cycle variables, and historical-generation features. After seasonally grouped median-absolute-deviation (MAD) outlier screening, GIS-based spatial matching, terrain extraction, and viewshed-derived shading analysis, regression models and climatology baselines were compared under both chronological validation and station-exclusion spatial cross-validation. Under the strict chronological validation, CatBoost achieved the best temporal performance among the tested models (R² = 0.3119, MAE = 2719.7 kWh, RMSE = 3245.6 kWh), slightly outperforming the monthly climatology baseline. In the station-exclusion spatial cross-validation, XGBoost achieved the highest mean R² (0.8659), indicating good spatial transferability to unseen stations. Correlation and partial-correlation analyses showed that the temperature-related variable group and monthly radiation were the dominant meteorological controls, whereas elevation, slope, and terrain shading showed weak direct correlations with monthly generation for already-sited stations. Annual 90% prediction intervals were further estimated using residual bootstrapping, with an empirical coverage of 94.9%. The proposed framework provides a practical basis for monthly generation forecasting and operational assessment of already-built distributed PV stations in mountainous regions, while its application to greenfield site selection requires additional site engineering and near-field obstruction information.

ERA5 100 m wind speed provides long-term and spatially continuous information for regional wind-resource assessment, but its grid-scale representation may introduce terrain- and wind-regime-dependent biases in complex terrain. This study evaluated ERA5 using 459,630 valid hourly observation–reanalysis pairs from 64 wind masts and developed a segmented residual-correction framework incorporating wind-regime, terrain, location, and temporal predictors. Three statistical correction methods and four tree-based models were compared under identical training and validation samples. The main analysis used a stratified random holdout, supplemented by chronological and tower-level spatial holdouts and repeated high-wind experiments. In the random validation subset, raw ERA5 yielded R = 0.669, R² = 0.307, RMSE = 2.261 m s⁻¹, MAE = 1.664 m s⁻¹, and ME = −0.904 m s⁻¹. Random forest achieved the best paired hourly performance, increasing R to 0.850 and reducing RMSE and MAE to 1.433 and 1.078 m s⁻¹, respectively. Its RMSE reductions remained positive but decreased to 14.15% and 14.47% under chronological and spatial holdouts. Terrain relief was strongly associated with tower-level raw ERA5 errors, although its controlled inclusion produced only a modest additional RMSE reduction of approximately 0.7%. Pooling the >9 m s⁻¹ training tail improved high-wind stability and outperformed a separately trained >12 m s⁻¹ model in 17 of 20 experiments. Random forest was preferable for paired hourly reconstruction, whereas quantile mapping more closely reproduced pooled theoretical wind-energy indicators. These results show that ERA5 bias correction should be selected according to the intended application and evaluated using complementary temporal, spatial, distributional, and high-wind diagnostics.