A Terrain-Aware Semi-Empirical Framework for Simulating Surface Solar Radiation in Data-Sparse Complex Terrain

Topographic slope, aspect, terrain obstruction, and cloud variability jointly produce pronounced spatiotemporal heterogeneity in surface solar radiation (SSR) over complex terrain, while sparse ground observations in mountainous areas limit high-resolution estimation and independent reliability assessment. We therefore developed and tested a terrain-aware semi-empirical SSR framework for data-sparse complex terrain. Its methodological novelty involves the use of one precomputed directional horizon to consistently constrain terrain shadowing, sky-view factor, and terrain-view factor, combined with time-step clear-sky irradiance; daily cloud forcing; and explicit direct, diffuse, and first-order terrain-reflected components. At multiple time scales, the framework was applied to Liancheng (LC), Wuyishan (WYS), and Jinggangshan (JGS), three regions with contrasting terrain and climatic settings. Evaluation combined analytical terrain tests, benchmarks against HORAYZON and r.sun, ground-station observations, leave-one-station-out validation, module ablation, pixel-wise error-frequency analysis, correlation tests, and uncertainty analysis. The directional-horizon scheme represented sky obstruction and terrain shading, with close agreement with HORAYZON results. Annual SSR was 1166.10, 1000.85, and 922.95 kWh/m2 in LC, WYS, and JGS, respectively, yielding the order LC > WYS > JGS; WYS and JGS were strongly dominated by diffuse radiation. Independent station validation produced R2 values of 0.588–0.704 and RMSE values of 40.4–46.6 W/m2, while absolute-error frequency analysis showed that 61.4–70.2% of held-out daily estimates had absolute errors below 40 W/m2. The framework reproduced the principal temporal variability of SSR across the three mountain regions, although performance remained site-dependent and showed systematic underestimation. Uncertainty analysis identified cloud-cover input and its empirical parameters as the main sources of simulation uncertainty. Overall, the framework provides a physically interpretable and computationally tractable approach for regional SSR estimation in observation-sparse mountain environments.