AI-Augmented Clear-Sky Algorithms for High-Granularity Solar Irradiance Modelling
Journal of Climate Change · 29.06.2026 · 12 (1) · 20
DOI: 10.70917/jcc-2026-009 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Climate Change, online erschienen am 29.06.2026. Verfasst von Kudzanayi Chiteka und 1 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Accurate short-term solar irradiance forecasting is important for reliable photovoltaic operation in regions affected by rapid cloud movement and atmospheric variability. However, clear-sky models cannot adequately respond to cloud-driven irradiance fluctuations, while purely data-driven models may suffer from weak physical consistency and reduced generalisability. The present study developed a physics-informed hybrid framework in which clear-sky irradiance was used as the physical baseline, while machine learning was employed to learn the residual atmospheric deviations caused by clouds, aerosols, moisture and near-surface meteorology. The proposed model was evaluated against persistence, clear-sky, AI-only and standalone Gated Recurrent Unit benchmark models using 10-min-ahead global horizontal irradiance prediction. During the Roodepoort summer test period, the hybrid model achieved an RMSE of 79 W/m2, MAE of 59 W/m2, MBE of −8 W/m2 and R2 of 0.91, corresponding to RMSE reductions of 32%, 44%, 27% and 18% when compared with the persistence, clear-sky, AI-only and Gated Recurrent Unit models, respectively. Furthermore, the model maintained seasonal RMSE values between 79 and 96 W/m2 without retraining, retained useful transferability at external Southern African Universities Radiometric Network stations with a mean external RMSE of 95.5 W/m2 and R2 of 0.86, and produced calibrated 95% prediction intervals with 93% empirical coverage. These findings showed that clear-sky constrained residual learning can improve deterministic accuracy, uncertainty reliability and seasonal robustness, thus supporting more reliable photovoltaic forecasting, grid integration and operational energy management under variable sky conditions.
Zitieren
APA
Chiteka, K., & Enweremadu, C. (2026). AI-Augmented Clear-Sky Algorithms for High-Granularity Solar Irradiance Modelling. Journal of Climate Change, 12(1), 20. https://doi.org/10.70917/jcc-2026-009
Vancouver
Chiteka K, Enweremadu C. AI-Augmented Clear-Sky Algorithms for High-Granularity Solar Irradiance Modelling. Journal of Climate Change. 2026;12(1):20. doi:10.70917/jcc-2026-009
BibTeX
@article{Chiteka2026,
author = {Chiteka, Kudzanayi and Enweremadu, Christopher},
title = {AI-Augmented Clear-Sky Algorithms for High-Granularity Solar Irradiance Modelling},
journal = {Journal of Climate Change},
year = {2026},
volume = {12},
number = {1},
pages = {20},
doi = {10.70917/jcc-2026-009}
}
EUVIMED