Value of Aerosol Information for Day-Ahead Solar Irradiance Forecasting in a Dust-Prone Climate: A Multi-Site Evaluation of Machine-Learning Post-Processing in Iraq
ReCIBE, Revista electrónica de Computación, Informática, Biomédica y Electrónica · 27.10.2026 · 15 (3)
DOI: 10.32870/2961bn58 ↗Kurz gefasst
Fachartikel in der Zeitschrift ReCIBE, Revista electrónica de Computación, Informática, Biomédica y Electrónica, online erschienen am 27.10.2026. Verfasst von omar Gheyath Ibrahim und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-SA 4.0).
Abstract (Originalsprache)
Day-ahead photovoltaic scheduling in Iraq relies on numerical weather prediction (NWP) models that represent aerosol through climatology, whereas mineral dust often departs strongly from climatology. This study quantifies whether aerosol information improves machine-learning post-processing of dayahead irradiance forecasts. Hourly ICON and GFS forecasts issued one day ahead, Copernicus Atmosphere Monitoring Service (CAMS) aerosol fields and SARAH-3 satellite irradiance were assembled for Mosul, Baghdad, Rutba and Basra (2024–2025). Ridge regression, gradient-boosted trees (GBM) and a multilayer perceptron were trained on 2024 and tested on 2025 with five aerosolinformation levels, including a physics-informed predictor combining the aerosol optical depth (AOD) anomaly with relative air mass. Without aerosol input, the GBM reduced the root-mean-square error (RMSE) of global horizontal irradiance (GHI) from 69.6 to 63.1 W m⁻² and of direct normal irradiance (DNI) from 186.3 to 131.3 W m⁻² relative to raw ICON. Under cloud-free forecasts, NWP overestimation increased with AOD. CAMS predictors reduced the RMSE in cloud-free dusty hours (AOD ≥ 0.6) by 4.5–5.3% for GHI and 6.3–10.6% for DNI (day-block bootstrap, p ≤ 0.001), but changed the annual GHI RMSE by less than 0.5%, because 71% of dusty hours were cloudy. Aerosol persistence at issue time gave no benefit. Dusty days doubled the daily photovoltaic energy error, and CAMS predictors reduced the dusty-day bias from 0.16 to 0.03 kWh kWp⁻¹. Dust information is therefore valuable mainly for beam irradiance, in clear dusty conditions and when a genuine aerosol forecast is available. Keywords: solar forecasting; dust aerosol; numerical weather prediction; post-processing; gradient boosting; direct normal irradiance; photovoltaic; Iraq
Zitieren
APA
Ibrahim, O. G., Majeed, M. T., & Habbib, A. M. (2026). Value of Aerosol Information for Day-Ahead Solar Irradiance Forecasting in a Dust-Prone Climate: A Multi-Site Evaluation of Machine-Learning Post-Processing in Iraq. ReCIBE, Revista electrónica de Computación, Informática, Biomédica y Electrónica, 15(3). https://doi.org/10.32870/2961bn58
Vancouver
Ibrahim OG, Majeed MT, Habbib AM. Value of Aerosol Information for Day-Ahead Solar Irradiance Forecasting in a Dust-Prone Climate: A Multi-Site Evaluation of Machine-Learning Post-Processing in Iraq. ReCIBE, Revista electrónica de Computación, Informática, Biomédica y Electrónica. 2026;15(3). doi:10.32870/2961bn58
BibTeX
@article{Ibrahim2026,
author = {Ibrahim, omar Gheyath and Majeed, Mohammed Tariq and Habbib, Abdulla Mohannad},
title = {Value of Aerosol Information for Day-Ahead Solar Irradiance Forecasting in a Dust-Prone Climate: A Multi-Site Evaluation of Machine-Learning Post-Processing in Iraq},
journal = {ReCIBE, Revista electrónica de Computación, Informática, Biomédica y Electrónica},
year = {2026},
volume = {15},
number = {3},
doi = {10.32870/2961bn58}
}
EUVIMED