A sensorless and explainable decision-support framework for risk-aware photovoltaic cleaning prioritization in desert environments using public environmental datasets

Photovoltaic (PV) systems in desert environments experience substantial soiling losses, while dedicated soiling sensors can increase monitoring cost and maintenance burden. This study develops a sensorless, risk-aware decision-support framework for PV cleaning prioritization using daily NASA POWER meteorological data, with PVGIS-SARAH3 used as an independent solar-resource consistency check at the target coordinate. The framework combines Consecutive Dry Days (CDD), a Mudding Risk Indicator (MRI), a Dust Soiling Potential (DSP) index, bounded dust accumulation, rainfall-driven natural cleaning, soiling-ratio estimation, and a hybrid maintenance-risk matrix. Application to Hafar Al-Batin for 2023–2025 produced mean estimated performance losses of 10.26–12.97%, with maximum losses near 18.1%. Parameter sensitivity and 3,000-run Monte Carlo analysis showed that the attenuation coefficient and dust-storage capacity mainly govern the absolute loss magnitude; the three-year mean performance loss was 11.77% with a 95% uncertainty interval of 9.63–14.01%. A Random Forest rule-emulation analysis, using only exogenous weather variables to avoid direct target leakage, achieved 84.09% holdout accuracy and 81.66% mean five-fold cross-validation accuracy. SHAP analysis provided formal post-hoc explanation of the rule-emulation model. A quantitative strategy benchmark demonstrated explicit trade-offs among cleaning frequency, estimated energy recovery, and break-even cleaning cost. The framework is therefore positioned as a low-data cleaning-prioritization tool rather than a field-calibrated optimizer; direct PV soiling measurements remain necessary for site-specific operational calibration.