Elaboration a Machine Learning-based Modelling of Dew Water as a Function of Meteorological Parameters

Soiling significantly reduces the performance of photovoltaic (PV) systems, particularly in arid and semi-arid regions where water scarcity limits conventional cleaning methods. Dew water represents a sustainable, passive alternative for surface maintenance, but its availability is highly sensitive to local meteorological conditions. This study develops a machine learning-based modeling framework to predict dew water using key meteorological variables including the ambient temperature, relative humidity, wind speed, and clarity index. A one-year experimental campaign was conducted in Rabat, Morocco, generating a high-resolution dataset of daily dew collection and corresponding environmental parameters. In this experimental study, a mean transmittance ratio of 0.973 was obtained. Five regression algorithms Support Vector Regression (SVR), Gradient Boosting, XGBoost, Decision Tree, and Random Forest were evaluated for predictive performance. Among them, the Random Forest model achieved the highest accuracy with an R² of 0.969, a mean absolute error (MAE) of 0.0056, and a root mean square error (RMSE) of 0.0100. These results underscore the model’s robustness in capturing the nonlinear dynamics of dew formation. The proposed approach offers a reliable, data-driven tool for forecasting dew availability, facilitating the integration of passive cleaning strategies in PV maintenance planning particularly in resource-constrained environments.