anasshoudo/U-Net-Model-for-PM10: Predicting particulate matter (PM10) values using the U-Net model

This notebook presents the implementation of the U-Net architecture employed in (Houdou et al., 2025) for forecasting PM10 concentrations across Morocco. The study (Houdou et al., 2025) enhances short-term predictions—up to five days ahead—of particulate matter with a diameter less than 10 µm (PM10), leveraging a deep learning approach based on U-Net. The model refines PM10 forecasts generated by the Copernicus Atmosphere Monitoring Service (CAMS) through post-processing, using CAMS reanalysis data as a reference for supervised learning. The U-Net architecture was modified to produce outputs at a different spatial resolution than the inputs, which removes the need for interpolation while maintaining important spatial features.   References Houdou, A., Khomsi, K., Delle Monache, L., Hu, W., Boutayeb, S., Belyamani, L., Abdulla, F., El Badisy, I., Al-Delaimy, W. K., & Khalis, M. (2025). Enhancing 5-Day Particulate Matter (Pm10) Forecasts in Morocco Using U-Net: A Deep Learning Approach. SSRN. doi.org