Related Experiment Video
Updated: Jan 18, 2026
08:16
Collecting and Processing Drone-based Remotely Sensed Data for Use in Forest Recovery Monitoring
Published on: October 24, 2025
494
Enhancing 5-Day Particulate Matter (PM10) Forecasts in Morocco Using U-Net: A Deep Learning Approach.
Anass Houdou1,2, Kenza Khomsi3, Luca Delle Monache4
1Mohammed VI International School of Public Health, Mohammed VI University of Sciences and Health, Casablanca, Morocco.
Atmospheric Research
|September 11, 2025
Summary
This study introduces a U-Net deep learning model to enhance particulate matter (PM10) forecasts in Morocco, significantly improving accuracy over existing methods. The model offers reliable predictions, aiding public health by providing crucial early warnings for PM10 pollution.
Area of Science:
- Environmental Science
- Atmospheric Science
- Artificial Intelligence
Background:
- Accurate prediction of particulate matter (PM10) is vital for public health and risk prevention.
- Existing PM10 forecasts, like those from Copernicus Atmosphere Monitoring Service (CAMS), require enhancement for improved accuracy.
- The Middle East and North Africa (MENA) region lacks advanced deep learning applications for PM10 forecasting.
Purpose of the Study:
- To improve the accuracy of five-day PM10 forecasts over Morocco using a novel U-Net deep learning model.
- To adapt the U-Net architecture for variable resolution outputs, preserving spatial details without interpolation.
- To establish the first deep learning-based PM10 forecasting system in the MENA region.
Main Methods:
- Utilized a modified U-Net deep learning model to post-process CAMS PM10 forecasts.
- Employed CAMS reanalysis data as a reference for model training.
- Modified the U-Net architecture to handle different input and output resolutions, avoiding interpolation.
- Compared the U-Net model against CAMS forecasts and the Analog Ensemble (AnEn) model.
Main Results:
- The U-Net model significantly outperformed CAMS and AnEn baselines in key metrics (MAE, RMSE, R², IOA, bias).
- Improvements were particularly notable in dust storm-prone regions before the CAMS forecast upgrade in mid-2023.
- The model demonstrated continued accuracy improvements in late 2023, though CAMS upgrade cycles impacted error patterns.
- U-Net accurately captured high pollution levels, predicting values up to 500 μg/m³.
Conclusions:
- The U-Net model offers a significant advancement in PM10 forecasting accuracy for operational use.
- Regular retraining of the U-Net model with updated data is crucial for maintaining forecasting reliability.
- This deep learning approach provides valuable tools for early warnings, mitigating health and environmental impacts of PM10 pollution.
