Evaluation of three chemical transport models for extreme PM10 events in Morocco and improvement of forecasts using hybrid CTM-Random Forest models.

Particulate matter (PM10), originating from natural sources such as dust storms and human activities, significantly impacts public health and the environment. Accurate forecasting of PM10 concentrations is therefore crucial. This study aims to predict PM10 levels during extreme pollution episodes in Morocco using chemical transport models (CTMs) and to enhance these predictions with machine learning (ML) models. Three CTMs, namely CHIMERE, SILAM, and CAMS were validated during a high PM10 pollution episode that occurred between 2020 and 2021 in five Moroccan cities. Subsequently, two hybrid models combining SILAM and CHIMERE with the Random Forest (RF) algorithm were developed to improve the predictions. The performance of these hybrid models was assessed by comparing observed in situ data with predictions using statistical measures. The results demonstrate that CHIMERE and SILAM accurately forecast peak pollution levels, with strong correlations across all cities. CHIMERE, in particular, showed excellent accuracy in simulating concentration trends. In contrast, the CAMS model performed poorly due to its low spatial resolution. On the other hand, the hybrid models of SILAM-RF and CHIMERE-RF greatly improved the bias and error reductions, giving a better fit for the observed data. CHIMERE-RF performed well, offering more precise and reliable forecasts for extreme pollution events. These findings are essential for air quality monitoring during severe PM10 episodes, aiding in implementing preventive measures to minimize health risks related to PM10.