Dust pollution presents a significant environmental challenge in arid and semi-arid regions, adversely impacting public health, agriculture, water resources, and infrastructure. This study aimed to forecast PM10 dust pollution in Ahvaz by utilizing MODIS satellite AOD data collected from 2016 to 2022, alongside four machine learning techniques: XGBoost, SGB, LSTM, and MLP. Initially, the HYSPLIT model was employed to track dust paths and analyze their frequency. Subsequently, true color satellite images were used to identify virtual stations along these dust paths, and the AOD values at these locations served as input data for the forecasting models. The results indicated a significant correlation between the satellite AOD data and actual PM10 data for 2021 and 2022, with a significance level of 0.01 and determination coefficients of 0.89 and 0.85, respectively. Dust tracking maps demonstrated that dust was being transported from Iraq to Ahvaz. Average AOD data from eight virtual stations, with a one-day lag, was used for weekly dust forecasting. Model evaluation results revealed that XGBoost outperformed the other models, achieving a mean absolute error of 0.08 and a coefficient of determination of 0.87. The SGB and LSTM models followed closely in performance, while the MLP model showed the lowest accuracy. Predictions for the following seven days indicated that boosting models, particularly XGBoost, effectively captured data fluctuations. This research demonstrates that AOD satellite data, combined with advanced machine learning techniques, are valuable tools for forecasting and managing dust pollution.