Dust storms are a destructive environmental phenomenon causing significant damage each year, particularly in arid and semi-arid regions. This study employed three hybrid metamodels, combining artificial intelligence, machine learning, and Box-Jenkins methods, to model and forecast dust events. The research provides a 20-year outlook on this environmentally damaging issue, focusing on Khuzestan Province, Iran. The models were evaluated for predicting the Frequency of Dust Stormy Days index. Results showed the K-Nearest Neighbors-Particle Swarm Optimization-Nonlinear AutoRegressive eXogenous model performed best, followed by the Support Vector Machine-Ant Colony Optimization-Seasonal AutoRegressive Integrated Moving Average model, and then the Gated Recurrent Unit-Long Short-Term Memory-Self-Exciting Threshold AutoRegressive model. Notably, incorporating extensive historical dust storm data from prior seasons into these hybrid models did not significantly enhance prediction accuracy. For instance, using the Frequency of Dust Stormy Days index from four prior seasons improved the Root Mean Square Error for the best performing model. Similarly, the Mean Absolute Error for another model was reduced from approximately 0.08-0.075 days by applying data from four to one previous seasons. These findings indicate a sharp rise in the Frequency of Dust Stormy Days index in Khuzestan Province in the coming years, which could lead to considerable social, economic, and environmental impacts.