Mitigating Air Pollution Risks with Deep Learning: A Quantum-Optimized Approach for Nitrogen Dioxide Prediction in Los Angeles

Air pollution causes about seven million pre mature deaths globally every year, making it a critical issue that requires urgent attention. The key to mitigating its devastating effects lies in understanding its nature, identifying sources and trends, and predicting its. Accurate Real-time air pollution forecasting is a challenging task due to its spatiotemporal dynamics, requiring sophisticated modeling approaches. In our study, employed the Sequential Array-based Convolutional LSTM (SACLSTM) framework, which captures spatial and temporal correlations by integrating deep CNNs for spatial analysis with deep LSTM models for temporal prediction. To further enhance the model's accuracy, optimized the SACLSTM parameters using the Quantum-based Draft Mongoose Optimization Algorithm (QDMOA). Using ten days of nitrogen dioxide (NO₂) data from Los Angeles County, developed a sequential encoder-decoder network capable of predicting air pollution levels ten days into the future. By reformatting satellite air quality images into a 5D tensor, achieved precise predictions of nitrogen dioxide concentrations across various locations and time periods in Los Angeles. Our results are thoroughly documented with metrics and visualizations, clearly demonstrating the factors behind the improved accuracy. The comparison of results highlights the effectiveness of our approach in providing reliable air pollution forecasts.