Abstract

Environmental pollution is a serious international issue with far reaching consequences on the health and biologist sustainability of the ecosystem and socio-economic progress especially in the fast urbanizing and industrializing areas. Traditional ground-based global air quality observation systems have real-valued measurements but have the disadvantage of sparsity in their spatial coverage as well as their inability to scale to large scales hindering extensive exposure measurement and assessment of their health impacts on a large population. To address these constraints, this paper suggests a more complex machine learning (ML) and deep neural network (DNN)-based system to detect massive environmental pollution, quantify population exposure, and compute biomedical effects of desalination plants with multisource remote sensing data. Aerosolar optical depth (AOD) and atmospheric trace gas concentrations (NO 2, SO 2, and O 3) derived by satellites are combined by using a multimodal deep learning architecture based on convolutional neural networks (CNNs) that extract spatial features, on top of which are transaction long short-term memory (BiLSTM) networks that perform temporal, and attention topics that assign adaptive weights to features. The suggested framework has generated high-resolution spatio-temporal maps of pollution concentration and has generated population-weighted exposure indices to measure short- and long-term patterns of exposure. In addition, health risk models which are based on data are used to evaluate the relationship between respiratory health outcomes and the level of pollution exposure. Experimental tests indicate that the suggested method performs significantly better in comparison to standard regression and individual ML models, and it leads to significant RMSE and the coefficient of determination (R 2) improvement. Exposure-response analysis indicates that there are significant correlations between high levels of pollution and the high health risks associated with respiratory health. In general, the findings indicate the efficiency of AI-based remote sensing systems to combine environmental pollution, exposure assessment and population health long-range surveillance activities, which provide significant information to support evidence-based policies and decision-making about the environment.