Spatio-Temporal Analysis and Modeling of Air Pollution in Almaty, Kazakhstan

This thesis investigates the spatial and temporal dynamics of fine particulate matter (PM2.5) in Almaty, a city characterized by strong winter pollution, complex mountainous terrain, and a sparse monitoring network. To overcome limitations of uneven sensor coverage, the analysis integrates air-quality, meteorological, topographic, and urban data from 2020 to 2025 and applies a multi-stage modelling framework that increases in spatial and temporal complexity. City-level dynamic regression models first quantify broad meteorological influences and seasonal heating effects. Daily sensor-level Bayesian hierarchical models then incorporate spatial heterogeneity through sensor-specific intercepts, static geographic predictors, and autoregressive temporal dependence. Finally, a weekly spatio-temporal INLA-SPDE model represents PM2.5 as a continuous latent field, enabling high-resolution prediction across the entire urban domain. Across all modelling stages, lower temperatures, stagnant high-pressure conditions, and absence of precipitation consistently emerge as key drivers of elevated PM2.5, while the heating season remains a strong positive predictor independent of meteorology. Spatial analyses reveal pronounced concentration gradients, with higher pollution in low-lying central and north-western districts and persistent hotspots near the coal-fired CHP-1 and CHP-2 plants. The INLA-SPDE model captures these structures and produces smooth prediction maps that highlight strong winter accumulation, cleaner conditions in summer, and stable interannual patterns shaped by emissions, topography, and atmospheric conditions. The results offer a comprehensive depiction of PM2.5 behaviour in Almaty and provide insights for air-quality management. They indicate that expanding monitoring along the city’s periphery, especially in higher-uncertainty areas, would improve spatial coverage, while policies targeting heating emissions are likely to produce substantial wintertime benefits. Overall, the thesis demonstrates the value of combining hierarchical and SPDE-based methods for analysing pollution in cities with complex terrain and limited monitoring networks, and establishes a statistical foundation for future predictive and policy-oriented applications.