Air pollution macro-regions identification using machine learning and spatio-temporal analysis

Abstract

Air pollution caused by suspended particulate matter (PM) remains one of the key environmental challenges in Poland, particularly in the context of public health and spatial planning. This study presents a spatio-temporal analysis based on data from 173 air quality monitoring stations collected between 2015 and 2023. Advanced unsupervised clustering methods based on the Dynamic Time Warping (DTW) metric were applied to identify spatial patterns of pollution at both daily and annual timescales. Based on over 13 million observations, four macroregions were delineated, along with a set of sixteen clusters allowing for assessment of local anomalies. A significant variation in median PM 10 concentrations has been observed across macroregions, ranging from 19.7 to 27.18 μ g / m 3 . The results revealed the significant role of topographic, urban, and microclimatic factors in shaping the spatial distribution of particulate matter. Urbanized areas in southern Poland (Silesia, Lesser Poland) formed distinctly isolated clusters with high PM levels, in contrast to the stable, low-emission northern lowlands. The analysis further demonstrated that regional policies may be ineffective without supra-regional coordination. These findings show the need of including the high-resolution data analyses into environmental and public health planning to effectively limit the impacts of air pollution.