Using Advanced GeoAI-Based Ensemble Mixed Spatial Prediction Model to Estimate Ambient Ammonia

Abstract Ambient ammonia (NH 3 ) contributes to odor and particulate matter (PM), making it essential to understand its characteristics for effective reduction of both. This study developed a Geo-AI-based ensemble mixed spatial model (EMSM) to predict NH 3 concentrations in Tainan, Taiwan. Morning average NH₃ data from 45 sites (2021–2022) were used to construct and validate the model. We employed Geo-AI techniques, integrating kriging, five machine learning algorithms, and ensemble methods to develop an EMSM. The EMSM estimated NH 3 variations over a two-year period by applying in-situ, geospatial, meteorological, and social factors. The results indicated the EMSM achieved a model performance of up to 94%, surpassing all other compared models. The results of the spatiotemporal resolution analysis suggest that fluctuations in NH 3 may be influenced by NH 3 kriging based, paddy, petrochemical raw material industry, all road, and public restroom quantity. These findings offer precise estimates that can inform pollution control strategies and support epidemiological study. Graphical abstract