An Innovative Geo-AI Approach in Estimating High-Resolution Urban Ambient Fungal Spore Variations

The spatial distribution of fungal spores varies seasonally and regionally based on meteorological conditions and surrounding land use patterns. To investigate the variation in ambient fungal spore concentration, this study is the first of its kind to develop geospatial-artificial intelligence (Geo-AI) models for estimating total fungal spore variation. Geospatial predictor variables including air pollutants, meteorological parameters, land use and land cover allocations, road networks, landmarks, and vegetation indices derived from spectral data surrounding sampling sites were collected for the Geo-AI model development. The Shapley Additive Explanations (SHAP) index was utilized as machine learning explainability tool for clarifying the importance of each variable. The most influential variables identified by SHAP were incorporated into machine learning algorithms, including random forest, gradient boosting machine (GBM), XGBoost, LightGBM, and CatBoost. The developed Geo-AI model achieved a high prediction accuracy with an R² value of 0.96 and a root mean square error (RMSE) value of 0.03 using GBM algorithm. The identified variables highlighted the significant influence of meteorological parameters, PM₁₀, spectrum-based vegetation index, and land-use/land covers allocations on fungal spore concentrations. Estimation maps revealed elevated fungal spore levels in mountainous areas, contrasting with lower levels observed in urban environments. The proposed Geo-AI model addressed previous limitations of estimating fungal spores in large areas. The estimation of fungal spore concentration can provide valuable exposure information for further environmental epidemiological analysis.