Space-time pattern mining for municipal-level short-term population forecasting in Serbia

Due to various demographic, socio-economic, and environmental factors, population fluctuations occur across different spatial scales. Understanding these changes is crucial for effective spatial management. For decades, demographers have used population forecasting techniques to improve comprehension and enhance management strategies. However, traditional models have often neglected the spatial dimension of population dynamics. An innovative GIS-based modeling framework integrating the Space-Time Pattern Mining (STPM) tool with extrapolative forecasting techniques is introduced to generate short-term, municipal-level population forecasts for Serbia by 2037. Using historical population data from 1991 to 2022, three forecasting models (Curve Fit, Exponential Smoothing, and Forest-based) are employed to forecast population trends over a 15-year period (2023–2037). The most accurate model for each municipality is identified using the Evaluating Forecasts by Location method. The results indicate that Serbia will experience further population decline, characterized by a decrease in medium- and large-sized municipalities and a simultaneous increase in small-sized ones, continuing the long-term depopulation trend. The proposed framework demonstrates the potential of geospatial analyses to enhance demographic forecasting by combining spatial and temporal dimensions within a unified analytical structure.