A Spatio-Temporal Orthophoto-Based Framework for Long-Term Land Cover Prediction

Long-term land cover dynamics constitute a major challenge for sustainable urban planning and environmental monitoring. Historical orthophotos provide high-resolution spatial information over several decades, but their exploitation for predictive modeling remains limited due to data heterogeneity, degradation, and complex spatio-temporal dependencies. This paper proposes ORTHO-STPM, a spatio-temporal framework for land cover prediction based on historical orthophotos. The framework integrates robust preprocessing, machine learningbased feature extraction, and hybrid probabilistic-deep learning temporal modeling. Experimental results demonstrate the effectiveness and stability of the proposed approach for long-term land cover forecasting.