A Landscape Metrics-Based Sample Weighting Approach for Forecasting Land Cover Change with Deep Learning Models

Unaddressed imbalance of multitemporal land cover (LC) data reduces deep learning (DL) model usefulness to forecast changes. To manage geospatial imbalance, there is a lack specialized cost-sensitive strategies available. Sample weights are typically derived from training instance frequencies, which disregard spatial pattern complexities. Therefore, this study proposes sample weighting approach underpinned by class-level landscape metrics (LSMs) assign importance categories based on relative indicators form. A case demonstrates the application and effects LSM-based for projecting LC changes region in British Columbia, Canada. Four spatiotemporal DL models provided weighted samples including explanatory factors. calculated patch density, shape irregularity, heterogeneity improved figure merit related measures over baseline configurations. This contributes change management models.