Effective short‐term forecasting strategies to improve LULC projections in threatened ecosystems

Abstract

Recent advancements in remote sensing imagery classification have greatly improved monitoring of land use/land cover (LULC) dynamics, deepening our understanding of their effects on ecosystems and terrestrial nutrient cycling. Forecasting LULC change remains challenging because it is strongly influenced by socioeconomic drivers and biogeochemical processes linked to land management and climate change. To address this complexity, a wide range of models has been developed, from process‐based to statistical approaches. Yet, comparisons at regional and global scales reveal large discrepancies, underscoring the need for more consistent calibration and validation with historical observations. Here, we leverage the increasing availability of annual LULC maps to evaluate the temporal performance of two independent data‐driven approaches: ARIMA time‐series forecasting and a deterministic Lotka–Volterra ecological‐inspired model, across the Río de la Plata Grasslands, a threatened South American ecosystem. Both methods outperformed memoryless Markov chain models in capturing annual LULC transitions without requiring time‐consuming processing spatial inputs. These results demonstrate that incorporating long‐term annual LULC histories can substantially improve predictive skill and provide a robust framework for model intercomparison, with clear implications for linking land‐cover change to ecosystem and Earth system modeling.