Quantifying and simulating above-ground biomass and carbon sequestration under various land cover scenarios in planted forest using machine learning and Earth observation
摘要
Quantifying and projecting above-ground biomass (AGB) and carbon dynamics are critical for understanding forest condition and supporting climate mitigation strategies, particularly in data-limited tropical systems. This study presents the first long-term assessment of land use/land cover (LULC) change, AGB, and carbon stocks in Kamatira Forest, West Pokot County, covering the period 1990–2035. Multi-temporal Landsat imagery for (1990–2024) was processed in Google Earth Engine and classified using machine-learning approaches, with accuracy assessed through confusion-matrix metrics. The 2024 LULC classification achieved an overall accuracy of 88.16% and a Kappa coefficient of 0.82, indicating strong agreement with field reference data. Above-ground biomass and carbon stocks were estimated using the pan-tropical allometric equation of (Chave et al., 20(10):3177-3190, 2014). Mean AGB ranged from 948 to 1040 Mg ha−1 between 1990 and 2024, with dense forest consistently contributing over 70% of total biomass and carbon stocks, despite a marked decline in its areal extent. Future LULC was simulated to 2035 using an ensemble of machine-learning models, which achieved an overall accuracy of 0.87 and strong class-wise performance (precision and recall > 0.84 for dense and open forest classes). Projections indicate continued redistribution of biomass from dense forest to moderately dense and open forest classes, highlighting the dominant role of LULC transitions in shaping future carbon dynamics. The integrated methodological framework provides a robust approach for assessing and forecasting forest carbon dynamics, particularly in areas with limited inventory data, thereby supporting carbon accounting and the sustainable management of planted forests.