Assessing Forest Structure and Biomass Loss in Mount Cameroon National Park Using Remote Sensing and Machine Learning

Abstract

Tropical montane forests of Cameroon represent disproportionately important per-unit-area carbon stocks and biodiversity repositories, yet remain among the least monitored protected landscapes in West and Central Africa. This study presents a comprehensive, multi-temporal remote sensing and machine learning assessment of forest structural dynamics and above-ground biomass (AGB) loss across the full extent of Mount Cameroon National Park (MCNP) between 2000 and 2023. Multi-source satellite data comprising Landsat 5/8, Sentinel-2, ALOS PALSAR-2 L-band SAR, and ICESat-2 ATL08 spaceborne LiDAR were integrated within Google Earth Engine. A Random Forest classifier achieved an overall land cover classification accuracy of 95.8% (Kappa = 0.942). AGB estimation models were calibrated against 6,842 GEDI L4A spaceborne retrievals and 4,614 ICESat-2 ATL08 segments across five altitudinal vegetation zones, yielding R² = 0.884 and RMSE = 20.31 Mg ha⁻¹, with multi-sensor fusion delivering an R² improvement of 0.150 over optical-only models. SHAP analysis identified PALSAR-2 HV backscatter as the dominant predictor (mean |SHAP| = 0.221), followed by ICESat-2 canopy height (0.179) and EVI (0.158). Dense closed-canopy forest declined from 37,309 ha (64.1%) in 2000 to 26,955 ha (46.3%) in 2023, representing a 176% increase in annual deforestation rate, from 358 ha yr⁻¹ to 987 ha yr⁻¹. Total AGB declined from 10.31 Tg to 7.04 Tg, generating cumulative carbon emissions of 5.04 Tg CO₂e. ICESat-2 ATL08 confirmed a mean canopy height reduction of 6.3 m in degraded forest zones between 2019 and 2023, with structural divergence between intact and degraded forest increasing from 8.4 m to 13.4 m. Road proximity was the dominant deforestation driver (Spearman ρ = -0.76), with the agricultural frontier migrating upslope by 350 m over the study period. These findings provide spatially explicit quantitative evidence supporting REDD+ Tier 2 carbon accounting and adaptive management in this critical Afromontane protected landscape.