A Quantitative Assessment of Forest Structure and Biomass Loss in Bakossi National Park using Remote Sensing and Machine Learning
Kato Samuel Namuene
*
Department of Forestry and Wildlife, University of Buea, PMB 63, Buea, Cameroon.
Ambo Beatrice Fonge
Department of Plant Science, University of Buea, PMB 63, Buea, Cameroon.
Mojoko Fiona Mbella
Department of Geography, University of Buea, PMB 63, Buea, Cameroon.
Colins Mesue Kome
Department of Forestry and Wildlife, University of Buea, PMB 63, Buea, Cameroon.
Mbua Abang Augustine
Divisional Delegation of Forestry and Wildlife, Kupe Muanenguba Division, Bangem, South West Region, Cameroon.
Tinyu Cyprian Mondoh
Regional Delegation of Forestry and Wildlife West, Baffoussam, Cameroon.
*Author to whom correspondence should be addressed.
Abstract
Tropical montane forests of West and Central Africa face escalating anthropogenic pressure, and Bakossi National Park (BNP), situated in the Cameroon Highlands at the biogeographic confluence of the Upper Guinea and Congo Basin forest blocks is not an exception. This study presents a systematic multi-temporal assessment of forest cover change and above-ground biomass (AGB) dynamics in BNP from 2000 to 2023. It employs a synergistic JavaScript multi-source remote sensing framework within Google Earth Engine integrated with supervised machine learning algorithms calibrated exclusively from spaceborne LiDAR observations without field inventory. Time-series composites from Landsat 5 TM, Landsat 8 OLI/TIRS, Sentinel-2 MSI, and ALOS PALSAR-2 were classified across six reference epochs using a Random Forest Classifier, achieving overall accuracy of 96.4% (κ = 0.951). AGB models were calibrated against 6,842 GEDI L4A footprints and 4,287 ICESat-2 ATL08 canopy height observations, with Random Forest Regression yielding the highest validation performance (R² = 0.893, RMSE = 18.72 Mg ha⁻¹). Incorporating L-band SAR backscatter improved AGB accuracy by R² = 0.112 over optical-only models, confirming the indispensability of microwave data above the optical saturation threshold in dense closed-canopy forests. Between 2000 and 2023, BNP sustained net forest cover loss of 5,374 ha (18.3% of gazetted area), with annual deforestation accelerating from 412 ha yr⁻¹ to 1,021 ha yr⁻¹, a 148% increase, and total AGB declined from 6.22 Tg to 3.50 Tg, generating cumulative emissions of approximately 11.4 Tg CO₂e. Deforestation intensity correlated most strongly with road proximity (ρ = -0.74) and settlement proximity (ρ = -0.68), with three discrete hotspots identified through kernel density estimation. ICESat-2 ATL08 independently confirmed a mean canopy height reduction of 6.3 m in degraded zones between 2018 and 2023. These results carry direct implications for REDD+ MRV frameworks, Cameroon's nationally determined carbon contributions, and the global 2030 biodiversity protection target.
Keywords: Above-ground biomass, Bakossi National Park, deforestation, forest degradation, GEDI, google earth engine, ICESat-2, PALSAR, random forest, REDD+, tropical montane forest.