Machine learning-based assessment of deforestation, successional stages, and carbon stocks in a tropical montane forest using radar and optical satellite imagery

Machine learning-based assessment of deforestation, successional stages, and carbon stocks in a tropical montane forest using radar and optical satellite imagery

Tropical montane forests (TMFs) play a crucial role in providing essential ecosystem services. However, these environments are increasingly threatened by deforestation and degradation. In the Philippines’ TMFs, the extent of deforestation, successional stages, and carbon stocks remain poorly explored. Hence, this study focuses on the Province of Benguet, Philippines, with three specific objectives: 1) to demarcate deforestation using the fusion of Sentinel-1,-2, and biophysical data through a traditional classifier, machine (ML) and deep learning (DL) algorithms; 2) to map the successional stages in different vegetation types through Interferometric Synthetic Aperture Radar (InSAR), Global Ecosystem Dynamics Investigation (GEDI), Sentinel products and biophysical data with ML; and, 3) to estimate the above-ground biomass (AGB) and above-ground carbon (AGC) through optical, radar, biophysical data and ML. In addition to the field assessments conducted from December 2023 to January 2024, a systematic review of spaceborne remote sensing applications in global TMFs reinforced the significance of this study. The following results are the highlights of this study: 1) generally, RS investigations on TMFs are concentrated in the Americas (62%), with optical sensors (85.76%) being used more frequently than SAR (12.70%); 2) the fusion of Sentinel-1-2 and biophysical data with U-Net DL algorithm effectively demarcated the deforestation (Overall Accuracy (OA) = 86.77%, Kappa Index (KI) = 78.89); 3) elevation emerged as a significant predictor of vegetation type distribution, with Random Forest’s (RF) top 10 features yielding the best predictive performance (OA = 84.22%, KI = 81.19%); and, 4) among the various algorithms utilized for AGB assessment, RF demonstrated the highest accuracy (r = 0.982; RMSE = 53.980 Mgha -1). Above-ground carbon density varied from 0 to 434.94 Mgha-1. This study underscores the urgency of formulating conservation and sustainable management policies. It also emphasizes the significance of Benguet’s TMF in the context of carbon sequestration initiatives like REDD+.

File reproduced in accordance with the copyright policy of the publisher/author/creator.