Deep learning U-Net classification of Sentinel-1 and 2 fusions effectively demarcates tropical montane forest's deforestation

Deep learning U-Net classification of Sentinel-1 and 2 fusions effectively demarcates tropical montane forest's deforestation

Tropical montane forests (TMF) play a vital role in providing numerous ecosystem services. This ecosystem is characterized by towering mountains, cold weather, tall trees such as pine, and dwarfed plants. Although the combination of optical and radar imagery has shown promise in land use and land cover (LULC) mapping, only a handful of studies have attempted to study the dynamics of land changes in tropical montasne forests by combining the two. In this study, we examined the usage of Sentinel-1 (S-1) and Sentinel-2 (S-2) and their fusion as input for LULC mapping with three modeling classifiers: traditional (maximum likelihood classification - MLC), machine learning techniques (random forest, k-nearest neighbor, and kd-tree nearest neighbor), and a deep learning approach (U-Net) in a TMF in the Philippines. Also, the deforestation was characterized in terms of proximity and topographic factors. The findings revealed that the combination of S-1 and S-2 provides LULC with high accuracy of image classification. In binary classification, traditional MLC supersedes other classifiers in correctly classifying the pixels of the input imagery (average overall accuracy (OA) = 95.22; Kappa index (KI) = 90.39). Random Forest (RF) stands out in machine learning classifiers (average OA = 94.49; KI = 88.8). However, U-Net deep learning loses in binary classification but proven more robust when LULC were classified into six complex classes with an OA of 86.77 and KI of 78.89. In addition, using deep learning modeling for LULC mapping of the research site, it was determined that from 2015 to early 2022, 417.93 km2 of the study area had been deforested. Further, this research reveals that the greater the proximity of a forest to a human settlement or agricultural zone, the greater the likelihood that it will be cleared for human habitation or agriculture. Deforestation also occurred in rural locations, far from roads and bodies of water. This analysis also supports the hypothesis that deforestation can occur even in high-elevation areas. The results of this study can be utilized by policy and law makers in order to better communicate the critical nature of implementing evidence- and science-based policies to strengthen protections, as well as developing conservation and management plan for tropical montane forests.

Files associated with this item cannot be displayed due to copyright restrictions.

This article is part of a UniSQ Thesis by publication. See Related Output.