Integrated multi-satellite data and machine learning approach in mapping the successional stages of forest types in a tropical montane forest

Abstract

Understanding the successional stages in tropical montane forests (TMF) is crucial for its conservation and management. This study integrated Sentinel-1, Sentinel-2, InSAR, GEDI, and machine learning to map the categorical successional stages of different forest types in a Philippines’ TMF. Field data collected from December 2022 to January 2023 were used to create and validate successional stages models. Sentinel-1 interferogram, unwrapped interferogram, and coherence exhibited moderate positive correlations with canopy height (r = 0.43). Incorporating GEDI with InSAR to predict canopy height yielded less accurate predictions (r = −0.20 to 0.04; RMSE = 12–13 m). Results show that canopy height, a widely accepted attribute for forest structure, appears secondary to other biophysical variables. Integrating optical, radar, and auxiliary variables achieved an overall accuracy of 79.56% and a kappa value of 75.74%. Feature importance analysis using Random Forest enhanced the overall accuracy (84.22%) and kappa value (81.19%). The integration of multi-satellite data with machine learning has proven effective for studying TMFs successional stages. Elevation emerged as the most significant predictor of forest type distribution, with mature and young pine forests dominating lower elevation (700–1,400m) and mossy forests dominating above 1,400m. Given the observed disturbances, the study underscores the need for robust conservation strategies and sustainable TMF management. Future research should focus on time-series analyses of successional stages, further optimization of machine learning models, and integrating additional data sources, such as LiDAR, to enhance canopy height predictions and forest monitoring efforts. The findings also provide valuable knowledge applicable to TMFs globally, supporting informed conservation and policies intended to protect biodiversity.