Mapping Forest Successional Stages Using Spaceborne GEDI LiDAR Canopy Height and Multi-Sensor Sentinel Integration in Bukidnon, Philippines

Mapping forest successional stages is essential for supporting forest restoration, biodiversity conservation, and sustainable land management, particularly in mountainous tropical landscapes where structural vegetation information is difficult to obtain. This study aims to develop a multi-sensor machine learning framework for estimating forest successional stages across Bukidnon Province, Philippines, a mountainous region characterized by extensive forest cover, rugged topography, and increasing agricultural encroachment. The proposed framework was implemented in Google Earth Engine by integrating Sentinel-1 synthetic aperture radar (SAR), Sentinel-2 optical imagery, SRTM topographic variables, and GEDI LiDAR observations. A Random Forest regression model was trained using multi-source predictors to estimate continuous RH98 (98th percentile relative canopy height) from GEDI footprints, which was subsequently used to classify forests into three successional stages: Early (0–5 m), Secondary (>5–15 m), and Advanced (>15 m). The model achieved a root mean square error (RMSE) of 13.30 m, indicating that uncalibrated large-footprint (25 m) GEDI LiDAR measurements are affected by terrain-induced vertical stretching in steep landscapes. A visual geodetic assessment of 299 validation points produced an overall accuracy of 48.16% and a Kappa coefficient of 0.22, with most classification errors occurring in agricultural transition zones characterized by mixed pixels. Although uncalibrated spaceborne LiDAR is insufficient for precise discrimination of individual successional stages, integrating continuous canopy height estimates with multi-sensor remote sensing data provides a computationally efficient and practical approach for coarse-scale forest succession monitoring across provincial landscapes.