Mapping Canopy Heights in Dense Tropical Forests Using Low-Cost UAV-Derived Photogrammetric Point Clouds and Machine Learning Approaches

Tropical forests are a key component of the global carbon cycle and climate change mitigation. Field- or LiDAR-based approaches enable reliable measurements structure above-ground biomass (AGB) tropical forests. Data derived from digital aerial photogrammetry (DAP) on unmanned vehicle (UAV) platform offer several advantages over field- in terms scale efficiency, DAP has been presented as viable economical alternative boreal deciduous However, detecting with ground dense forests, which is required for estimation canopy height, currently considered highly challenging. To address this issue, we present generally applicable method that based machine learning methods to identify forest floor DAP-derived point clouds We capitalize high-resolution vertical inform detection. conducted UAV-DAP surveys combined field inventories Congo Basin. Using airborne LiDAR (ALS) truthing, height model (CHM) generation workflow constitutes detection, classification interpolation points using combination local minima filters, supervised algorithms TIN densification classifying spectral geometrical features UAV-based 3D data. demonstrate our DAP-based provides estimates tree heights identical (conservatively estimated NSE = 0.88, RMSE 1.6 m). An external validation shows capable providing accurate precise AGB (DAP vs. old forest: r2 0.913, 31.93 Mg ha?1). Overall, study demonstrates application cheap easily deployable platforms can be deployed without expert knowledge generate biophysical information advance monitoring