Revealing forest structural "fingerprints": An integration of LiDAR and deep learning uncovers topographical influences on Central Amazon forests

• DOI: 10.1016/j.ecoinf.2024.102628
• Publication Date: 2024-05-05T14:00:32Z
• Authors: Nathan Borges Gon...; Diogo Martins Rosa; Dalton Freitas do...; Marielle N. Smith; Ricardo Dalagnol; Danilo Roberti Al...; Bruce W. Nelson; Scott C. Stark

Amazon forests are characterized by rich structural diversity. However, the influence of factors such as topography, soil attributes, and external disturbances on variability is not always well characterized, traditional metrics may be inadequate to capture this type complexity. While LiDAR offers expanded insights, parameters used in analysis, mean or maximum canopy height, directly linked environmental variables like topography. Emerging approaches merge with machine learning uncover deeper complexities. work date fail fully utilize potential fine-scale information. Here we introduce a novel approach, leveraging 2D point cloud images derived from profiling (PCL). The technique targets intricate details within clouds using deep algorithms. With dataset Central comprising 18 multitemporal transects 450 m length, our objective was detect "fingerprints" varied topographical types along hillslope, comprising: Riparian, White-sand, Plateau, any gradient shifts based terrain variations here represented height above nearest drainage (HAND). trained tested leave-one-group-out approach (LOGO) which, for each iteration, complete transect excluded training after iteration. fast.ai platform ResNet-34 architecture, coupled transfer learning, were perform classification distinguish between three types. Furthermore, hybrid model combining Convolutional Autoencoder, Partial Least Square (PLS) regression designed forest correlations HAND variation. Cross-validation achieved promising high weighted F1 score 0.83 classify Additionally, combined Autoencoder PLS revealed strong correlation (R2 = 0.76) actual predicted HAND. Innovatively ground-based PCL LiDAR, study Forest structures connected topographic Our findings underscore transformative integrative investigating dynamics promise powerful new tool understanding climate-related structure change.