Bridging Nadir Observations and Three-Dimensional Tree Structure through Species-dependent Generative Modeling
Authors:
- Raghav SharmaRSRaghav Sharma
- Kongwen (Frank) ZhangKongwen (Frank) Zhang
- Jane LiuJLJane Liu
- Baoxin HuBHBaoxin Hu
Preprint
- 14 March 2026
- preprint
- Published by Elsevier in SSRN Electronic Journal
Abstract
Reconstructing three-dimensional (3D), species-dependent tree structure from widely available two-dimensional (2D) nadir-view remote sensing data represents a critical yet unresolved challenge in forest monitoring and ecological analysis. While decades of airborne and satellite imagery provide rich spatial and temporal coverage, their inherent top-down perspective limits direct access to lateral crown structure, constraining the characterization of individual trees and canopy dynamics. In this study, we propose a novel 2D-to-3D inference framework that synthesizes realistic lateral (fac¸ade) views of individual tree crowns from single nadir-view LiDAR data as proxy, establishing an important intermediate step toward optical 3D forest modeling. Using a conditional diffusion framework, we demonstrate that integrating ancillary biological attributes, specifically tree species and height, resolves the inherent ambiguity in mapping nadir observations to lateral structure. The model learns species-dependent structural priors that enable the generation of physically plausible fac¸ade views, effectively transmuting 2D observations to infer missing 3D characteristics. Our quantitative evaluation demonstrates high pixel-wise fidelity between generated and measured ground-truth LiDAR data, achieving a Mean Absolute Error (MAE) of 0.077. Currently, nadir data alone are reported as insufficient to recover meaningful fac¸ade geometry, whereas conditioning on species identity and tree height enables robust reconstruction of characteristic crown shapes and vertical profiles. This study highlights the role of attribute-conditioned generative modeling as a scalable mechanism for transforming 2D remote sensing archives into enriched 3D structural representations. By bridging the gap between nadir imagery and species-dependent 3D tree models, the proposed approach lays the groundwork for future reconstruction of individual tree structure from historical 2D imagery, supporting long-term analyses of tree growth, canopy evolution, and ecosystem dynamics at landscape scales.Keywords
This publication has 12 references indexed in Scilit:
- Individual tree species prediction using airborne laser scanning data and derived point-cloud metrics within a dual-stream deep learning approachInternational Journal of Applied Earth Observation and Geoinformation, 2025
- Benchmarking tree species classification from proximally sensed laser scanning data: Introducing the FOR‐species20K datasetMethods in Ecology and Evolution, 2025
- A Review of Individual Tree Crown Detection and Delineation From Optical Remote Sensing Images: Current progress and futureIEEE Geoscience and Remote Sensing Magazine, 2024
- SegmentAnyTree: A sensor and platform agnostic deep learning model for tree segmentation using laser scanning dataRemote Sensing of Environment, 2024
- Aboveground biomass density models for NASA’s Global Ecosystem Dynamics Investigation (GEDI) lidar missionRemote Sensing of Environment, 2022
- Terrestrial LiDAR: a three‐dimensional revolution in how we look at treesNew Phytologist, 2018
- Quantifying stand structural complexity and its relationship with forest management, tree species diversity and microclimateAgricultural and Forest Meteorology, 2017
- Allometric equations for integrating remote sensing imagery into forest monitoring programmesGlobal Change Biology, 2016
- Nondestructive estimates of above‐ground biomass using terrestrial laser scanningMethods in Ecology and Evolution, 2014
- Image quality assessment: from error visibility to structural similarityIEEE Transactions on Image Processing, 2004