FLORA: A deep learning approach to predict forest attributes from heterogeneous LiDAR data

Emilie Vautier, Clément Mallet, Cédric Vega

Forest attributes are essential for national-scale resource monitoring. Airborne LiDAR metrics are among the auxiliary variables most strongly correlated with forest attributes used in National Forest Inventory (NFI) estimates. However, producing wall-to-wall predictions remains challenging when LiDAR data are acquired under heterogeneous conditions. As national LiDAR programs expand across Europe, variability in sensors, flight parameters, seasons, and scan angles limits the robustness of existing models, which are often calibrated for local conditions. We present FLORA (Forest LiDAR Octree Regression with Auxiliary Data), a deep learning framework that predicts six forest attributes: dominant height, total volume, deciduous volume, coniferous volume, basal area, and stem density from heterogeneous LiDAR point clouds. FLORA combines an octree-based backbone with ecological and spatiotemporal auxiliary variables through a late-fusion gating mechanism. Models are trained and evaluated on 32,052 National Forest Inventory plots across mainland France using data from the French LiDAR HD program. A single model trained on both leaf-on and leaf-off acquisitions outperforms season-specific models and improves cross-season robustness. Auxiliary variables provide modest overall gains but contribute more strongly to species-specific volume prediction. FLORA achieves an rRMSE of about 12.3% (R2 = 0.88) for dominant height and 39% (R2 = 0.74) for total volume, providing a robust baseline for large-scale forest attribute estimation from heterogeneous national LiDAR programs.

Open Source

Research Brief

FLORA is a deep learning framework that accurately predicts six key forest attributes from diverse LiDAR data, improving robustness across seasonal variations for national forest monitoring.

Researchers have developed a new artificial intelligence system called FLORA to more accurately map and understand forests across entire countries. Current methods struggle because the laser-scanning data (LiDAR) used to measure forests can vary widely depending on the equipment, weather, or time of year it was collected. FLORA uses a special AI architecture, an octree-based backbone, that can process these diverse LiDAR datasets, even combining them with ecological and location-based information, to predict crucial forest details like tree height, total wood volume (including deciduous and coniferous types), basal area, and stem density. Tested on a massive dataset from over 32,000 plots in French forests, FLORA proved robust even when mixing data from different seasons (leaves on vs. leaves off trees), often outperforming season-specific models. The system offers a more reliable and robust way to monitor forest resources on a national scale, achieving high accuracy for dominant height (rRMSE ~12.3%, R2 = 0.88) and total volume (rRMSE ~39%, R2 = 0.74).

Potential Applications

National Forest Inventory (NFI) updates and enhanced national-scale resource monitoring.
Precise monitoring of deforestation, afforestation, and reforestation efforts for environmental policy.
Supporting sustainable forestry management, timber yield estimation, and biodiversity assessment.
Quantifying forest carbon sequestration for climate change mitigation and carbon credit markets.

25/100

Paper Trustworthiness Index

High Skepticism

High Skepticism / Self-Published

This document should be treated with critical skepticism. It contains unverified scientific claims or was self-published.

Verified AI Assessment: This credibility analysis was generated by Gemini 2.5 Flash analyzing the full paper text, references, and metadata.

Core Pillars Breakdown

Author & Institutional Track Record

0 / 25

The abstract does not provide any information about the authors, their affiliations, or their previous work. Therefore, no assessment of author track record can be made from the provided text.

Technical Rigor & Methodology

25 / 30

The abstract describes a specific deep learning architecture ('octree-based backbone' with 'late-fusion gating mechanism'), mentions training and evaluation on a large and heterogeneous dataset ('32,052 National Forest Inventory plots' from 'French LiDAR HD program'), and provides quantitative performance metrics (rRMSE, R2). It also highlights comparative performance against 'season-specific models', suggesting a robust methodological approach.

Reproducibility & Openness

0 / 25

The abstract does not provide any information regarding the public availability of code, data, trained models, or specific URLs. Without this information, independent reproduction cannot be assessed.

Community Vetting & Peer Review

0 / 20

The abstract does not mention if the paper has been peer-reviewed, accepted into a conference, published in a journal, or is currently a preprint. Therefore, its community vetting status cannot be determined from the provided text.

Detailed Evidence Assessment

Verified Evidence & Citations

FLORA predicts six specific forest attributes.

“We present FLORA (Forest LiDAR Octree Regression with Auxiliary Data), a deep learning framework that predicts six forest attributes: dominant height, total volume, deciduous volume, coniferous volume, basal area, and stem density...”

FLORA utilizes an octree-based backbone and auxiliary variables.

“FLORA combines an octree-based backbone with ecological and spatiotemporal auxiliary variables through a late-fusion gating mechanism.”

Models were trained and evaluated on a large French dataset.

“Models are trained and evaluated on 32,052 National Forest Inventory plots across mainland France using data from the French LiDAR HD program.”

A single model trained on both leaf-on and leaf-off acquisitions outperforms season-specific models.

“A single model trained on both leaf-on and leaf-off acquisitions outperforms season-specific models and improves cross-season robustness.”

Auxiliary variables provide modest overall gains but contribute more strongly to species-specific volume prediction.

“Auxiliary variables provide modest overall gains but contribute more strongly to species-specific volume prediction.”

FLORA achieves specific rRMSE and R2 for dominant height and total volume.

“FLORA achieves an rRMSE of about 12.3% (R2 = 0.88) for dominant height and 39% (R2 = 0.74) for total volume...”

Uncertainties & Omissions

• Omission:No codebase repository link provided.

• Omission:Specific deep learning architecture details beyond 'octree-based backbone' and 'late-fusion gating mechanism' are not detailed.

• Omission:Detailed methodology for handling 'heterogeneous LiDAR data' beyond combining acquisitions is not provided.

• Omission:Comparison with specific existing baseline models beyond 'season-specific models' is not explicitly mentioned.

• Omission:No mention of computational resources or training time.

• Uncertainty:The precise nature and definition of 'heterogeneous conditions' beyond sensors, flight parameters, seasons, and scan angles are not fully elaborated.

• Uncertainty:The specific types of 'ecological and spatiotemporal auxiliary variables' used are not detailed.

• Uncertainty:The generalizability of FLORA beyond mainland France and the French LiDAR HD program is not discussed.

• Uncertainty:The exact impact of the 'modest overall gains' from auxiliary variables on the overall R2/rRMSE for all attributes is not quantified, beyond species-specific volume prediction.