Ventura et al. (2026) Optimizing canopy cover evaluation: A machine learning approach using LiDAR data

Identification

• Journal: Environmental Modelling & Software
• Year: 2026
• Date: 2026-04-01
• Authors: Pau Bosch Ventura, Carles Carrillo, Alejandro Donaire, Eric Sánchez
• DOI: 10.1016/j.envsoft.2026.106982

Research Groups

Computer Architecture & Operating Systems Department, Universitat Autònoma de Barcelona, Barcelona, Spain

Short Summary

This study develops AI-CanopyMapper, a machine learning framework leveraging LiDAR data for efficient and accurate prediction of canopy cover, achieving a mean absolute error of 6.47% and an R² of 0.88 for the full model in Catalonia. The framework demonstrates strong generalization capabilities and computational efficiency, even with limited data, offering a fast and scalable alternative to traditional methods.

Objective

• To develop a fast, scalable, and accurate machine learning framework (AI-CanopyMapper) that uses LiDAR data to predict canopy cover, addressing the limitations of traditional methods in balancing speed and accuracy.

Study Configuration

• Spatial Scale: Catalonia, Spain. LiDAR data organized in 2 km × 2 km blocks with a minimum density of 0.5 points/m². Ground truth and derived maps (Slope, Aspect) at 20 m × 20 m resolution. LandUse data at 10 m × 10 m resolution.
• Temporal Scale: LiDAR and Canopy Cover ground truth data collected between 2016 and 2017. LandUse data collected in 2017. Slope and Aspect data collected in 2020.

Methodology and Data

• Models used: XGBoost (gradient-boosting algorithm) for canopy cover prediction. Mean imputer for handling missing values. Standard scaler for feature normalization.
• Data sources:
  • LiDAR data: Institut Cartogràfic i Geològic de Catalunya (ICGC).
  • Ground truth (Canopy Cover map): Institut Cartogràfic i Geològic de Catalunya (ICGC).
  • LandUse data: Impact Observatory, Microsoft, and Esri.
  • Slope and Aspect data: Derived from Digital Terrain Model (ICGC).

Main Results

• The full AI-CanopyMapper model achieved a Mean Absolute Error (MAE) of 6.47% and an R² of 0.88 on the testing set.
• A partial model, trained with only 1.3% of the available data, achieved an MAE of approximately 15%, demonstrating strong generalization capabilities under data-limited conditions.
• The optimized and parallelized methodology processed all 8424 blocks of Catalonia in 4 hours and 30 minutes, a significant reduction from an estimated 41 to 53 days for sequential processing.
• Feature selection identified 20 optimal features, with height and number of returns being the most important.
• Over 50% of predictions had an error within ±2%. The model showed a slight tendency to overestimate for smaller errors and underestimate for larger errors.
• AI-CanopyMapper's performance (R² of 0.88 with 0.5 points/m² LiDAR density) is notably strong compared to existing approaches in the literature, which typically report R² values between 0.60 and 0.70, often using denser LiDAR data or combined sources.

Contributions

• Introduction of AI-CanopyMapper, a novel, fast, scalable, and accurate machine learning framework for canopy cover estimation using LiDAR data.
• Optimized feature engineering, selection, and parallel processing techniques significantly enhance computational efficiency and reduce processing time for large datasets.
• Demonstrated state-of-the-art accuracy (R² 0.88, MAE 6.47%) using a relatively low LiDAR point cloud density (0.5 points/m²), outperforming many existing methods that require denser data or additional data sources.
• Validation of a "partial model" approach, showing that reliable canopy cover predictions can be obtained with significantly less training data (1.3% of total), enabling rapid preliminary assessments and efficient resource allocation.
• The framework is designed to be density-agnostic and transferable, allowing for adaptation and retraining in diverse ecosystems.
• Addresses the critical challenge of balancing speed and accuracy in canopy cover mapping, providing updated values essential for applications such as forest management and wildfire risk assessment.

Funding

• Ministerio de Ciencia e Innovación MCIN AEI/10.13039/501100011033 under contract PID2020-113614RB-C21 and PID2023-146193OB-I00.
• CPP2021-008762 by the European Union-NextGenerationEU/PRTR.
• Catalan Government under grant 2021-SGR-574.

Citation

@article{Ventura2026Optimizing,
  author = {Ventura, Pau Bosch and Carrillo, Carles and Donaire, Alejandro and Sánchez, Eric},
  title = {Optimizing canopy cover evaluation: A machine learning approach using LiDAR data},
  journal = {Environmental Modelling & Software},
  year = {2026},
  doi = {10.1016/j.envsoft.2026.106982},
  url = {https://doi.org/10.1016/j.envsoft.2026.106982}
}