Open AccessA machine learning approach to model leaf area index in Eucalyptus plantations using high-resolution satellite imagery and airborne laser scanner data
Author(s) -
Andrés Hirigoyen,
Cristina Acosta,
Antonio Ariza,
Maria Angeles Vero-Martinez,
Cecilia Rachid,
Jorge Franco,
Rafael Navara-Cerrillo
Publication year - 2022
Publication title -
annals of forest research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.466
H-Index - 18
eISSN - 2065-2445
pISSN - 1844-8135
DOI - 10.15287/afr.2021.2073
Subject(s) - remote sensing , leaf area index , lidar , multispectral image , environmental science , canopy , random forest , satellite , mean squared error , coefficient of determination , support vector machine , normalized difference vegetation index , vegetation (pathology) , computer science , mathematics , machine learning , geography , statistics , medicine , ecology , archaeology , pathology , aerospace engineering , engineering , biology
As a forest structural parameter, leaf area index (LAI) is crucial for efficient intensive plantation management. Leaf area is responsible for the energy absorption needed for photosynthetic production and transpiration, both affecting growth. Currently, LAI can be estimated either by remote-sensing methods or ground-based methods. However, unlike ground-based methods, remote estimation provides a cost-effective and ecologically significant advance The aim of our study was to evaluate whether machine learning algorithms can be used to quantify LAI, using either optical remote sensing or LiDAR metrics.in Eucalyptus dunnii and Eucalyptus grandis stands First, empirical relationships between LAI and remote-sensing data using LiDAR metrics and multispectral high-resolution satellite metrics, were assessed. Selected variables for LAI estimation were: LiDAR forest canopy cover, laser penetration index, and canopy relief ratio - from among the LiDAR data and the green normalized difference vegetation index and normalized difference vegetation index - from among the ground-based data we compared the accuracy of three machine learning algorithms: artificial neural networks (ANN), random forest (RF) and support vector regression (SVR). The coefficient of determination ranged from 0.60, for ANN, to 0.84, for SVR. The SVR regression methods showed the best performance in terms of overall model accuracy and RMSE (0.60). The results show that the remote sensing data applied through machine learning algorithms provide an effective method to estimate LAI in eucalyptus plantations. The methodology proposed is directly applicable for operational forest planning at the landscape level.