1- guilan university
2- tarbiat modares university
2- tarbiat modares university
Abstract: (167 Views)
Introduction and Objectives: The Zagros forests play a vital role in climate regulation, biodiversity conservation, and carbon sequestration. The Leaf Area Index (LAI), as one of the most crucial ecological indicators, reflects the productive capacity of forest ecosystems and is essential for modeling photosynthesis, carbon cycle, and evapotranspiration processes. Given the challenges of direct LAI measurement, employing non-destructive machine learning-based methods for LAI estimation becomes imperative. This study aimed to model LAI based on tree structural characteristics (including diameter at breast height [DBH], tree height, and canopy cover percentage) using machine learning algorithms in the northern Zagros forests (Marivan region, Kurdistan Province). The research seeks to provide an accurate and cost-effective solution for monitoring ecosystem changes and sustainable forest management.
Materials and Methods: The study was conducted in the northern Zagros forests (Marivan County) characterized by cold and semi-humid climate. Field data were collected from 80 systematically-randomized square plots (20×20 m). In each plot, structural characteristics including DBH, tree height, canopy cover percentage, and LAI were measured using hemispherical photography and Gap Light Analyzer (GLA) software. Aboveground biomass (AGB) was estimated using a power regression model with DBH as the independent variable. Four machine learning algorithms—Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Gaussian Process Regression (GPR)—were compared for LAI modeling. The dataset was divided into training (70%) and validation (30%) subsets, with model performance evaluated using determination coefficient (R²), root mean square error (RMSE), relative RMSE (rRMSE), and mean absolute error (MAE).
Results: LAI values in the study area ranged from 0.151 to 4.623, indicating high vegetation density variability. LAI showed very strong correlation with canopy cover percentage (r = 0.92), and moderate correlations with DBH (r = 0.70) and tree height (r = 0.78). Aboveground biomass demonstrated moderate correlation with LAI (r = 0.64). Among the machine learning algorithms, Random Forest achieved the highest accuracy (R² = 0.96 in training, R² = 0.90 in validation) and was identified as the optimal model. GPR and KNN algorithms performed similarly (R² = 0.91), followed by SVM (R² = 0.88). The correlation matrix confirmed the strong influence of structural characteristics on LAI, while Shannon-Wiener diversity index showed weak correlation with other variables.
Conclusion: This study demonstrated that the Random Forest algorithm, with its capability to model complex nonlinear relationships, serves as an effective tool for indirect LAI estimation in Zagros forests. The strong correlation between LAI and both canopy cover and structural tree characteristics enables the use of these variables as physical proxies for leaf density. The findings provide a scientific basis for sustainable management planning of Zagros forests, climate change monitoring, and environmental stress assessment. Future studies should consider integrating remote sensing data with deep learning approaches to enhance model accuracy. This research represents a significant step toward conserving the valuable Zagros ecosystems and promoting evidence-based decision making.
Materials and Methods: The study was conducted in the northern Zagros forests (Marivan County) characterized by cold and semi-humid climate. Field data were collected from 80 systematically-randomized square plots (20×20 m). In each plot, structural characteristics including DBH, tree height, canopy cover percentage, and LAI were measured using hemispherical photography and Gap Light Analyzer (GLA) software. Aboveground biomass (AGB) was estimated using a power regression model with DBH as the independent variable. Four machine learning algorithms—Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Gaussian Process Regression (GPR)—were compared for LAI modeling. The dataset was divided into training (70%) and validation (30%) subsets, with model performance evaluated using determination coefficient (R²), root mean square error (RMSE), relative RMSE (rRMSE), and mean absolute error (MAE).
Results: LAI values in the study area ranged from 0.151 to 4.623, indicating high vegetation density variability. LAI showed very strong correlation with canopy cover percentage (r = 0.92), and moderate correlations with DBH (r = 0.70) and tree height (r = 0.78). Aboveground biomass demonstrated moderate correlation with LAI (r = 0.64). Among the machine learning algorithms, Random Forest achieved the highest accuracy (R² = 0.96 in training, R² = 0.90 in validation) and was identified as the optimal model. GPR and KNN algorithms performed similarly (R² = 0.91), followed by SVM (R² = 0.88). The correlation matrix confirmed the strong influence of structural characteristics on LAI, while Shannon-Wiener diversity index showed weak correlation with other variables.
Conclusion: This study demonstrated that the Random Forest algorithm, with its capability to model complex nonlinear relationships, serves as an effective tool for indirect LAI estimation in Zagros forests. The strong correlation between LAI and both canopy cover and structural tree characteristics enables the use of these variables as physical proxies for leaf density. The findings provide a scientific basis for sustainable management planning of Zagros forests, climate change monitoring, and environmental stress assessment. Future studies should consider integrating remote sensing data with deep learning approaches to enhance model accuracy. This research represents a significant step toward conserving the valuable Zagros ecosystems and promoting evidence-based decision making.
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |