Estimation of Dendrocalamus giganteus leaf area index by combining multi-source remote sensing data and machine...

Area of Science:

• Ecology and Remote Sensing
• Forestry and Environmental Science

Background:

• Leaf Area Index (LAI) is crucial for canopy energy and material exchange.
• Accurate, large-scale LAI estimation is vital for environmental monitoring.
• Existing machine learning models for LAI estimation have limitations.

Purpose of the Study:

• To develop a novel, accurate method for large-scale estimation of Dendrocalamus giganteus LAI.
• To integrate ICESat-2/ATLAS and Sentinel-1/-2 data for enhanced LAI estimation.
• To refine machine learning models using optimization algorithms.

Main Methods:

• Spatial interpolation using Sequential Gaussian Conditional Simulation (SGCS).
• Feature variable optimization via Pearson correlation coefficient with multi-source remote sensing data.
• Application of Bayesian Optimization (BO), Particle Swarm Optimization (PSO), Genetic Algorithms (GA), and Simulated Annealing (SA) to Random Forest Regression (RFR), Gradient Boosting Regression Tree (GBRT), and Support Vector Machine Regression (SVR) models.

Main Results:

• The Bayesian Optimization-Gradient Boosting Regression Tree (BO-GBRT) model demonstrated superior performance.
• Achieved high accuracy with R² of 0.922, RMSE of 0.263, MAE of 0.187, and P₁ of 92.38%.
• The proposed method outperformed existing machine learning approaches.

Conclusions:

• The BO-GBRT model offers a highly accurate and efficient solution for large-scale forest LAI estimation.
• This integrated remote sensing and machine learning approach shows significant potential for forest LAI inversion.
• The methodology can be extended to estimate other forest structure parameters.