Optimizing Forest Aboveground Biomass Models with Multi-Parameter Integration.

Area of Science:

• Forest Ecology
• Remote Sensing
• Biogeochemistry

Background:

• Forests are vital terrestrial carbon sinks, essential for climate change mitigation via carbon sequestration.
• Accurate estimation of aboveground biomass (AGB) is critical for carbon budget assessments and ecosystem modeling.
• Mountainous regions with diverse forest types present unique challenges for biomass estimation.

Purpose of the Study:

• To develop and evaluate models for estimating aboveground biomass (AGB) in Wolong Nature Reserve, China.
• To compare the performance of univariate, multivariate regression, and machine learning models for AGB estimation.
• To identify the most accurate predictors and methods for large-scale AGB assessment.

Main Methods:

• Construction of univariate models relating canopy height to field-measured AGB.
• Development of multivariate regression models incorporating vegetation indices (VIs), leaf area index (LAI), and topographic variables.
• Implementation and optimization of a decision tree-based machine learning framework using a combined predictor set.

Main Results:

• Univariate and conventional multivariate regression models showed a tendency to overestimate AGB.
• These overestimations limit the applicability of traditional models for large-scale biomass assessments.
• The optimized decision tree model demonstrated superior predictive accuracy after parameter tuning and cross-validation.

Conclusions:

• Machine learning, specifically decision tree models, offers enhanced accuracy for aboveground biomass estimation compared to traditional methods.
• Integrating multiple data sources including canopy height, LAI, VIs, and topography improves AGB prediction.
• Accurate AGB estimation using optimized models is vital for effective forest carbon management and climate change research.