# Optimizing Forest Aboveground Biomass Models with Multi-Parameter Integration

## TL;DR

This study compares different models for estimating forest aboveground biomass and finds that a decision tree machine learning model performs best.

## Contribution

The novel contribution is demonstrating that a decision tree model outperforms traditional univariate and multivariate models in AGB estimation.

## Key findings

- The decision tree model achieved an R2 of 0.8 without overestimation bias.
- Univariate and multivariate regression models showed overestimation or underestimation issues.
- Multi-parameter integration with machine learning improves AGB estimation in heterogeneous landscapes.

## Abstract

What are the main findings?
Three forest AGB estimation models (univariate function, multivariate regression, decision tree) were developed. The optimized decision tree model achieved the highest accuracy (R2 = 0.8) without overestimation bias.Traditional univariate (power function: R2 = 0.7349) and multivariate regression (best R2 = 0.517) models showed insufficient precision with overestimation or underestimation.

Three forest AGB estimation models (univariate function, multivariate regression, decision tree) were developed. The optimized decision tree model achieved the highest accuracy (R2 = 0.8) without overestimation bias.

Traditional univariate (power function: R2 = 0.7349) and multivariate regression (best R2 = 0.517) models showed insufficient precision with overestimation or underestimation.

What are the implications of the main findings?
Multi-parameter integration combined with machine learning effectively captures non-linear relationships, providing a reliable paradigm for AGB estimation in heterogeneous landscapes.Incorporating ecological parameters (e.g., LAI, CIg) enhances estimation completeness, supporting regional carbon stock assessments and forest management under carbon neutrality goals.

Multi-parameter integration combined with machine learning effectively captures non-linear relationships, providing a reliable paradigm for AGB estimation in heterogeneous landscapes.

Incorporating ecological parameters (e.g., LAI, CIg) enhances estimation completeness, supporting regional carbon stock assessments and forest management under carbon neutrality goals.

Forests constitute a fundamental component of terrestrial carbon stocks and play a pivotal role in mitigating climate change through carbon sequestration. Accurate estimation of aboveground biomass (AGB) is essential for quantifying carbon budgets and informing ecosystem models. This study takes Wolong Nature Reserve in Sichuan Province, China, a mountainous area with high vegetation coverage and diverse forest types dominated by coniferous and mixed forests, as the study area, and constructs and evaluates AGB estimation models by integrating canopy height, leaf area index (LAI), vegetation indices (VIs), and topographic variables. Initially, univariate parametric models (linear, exponential, logarithmic, power, and polynomial) were established to relate canopy height to field-measured AGB. Subsequently, multivariate regression models incorporating VIs, LAI, and topographic metrics were developed. Finally, a decision tree-based machine learning framework was implemented to exploit the combined predictor set. Comparative analysis revealed that both canopy height-based and conventional multivariate regression models tended to overestimate AGB, limiting their applicability for large-scale assessments. In contrast, the optimized decision tree model, following parameter tuning and cross-validation, achieved superior predictive accuracy.