Tree Height Prediction Using a Double Hidden-Layer Neural Network and a Mixed-Effects Model.

Area of Science:

• Forestry Science
• Computational Modeling
• Machine Learning

Background:

• Accurate tree height estimation is crucial for forest management and biomass assessment.
• Traditional models face challenges in capturing complex growth dynamics and spatial dependencies.
• Double hidden-layer neural networks show promise for enhancing prediction accuracy.

Purpose of the Study:

• To compare the predictive performance of a double hidden-layer neural network against a nonlinear mixed-effects model for tree height estimation.
• To introduce a novel, high-precision method for predicting tree height in forest plantations.
• To evaluate model accuracy using key statistical metrics.

Main Methods:

• Developed a double hidden-layer back propagation (BP) neural network using optimization strategies like k-fold cross-validation.
• Constructed a nonlinear mixed-effects model considering data differences and autocorrelation.
• Compared model performance using coefficient of determination (R²), RMSE, and MAE.

Main Results:

• The double hidden-layer BP neural network achieved an R² of 0.9068, outperforming the mixed-effects model (R² = 0.8590).
• The BP neural network demonstrated significantly lower RMSE (1.3197 vs. 1.6230) and MAE (1.2736 vs. 2.2658).
• Performance improvements for the BP network included a 5.56% increase in R², 18.69% decrease in RMSE, and 43.79% decrease in MAE.

Conclusions:

• The double hidden-layer BP neural network provides a more accurate and precise method for tree height prediction than nonlinear mixed-effects models.
• This study validates the efficacy of advanced neural network architectures for ecological modeling.
• The findings offer a valuable tool for improving forest inventory and management practices.