Nonparametric Individual-Tree Growth Models for the Northern Forest
| Authors: | Neal F. Maker, John D. Foppert |
| Year: | 2023 |
| Type: | Proceedings - Paper |
| Station: | Northern Research Station |
| DOI: | https://doi.org/10.2737/NRS-GTR-P-211-paper19 |
| Source: | In: Kern, Christel C.; Dickinson, Yvette L., eds. Proceedings of the first biennial Northern Hardwood Conference 2021: Bridging science and management for the future. Gen. Tech. Rep. NRS-P-211. Madison, WI: U.S. Department of Agriculture, Forest Service, Northern Research Station: 83–95. |
Abstract
Tree growth modeling is important for forest managers, but is difficult in structurally and compositionally diverse forests where growth depends on complex ecological relationships. Many of those relationships are not well studied, and traditional parametric models struggle to adequately account for them. We describe a nonparametric machine learning approach to modeling, which uses data to determine the methodology, accommodates high dimensional data, and could be well suited to capturing complicated, poorly understood interactions between predictors. We compare this approach to the traditional parametric approach by constructing diameter increment models for the United States northern forest region. In this case, the approach led us to build a gradient boosting machine that decreased the root-mean-squared error by 6.5 percent over the traditional, nonlinear least squares model, and which was especially accurate in predicting the growth of trees in less common situations, such as especially large trees and trees in forests with below- and above-average basal areas. Analysis of the model structure showed that its predictions are driven by multiple complex interactions. We explore several of them and discuss the potential of the nonparametric machine learning approach for predictive tree growth modeling and for studying ecological processes underlying tree growth.