Generating Tree-Level Harvest Predictions from Forest Inventories with Random Forests

Abstract

Wood supply predictions from forest inventories involve two steps. First, it is predicted whether harvests occur on a plot in a given time period. Second, for plots on which harvests are predicted to occur, the harvested volume is predicted. This research addresses this second step. For forests with more than one species and/or forests with trees of varying dimensions, overall harvested volume predictions are not satisfactory and more detailed predictions are required. The study focuses on southwest Germany where diverse forest types are found. Predictions are conducted for plots on which harvests occurred in the 2002–2012 period. For each plot, harvest probabilities of sample trees are predicted and used to derive the harvested volume (m³ over bark in 10 years) per hectare. Random forests (RFs) have become popular prediction models as they define the interactions and relationships of variables in an automatized way. However, their suitability for predicting harvest probabilities for inventory sample trees is questionable and has not yet been examined. Generalized linear mixed models (GLMMs) are suitable in this context as they can account for the nested structure of tree-level data sets (trees nested in plots). It is unclear if RFs can cope with this data structure. This research aims to clarify this question by comparing two RFs—an RF based on conditional inference trees (CTree-RF), and an RF based on classification and regression trees (CART-RF)—with a GLMM. For this purpose, the models were fitted on training data and evaluated on an independent test set. Both RFs achieved better prediction results than the GLMM. Regarding plot-level harvested volumes per ha, they achieved higher variances explained (VEs) and significantly (p < 0.05) lower mean absolute residuals when compared to the GLMM. VEs were 0.38 (CTree-RF), 0.37 (CART-RF), and 0.31 (GLMM). Root means squared errors were 138.3, 139.9 and 145.5, respectively. The research demonstrates the suitability and advantages of RFs for predicting harvest decisions on the level of inventory sample trees. RFs can become important components within the generation of business-as-usual wood supply scenarios worldwide as they are able to learn and predict harvest decisions from NFIs in an automatized and self-adapting way. The applied approach is not restricted to specific forests or harvest regimes and delivers detailed species and dimension information for the harvested volumes.

The Impacts of Calamity Logging on the Development of Spruce Wood Prices in Czech Forestry

“…The introduction of new technologies should allow for the rapid recovery of a forest [20] that can bind carbon from the atmosphere [19,20], produce wood that can serve as a biofuel [21], and retain more water in the landscape, thereby cooling the atmosphere. The secondary impact of rapid forest recovery is the long-term stabilization of annual logging [22][23][24] and hence also the stabilization of wood prices [25][26][27][28], including the sustainability of the whole foresty sector [29][30][31][32][33][34].…”

Estimating Forest Stock Volume in Hunan Province, China, by Integrating In Situ Plot Data, Sentinel-2 Images, and Linear and Machine Learning Regression Models

“…The RF is a decision tree algorithm and an effective machine learning model for predicting a forest of variables. Based on its powerful modeling capabilities, the RF regression has been widely used in scientific research [94][95][96][97][98][99]. The principle of the RF algorithm is to use the bootstrap method to randomly extract multiple samples to generate a group of regression trees (ntree) from the original sample population.…”

Prediction With Mixed Effects Models: A Monte Carlo Simulation Study

“…While this accuracy differential between multilevel models has been demonstrated universally in regression contexts, it is presently shown to remain relevant in classification contexts. Furthermore, while results by Speiser et al (2019, 2020) and Kilham et al (2019) have demonstrated that conditions may exist in which single-level classification trees and random forests may outperform their multilevel analogues, this study adds to the debate of the necessity of multilevel classifiers through a demonstration of NB’s efficacy under multilevel data conditions. These results bolster the findings of Demichelis et al (2006) and Zhang et al (2018) illustrating that NB does perform comparably to or greater than the multilevel Bayes classifier (among others) under the same conditions.…”

“…The GMERT models consistently outperformed the GLMML variants on measures of predictive mean absolute deviation and predictive misclassification rates; to date, this represents the most comprehensive simulation study on multilevel classifiers. Expanding on this finding, Kilham et al (2019) compared REEMTree, MERF, and GLMML, with MERF performing best on measures of both variance explained and root mean squared error in harvest prediction data. Capitaine et al (2019) also found substantially reduced bias in estimates for MERF above REEMTree and GLMML in the context of both simulated and archival low- and high-dimensional genetic data.…”

“…Of the literature described above, only three studies focused specifically on classification situations—namely Hajjem et al (2017), Kilham et al (2019), and Ngufor et al (2019)—rather than regression situations. Given the sparsity of comparisons among the current algorithms in the classification context, this work stands as a novel contribution via the relatively comprehensive set of conditions on which the data were simulated, and the breadth and relative novelty of algorithms employed.…”