Tree mortality in western U.S. forests forecasted using forest inventory and Random Forest classification

Abstract

Climate change is projected to significantly affect the vulnerability of forests across the western United States to wildfires, insects, disease, and droughts. Here, we provide recent mortality estimates for large trees for 53 species across 48 ecological sections using an analysis of 23,215 Forest Inventory plots and a Random Forest classification model. Models were also used to predict mortality in future FIA inventories under the RCP 4.5 emissions scenario. Model performance indicated species identity as the most important predictor of mortality under both current and future scenarios, with contributions from climate and soil variables. Our results show relatively high levels of recent mortality in the Middle and Southern Rocky Mountains driven by high mortality in Populus tremuloides, Pinus contorta, Pinus albicaulis, and Abies lasiocarpa. Low levels of mortality were observed in several species, with <1% annual mortality observed throughout all other sections. Future mortality was predicted to increase significantly for most species and correlated well with recent mortality at the species level, but not at the plot level. These results suggest that future attempts to model or predict widespread forest mortality would benefit from more research on vulnerable species and that significant mortality events in some species may not be important for dynamics across all systems.

Future climate risks from stress, insects and fire across US forests

“…Barbero et al, 2014) or drought risks (Buotte et al, 2019). The spatial patterns of insect model projections are consistent with previous projections for several major insect species (Bentz et al, 2010) and overall magnitude is similar to coarse‐level ecoregion projections in parts of the western US (McNellis et al, 2021) (Figure 4). Further, the climate sensitivities of insect mortality for several western US pine species with the highest historical insect‐driven mortality were consistent with estimates in the literature (Figure ) (Bentz et al, 2010; Creeden et al, 2014).…”

Predicting Regional-Scale Elevated Groundwater Nitrate Contamination Risk Using Machine Learning on Natural and Human-Induced Factors

“…For classification tasks, the dependent variable is classified into binary classes by each classification tree based on the probabilities estimated by the trees and a probability cutoff, and finally, the most popular class gets voted . The RF model was selected in this study as it is highly robust against problems like multicolinearity among the predictors, overfitting, bias, noise, and outliers. , RFs are popularly applied in hydrological, ,, extreme-climate events, , and climate change studies. , …”

“…81,82 RFs are popularly applied in hydrological, 6,63,83 extreme-climate events, 84,85 and climate change studies. 86,87 Maximum prediction accuracy of the RF model was achieved by creating 1000 trees, fixing the number of predictors during variable selection step at each splitting node as 4 (closest integer to the square-root of the number of predictors selected), taking one-third of the sampled data as out-of-bag (OOB) sample by bagging method and using 10fold cross-validation (CV) process. As our gridded nitrate data had class imbalance (78% 0-data and 22% 1-data), we applied embedded downsampling with replacement technique, i.e., each tree was supplied with an equal number of 1 and 0 data (1836 were 1-data in the training dataset).…”

Tree mortality response to drought‐density interactions suggests opportunities to enhance drought resistance

“…Predicting tree mortality has proven difficult (Rowland et al., 2021) and the specific physiological mechanisms responsible for tree mortality resulting from these hot droughts are still being investigated (Breshears et al., 2018; Choat et al., 2018). Nevertheless, tree mortality associated with hot drought has been clearly demonstrated in both manipulative experiments (Adams et al., 2017) and observational studies (Paz‐Kagan et al., 2017), and is expected in increase as temperatures rise (Allen et al., 2010; Bradford & Bell, 2017; McNellis et al., 2021).…”