248 Walker and Dahle: Likelihood of Failure of Trees Along Utility Rights-of-Way Wellpott et al. (2006) suggested that the approach used in Ancelin et al. (2004) is not a realistic repre- sentation of wind loading on individual trees (Gar- diner et al. 2008). A possible alternative approach to modeling wind risk of individual trees is to make use of the competition indices developed for predicting growth conditions of individual trees within stands, which Achim et al. (2007) demonstrated are extremely well correlated to the wind loading of individual trees within a mature Sitka spruce plantation (Gardiner et al. 2008). To become more than research tools, these predic- tive mechanistic models must be incorporated into forest-management systems in ways that are useful and practical (Gardiner et al. 2008). Yet currently, due to the need of numerous, precisely measured parame- ters, these models are not practical in many cases. While these tools have not been widely utilized in practice, Gardiner et al. (2008) suggest that, first, their operation must be simple and interpretation of the results routine (Gardiner et al. 2008). Future research into predictive mechanistic models should integrate decision-support tools to simplify each model’s oper- ation, such that the requirements are a hierarchical set of questions on the characteristics of the trees and site, and outputs are different levels of risk, low to high (Gardiner et al. 2008; Kamimura et al. 2008). Moreover, the integration of other remote-sensing data and additional geographic information system (GIS) layers to enhance location-specific conditions may be useful for the prediction of tree failure along utility ROWs. Statistical Approaches Statistical approaches, much like explanatory approaches, utilize geographic characteristics and physical prop- erties of trees as variables to aid in the prediction of windthrow (Kabir et al. 2018). However, instead of utilizing a single statistical tool, such as linear regres- sion, statistical approaches examine the relationships of the measured properties through the lens of multi- ple statistical tools to see which tool best predicts windthrow (Kabir et al. 2018). Examples of such properties include Generalized Linear Models (GLMs), Monte Carlo simulation (MC), classification and regression trees (CART), Random Forests (RF), and Artificial Neural Networks (ANN). Ciftci et al. (2014a) utilized a Monte Carlo–based methodology for the prediction of individual tree fail- ure. Their study attempted to quantify the probability ©2022 International Society of Arboriculture of failure of 2 maple trees in Massachusetts. Although one of the first and more novel methods for the pre- diction of likelihood of failure of individual trees, this study is limited in that it was computationally inten- sive and not well-suited for the large data sets that would be associated with trees along electric distribu- tion ROWs (Ciftci et al. 2014a). Kamimura et al. (2016) developed a logistic regression model and utilized a GALES-based model for individual tree failure from 1 storm at an Aquitaine forest in southwestern France, then validated the model against the next storm at that location. Their results suggested that GALES was capable of pre- dicting wind-damage risk of trees on certain soils, while their statistical models were not able to be gen- eralized to other locations or storm events (Kamimura et al. 2016). Kabir et al. (2018) used the covariates location, height, DBH, existence of severe defects, whether or not a tree had been pruned, and whether or not a tree had been removed in the immediate proximity of the tree in question to demonstrate that tree failure can be statistically estimated. Kabir et al. (2018) utilized several statistical tools, including a GLM with a Ber- noulli response, CART, a multivariate adaptive regres- sion spline (MARS), ANN, Naïve-Bayes Classifier, boosting, RF, and an ensemble model of RF and boosting. The ensemble model yielded the best pre- diction accuracy for estimating the failure probability of trees for their data set (Kabir et al. 2018). This was a novel approach to predicting windthrow of individ- ual trees and contributed to the literature, primarily by demonstrating the potential predictability of tree failure using statistical models. However, the results of this study cannot be used to estimate tree-failure probabilities for either other storms at the study site or at other locations because the models implemented included data from only 1 storm, at the 1 study site (Kabir et al. 2018). Thus far in likelihood-of-failure research, most statistical analyses have limited their statistical tools to linear or logistic regression (Kabir et al. 2018). Nevertheless, Ciftci et al. (2014a) and Kabir et al. (2018) have demonstrated the utility of other statisti- cal tools. Additionally, most studies are not able to be generalized as the models developed only apply to 1 location or 1 storm due to the lack of validation in subsequent locations or storms. Yet, Kamimura et al. (2016) developed models, both statistical- and GALES-based, in 1 storm and validated them against
July 2022
| Title Name |
Pages |
Delete |
Url |
| Empty |
Ai generated response may be inaccurate.
Search Text Block
Page #page_num
#doc_title
Hi $receivername|$receiveremail,
$sendername|$senderemail wrote these comments for you:
$message
$sendername|$senderemail would like for you to view the following digital edition.
Please click on the page below to be directed to the digital edition:
$thumbnail$pagenum
$link$pagenum
Your form submission was a success.
Downloading PDF
Generating your PDF, please wait...
This process might take longer please wait
