An Analysis of Electric Distribution Tree-Caused Outages: Retrospective Case Series

  • Arboriculture & Urban Forestry (AUF)
  • June 2026,
  • jauf.2026.016;
  • DOI: https://doi.org/10.48044/jauf.2026.016

Abstract

Background Electric utilities spend billions of dollars annually clearing trees around power lines, while vegetation remains the leading cause of power outages at many utilities. Limited research exists to determine the cause of most power outages due to trees, particularly distribution-level outages.

Methods A team of utility urban foresters at a Mid-Atlantic electric utility in Virginia, USA, investigated over 1,104 outages caused by tree failures to identify characteristics of failed trees. Using an observational retrospective case series approach, data from afflicted (naturally fallen) trees directly involved in power outages were collected on a mobile phone application for each individual outage, including tree type (hardwood, softwood), genus, size (both diameter and height), tree defects and health status prior to the failure, and the type of failure—root, stem, or branch. After exclusion of incomplete or duplicated records, statistical tests on 844 outages evaluated simple associations using Chi-square and Cochran-Mantel-Haenszel (CMH) test, as well as probabilities using multiple (more than one factor), polytomous (more than two levels of response), nominal (categorical response levels), and logistic regression, with the main objective to identify factors associated with frequency of failed tree parts.

Results This study found that most tree-caused power outages (TCOs) were the result of live trees failing at the trunk, and 46% of the failures had no visible defect prior to failure. While hardwood trees had more branch failures than softwood trees, overall, there was a low likelihood of branch failures. Some hardwood species were more likely than others to have branch failure; specifically, yellow poplar and maple, while pines were less likely, which was unexpected. Defects prior to failure were observed in only half of TCOs; however, of all trunk failures (n = 450), dead trees were 47.11%.

Conclusions This study highlights the importance of implementing the Utility Tree Risk Assessment Best Management Practices (BMPs) and expanding on existing methods to identify trees with either an elevated likelihood of failure and/or likelihood of impact. Purposefully managing an urban forest at a landscape scale requires knowledge of the species composition, size class, and how trees react to biological and physical factors in the environment.

Keywords

Introduction

Tree failure is the cause of most distribution power outages in the United States (Guggenmoos 2003; Walker and Dahle 2023), particularly trees outside of the right-of-way (ROW)(areas that electric utilities have a legal easement to build the infrastructure to deliver power) that are often situated in linear miles of forests adjacent to overhead power facilities. Although utilities spend valuable resources pruning trees away from poles, wires, and other facilities to keep ROWs clear for access, millions of trees outside of ROWs have the potential to cause lengthy outages, especially during extreme weather events (Lott and Ross 2005; Guikema et al. 2006; Guggenmoos and Sullivan 2012; Executive Office of the President 2013; Wanik et al. 2018; Taylor et al. 2022). In 2013, the White House Report on Grid Resiliency stated that weather-related outages are estimated to cost the US economy an inflation-adjusted annual average of $18 billion to $33 billion (Radmer et al. 2002; Executive Office of the President 2013).

Utility vegetation management research has focused on the cost effectiveness of cyclical pruning and clearing maintenance programs (Nowak and Ballard 2005; Poulos and Camp 2010; Graziano et al. 2020); storm resilience (a measure of how quickly power is restored after major storm outages, or how a severe weather event will impact outages)(Reed 2008; Wanik et al. 2018; Coder 2022a; Coder 2022b; Suttle et al. 2022); and the effects of pruning on tree-caused outages (TCOs)(Guggenmoos 2007). Studies have focused on predicting outages by analyzing tree growth (Radmer et al. 2002) and precipitation (Apostolov et al. 2023; Santos et al. 2024) but primarily use existing vegetative cover and historical outage data to predict future outages. However, there is limited published research that identifies the commonalities within TCOs which occur during nonstorm conditions or that describes best practices to identify and mitigate risk posed by those common TCO sources. Simpson and Van Bossuyt (1996) drew attention to the removal of defective or potentially high risk trees almost 30 years ago after a consultant study, but resource limitations and storm restoration costs have limited implementation of extensive elevated risk tree removal programs.

There are several ways trees cause power outages. The most straightforward occurs when an individual stem of a tree grows from underneath into the overhead facilities, which are approximately 10 m to 15 m from the ground. Another way for the tree to cause an outage is for a branch to grow in from a tree along the ROW edge, which typically causes an intermittent outage as it will burn off the tips of the branch. Long term outages due to vegetation are of ten caused by either a branch or section of a tree breaking and failing across one or more phases of the circuit or an entire tree uprooting or breaking along the trunk and failing, taking electric poles and wires down along its path to the ground.

The most widely used classification for TCOs is which part (root, trunk, or branches) failed and caused the disruption. Yet, literature has of ten centered around data sets narrowly focused on a specific species, size classes, or microclimates, rather than multiple species and forest cover at a landscape scale (Klein et al. 2020). Furthermore, the landscape-scale knowledge that utility vegetation managers can gain from such studies is limited and generally not disseminated in a straightforward format or originates from a state extension agency with limited viewing by arborists outside of that state (University of Florida IFAS 2015)

van Haaften et al. (2021) conducted a meta-analysis of 161 studies related to tree failure. They reported that there have been 142 factors studied in relation to tree failure covering 320 species for stem failure, 102 species for root failure (each was conducted on only 2 to 3 species), and 32 species for branch failure. In their meta-analysis, no studies reported correlations between any site factor and branch failure. Most studies focused on diameter at breast height (DBH, 1.37 m from the ground), tree height, or tree mass. Kane (2008) examined tree failure in campgrounds in Cape Cod, Massachusetts, USA, after a 45-m/s windstorm and found that softwood trees failed more often than hardwood trees during leaf-off conditions. The study also found that pruning and normal defects such as fire scars, cavities, etc., did not predict failure, although that was more likely due to lack of maintenance.

Stem failures have also been studied extensively in a controlled environment or urban setting (Jodas et al. 2024) using forces such as tensile pulls using ropes (Lilly and Sydnor 1995; Dahle et al. 2014). While stem failure was mentioned in a meta-analysis of 92 studies by van Haaften et al. (2021), no overall conclusions were found by the researchers about causes. Research related to stem failure found that trees with defects such as cavities were more likely to fail at the stem (Kane 2008).

Root failure, anchorage, and subsequent tree stability have also been examined on a micro level. Root cohesion (Sakals and Sidle 2004), anchorage (Lundström et al. 2007; Tsen-Tieng et al. 2018), and architecture have been examined in different climates, geography, and forest ecotypes. Much of the work conducted around root failure has focused on mechanical means of failure using machines, which is not necessarily reflective of real field conditions (Stokes et al. 2000; Dupuy et al. 2005; Stokes et al. 2007; Kim et al. 2020), or after storm events (Hamilton et al. 2006; Kabir et al. 2018), and sometimes in specific types of soils or within certain species of trees (Ennos 1993; Cucchi et al. 2004; Dorval et al. 2016). While these studies can inform how tree root failures occur inside a narrow range of variables, they do not provide enough information to be predictive of failure potential across a variety of species, size classes, growing conditions, and various types of seasonal weather (Coder 2010).

Like root failures, much of the branch failure research has utilized mechanical testing to simulate branch failure (Smiley 2003; Dahle et al. 2006; Kane and Clouston 2008; Kane et al. 2008; Slater and Ennos 2013; Kane and Finn 2014; Dahle et al. 2022; Eckenrode et al. 2022). These researchers investigated branch failures after a large snowstorm and concluded that the larger DBH trees were correlated with more surface area on branches, which allowed for greater snow loading, causing the larger branches to fail. Yet, how branch surface area correlates to other environmental loading events is not clear.

Moreover, many studies of tree failure have concentrated on the broader urban forest rather than the utility urban forest. While there are certainly similarities between the two, utility urban foresters would benefit from a targeted approach due to the unique configurations across urban, suburban, and rural sites, as well as the severity of consequences associated with TCOs. The meta-analysis of most urban forestry studies by van Haaften et al. (2021) concluded that tree failure is complicated and that there is not currently a model that includes all the relevant factors.

Industry and utilities have enacted “circuit hardening” utilizing “ground to sky” techniques to attempt to proactively minimize TCOs by removing all overhanging limbs, clearing every tree underneath and overhead circuits and aggressively pruning to prevent outages from grow-ins. It is our contention that this solution has been mistakenly overapplied to a problem that cannot be solved by these techniques alone, particularly when many TCOs are not prevented by those measures.

The objective of this study was to investigate common failure types of individual trees associated with TCOs along an electrical distribution system to examine the relationships between TCOs and environmental and structural factors. This information was utilized to develop a set of common TCO profiles to inform actionable plans for vegetation management practices to lower incidences of TCOs on the power grid. We hypothesize that environmental and structural factors can be used to develop profiles from trends in TCO tree failure patterns experienced along heavily forested distribution ROW corridors in the Mid-Atlantic. It is important to note that this study does not consider differences between standing and failed trees. Rather, in a series of TCO investigations spanning 33 months in central Virginia, with exclusive focus on failed trees, we sought and synthesized associations between tree and environmental characteristics. We included variables such as slope, combined with tree genera, tree size (height and diameter), health status and visible defects before the failure, and the month of the year in our analyses in order to identify trends. The overarching goal of this research was to develop a better understanding of the patterns of TCOs in electricity distribution systems, thus aiding utility urban foresters in managing their forests.

Methodology

Study Area

The TCO data was collected between June 2020 through February 2023 along a distribution system in central Virginia (Figure 1). The extent of the service area encompasses 22 counties and over 7,500 mi (12,070 km) of overhead conductor. It should be noted that the species composition and forest cover type vary greatly regionally as well as locally. It is also important to note the existing practices of this utility in terms of branch removal, which varies depending on species, construction (single phase or three phase), and whether the overhanging branches are within the zone of the substation and the first protective device, which are generally kept clear. However, the zone between substations and the first protective devices makes up only a small amount of the total mileage actively managed by the utility. Large overhanging limbs are common in areas with mature forest canopy, particularly with the genera Quercus. In practice, it is often difficult to reach the highest limbs with existing equipment, particularly in steep terrain. Specialty utility tree climbers have become scarce in areas close to metropolitan cities, where they can demand higher salaries by doing high-end private arboriculture work, which is creating a market demand for taller buckets as forest canopies mature.

Map of Virginia (a) and sampling location (b) of Rappahannock Electric Cooperative service territory in central Virginia, USA.
Figure 1.

Map of Virginia (a) and sampling location (b) of Rappahannock Electric Cooperative service territory in central Virginia, USA.

Existing practices also include a 5-year pruning cycle length, which includes clearing tall growing species from edge to edge of a (generally) 40-ft (12.19-m) ROW, taking down dead and other trees determined likely to fail before 5 years, and directional pruning of yard trees. One year after pruning, ROWs are sprayed with a low-volume foliar herbicide application, targeting tree species with the potential to grow tall enough to contact the overhead facilities as well as brushy invasive species that could limit access to the infrastructure or impede restoration efforts. Additionally, low-performing circuits are selected for a midcycle hazard tree evaluation and removal at approximately the midpoint of the maintenance cycle.

While obvious that a utility’s past and present vegetation management practices influence the types of tree failures that induce TCOs, future research could examine TCO tree failure profiles across various maintenance practices, which may further illuminate how past and present vegetation management impacts trends in TCOs.

Field Work

This study investigated TCOs, all of which were greater than 5 minutes. Most TCOs were investigated retrospectively after power restoration and often during fair weather conditions. Each outage incident was initially recorded by a dispatcher and reported by a line worker; later, outages identified as having been caused by a tree were extracted and then compiled into a spreadsheet for the director of the vegetation department (lead author of this study). Information reported by the line worker included if the offending tree was alive/dead prior to the failure, position (inside or outside ROW), and if a limb failure was from an overhanging limb.

This initial outage information was then sent to the appropriate utility service center and utility foresters for a TCO field investigation. The utility urban foresters that conducted the TCO field investigations were International Society of Arboriculture (ISA) Certified Arborist Utility Specialist® with 3 to 30 years of experience.

Due to cost constraints, the local utility was most interested in the investigation of outages that impacted the electric grid severely and consequently affected the customers the most. Thus, inspection priority was based on the number of customers affected and the type of device impacted. The workload of the utility urban forestry staff played a role in how many and which TCOs were investigated. If a large restoration event was underway, the priority for the utility was to restore power to customers as quickly as possible, and thus TCO investigations were not conducted. This may have biased the observations, but it is not feasible or appropriate for the utility to focus on anything but its customers during restoration events. It is possible that this convenience sampling protocol skewed the data towards smaller storm events or non-weather-related TCOs.

Field data were collected by utility foresters using a custom smart phone application, which was later modified into a web access portal. An offending tree’s failure part was recorded as occurring in one of the following tree part categories: trunk (top, middle, or base); branch (overhang, side, or top); or roots (decay, lifted root plate, or broken). Due to the constraints of the assumptions for statistical tests, such that all expected cell counts of row and column marginal totals in the contingency tables should be greater or equal to 5 (Stokes et al. 2012b), the data from 9 subcategories were pooled into 3 main (branch, trunk, or roots) categories. Tree health of each fallen tree was recorded as dead, alive, or in decline, and it was the assumed estimation of the health status prior to the failure, as visible tree health issues develop over several months or years. The retrospective approach with regard to health was chosen because of its empirical validation in silviculture internationally, including poplar in China (Lu et al. 2020; Wei et al. 2022); pine in Finland (Kurkela et al. 2005); cedars, eucalyptus, and ash in Mexico (Saavedra-Romero and Alvarado-Rosales 2024); resak in Indonesia (Rachmadiyanto et al. 2024); as well as ash, stringybark, messmate, and gum trees in Australia (Gibbons et al. 2008). The data collection protocol extended existing workload for the utility company staff, particularly because an exact outage location was not recorded by the utility worker originally recording it. However, the utility places a great value on analytics, return on investment (ROI), and cause and effect of vegetation-related work.

Diameter at breast height (DBH)(1.4 m) was estimated into 5 size classes: 0 cm to 30.5 cm; 30.6 cm to 45.7 cm; 45.8 cm to 61.0 cm; 61.1 cm to 91.4 cm; and > 91.4 cm. Tree heights were categorized into 3 size classes: 0 m to 15.2 m; 15.3 m to 30.5 m; and > 30.5 m. Due to the low number of trees that were > 30.5 m, these were combined with the middle category and named > 15.2 m. Slope was estimated into 4 categories: 0° to 10°; 11° to 30°; 31° to 40°; and > 40°.

Tree taxonomy was collected at the genus level. There were initially 11 genera collected in the field. An additional grouping created a new field titled Tree Type by combining trees as either hardwood (angiosperm) or softwood (gymnosperm). During exploratory analysis, genera with fewer than 10 trees were pooled into an “Other” category to allow meaningful statistics. The Other category included the following genera: Carya, Liquidambar, Platanus, Fagus, Robinia, Gleditsia, and unidentified trees (n = 65, 7.7%). The remaining genera were: Fraxinus (ash), Acer (maple), Quercus (oak), Pinus (pine), and Liriodendron (yellow poplar).

The range of the months included in the final dataset was from June 2020 to February 2023, resulting ineach calendar month being represented for at least two consecutive years. The months were combined into quarters. January through March were Q1; April through June were Q2; July through September were Q3; and October through December were Q4 (Table 1). Precipitation data is included for Ashland, Virginia, USA (Table 1)(NOAA 2026).

View this table:
Table 1.

Monthly precipitation (cm) by quarter (2020 to 2023) from Ashland, Virginia, USA (NOAA 2026).

The presence and types of pre-existing defects were recorded in the field. Categories included none, unknown, codominant stems, cracks, lean, and wounds. Decay was not noted specifically, as the only observations were visual, and decay was expected to be present when wounds were present. No tools were used to observe defects, simply visual observation and recording into the application.

There were 1,104 TCOs examined and recorded in the field. Of these, 260 records had fundamental flaws, such as missing data fields related to the type of failure, and were removed, leaving 844 unique records. The sample sizes for individual analyses varied (n = 817 to 844) based on additional individual missing observations in variables. The most complete dataset (n = 844) represented 77 substations and 176 circuits.

The selection of the trees was not at random but an exhaustive cohort of the specific cases of trees that collapsed and caused power outages, without taking data on control or standing trees. This type of study is analogous to observational retrospective case series in epidemiology when the disease-afflicted patients (or their medical history) are studied for common exposure in their past to generate hypotheses. The cause and effect can’t be established. This creates the restriction such that we can’t calculate the probability of tree failure among standing trees surrounding the power lines. This, however, does not prevent us from studying the circumstances and characteristics of the collapsed trees with relation to failed tree parts, as it provides valuable information to the utility company as well as to the arborists. The strength of this study is that the subjects (afflicted trees) were studied in their entire possible population of the cohort in the defined period of 33 continuous months and not the predetermined or limited proportion of the afflicted trees. Therefore, the error of estimated proportions is not associated with sampling error but in the biological variability and human error.

Data Analysis

The analysis of TCO data followed a two-step process. First, we utilized both Chi-square test of homogeneity of proportions and Pearson’s Chi-square tests, CMH statistics, and simple logistic regression to serve as a screening criterion for the second stage. In the second step, we used the significant factors from the first set of analyses to build models to predict the tree failure part (branch, trunk, or roots) based on the combination of the specified variables using multiple polytomous nominal logistic regression.

To ensure the validity of statistical estimates from the Chi-square tests and CMH statistics, all contingency tables were required to meet the minimum expected cell frequency of 5 (Stokes et al. 2012a). Meeting this assumption required the ‘pooling’ of categorical variable levels across multiple variables. The month of year variables were pooled into quarters, such that January through March were Q1; April through June were Q2; July through September were Q3; and October through December were Q4.

The homogeneity of proportions of the response, tree failure part (branch, trunk, or roots), was evaluated by the Chi-square probability test. The null hypothesis of this test states that the 3 tree failure parts occurred with equal frequency (0.333…). Additionally, the Pearson Chi-square test was utilized to examine the association of individual categorical variables (tree genera, tree height, slope, existing prior defects, month and quarter of the year) with tree failure part categories. The association between each of the ordinal categorical variables (tree health, DBH, and tree height) and the tree failure part were examined using the CMH statistics, using the comparison of the row means scores to detect shifts in the ordinal distribution across different tree failure parts.

A common response function model for categorical data is the logit (Stokes et al. 2012b). Logistic regression is a modeling strategy that relates the logit function for a set of explanatory variables to a linear model. The estimates of odds ratios and measures of association can be obtained from the parameter estimates (slopes and intercepts). Maximum likelihood estimation is used to provide those estimates (Stokes et al. 2012b). The effect of average monthly precipitation on tree failure part was examined using simple logistic regression.

Next, a more detailed subanalysis of association between individual tree failure parts (branch, trunk, or roots) followed, using the significant factors identified in the earlier Pearson’s Chi-square, CMH, and simple logistic regression analyses. In this second stage of analysis, we built models to predict the tree failure part (branch, trunk, or roots) based on the combination of the specified variables using multiple polytomous nominal logistic regressions. Specifically, model 1 included the factors tree health and pre-existing defects, while model 2 included the factors genera, DBH, and height.

For the response variable tree failure part with 3 nominal response levels (branch, trunk, or roots), 2 logits were calculated using the Trunk as the reference tree part. The 2 logits were the probability of Failure of Branch over the Failure of Trunk (B:T) and the probability of Failure of Root over the Failure of Trunk (R:T) for each of the specific combinations of independent variables.

Logit 1=Logithealth, defect, Branch=Log [Phealth, defect, Branch/Phealth, defect, Trunk]Logit 2=Logithealth, defect, Root=Log [Phealth, defect, Root/Phealth, defect, Trunk]

For 3 nominal response levels, these 2 logits were calculated for each of the combinations (18 combinations = 3 health × 6 defects); similarly, for the second model these were the combinations of 6 genera, 5 DBH, and 2 height (60 combinations). Proc LOGISTIC, Proc CATMOD, and Proc GLIMMIX of SAS (SAS Institute, Cary, NC, USA) were used to obtain the P-values, odds ratios, predicted probabilities, and frequencies of the generalized linear mixed models.

The factors were selected based on the significance in the earlier simple, association-based analyses (Chisquare, CMH, and simple logistic regression); the reference category for each factor was the category with the largest frequency in the entire dataset. The month and slope were not included for the final models since the low frequency of trees in the numerous resulting combinations did not meet the assumption criteria (Stokes et al. 2012b).

If one has a dichotomous response and generally represents the proportion of those subjects with an event (versus no event) outcome as P, then the logit response function can be written as log of the probability of event happening over the probability of that not happening, or log[P / (1 – P)], or log of odds. For the polytomous response (3 tree failure parts response levels: Branch, Trunk, or Roots), there are 2 logits available to calculate with respect to a reference category (for instance, branch to trunk B:T and root to trunk R:T). For illustration of the notation used, one of the associations we were interested in evaluating was if the odds of failing branches over the trunk (B:T) are different between two different genera, like oak and pine. The log difference of the odds of B:T between oak and pine can be expressed or reported as the Odds Ratio (ORB:T oak vs. pine). If the odds ratio is one, there is no difference between the odds of oak vs. pine; in other words, the ratio of failed branches to trunks was similar in both oaks and pines. The inclusion of value “one” in the 95% confidence limits on odds ratio estimates represents the lack of significant differences between odds of B:T for oaks vs. pines. On the contrary, if the odds ratio confidence limits do not include the value “one”, that is evidence of difference between the two odds.

Statistical analyses were conducted in JMP® (Version Pro 18.0.2; SAS Institute, Cary, NC, USA) and SAS (Version 9.4, SAS Institute, Cary, NC, USA). The significance criterion alpha for all tests was 0.05.

Results

Tree Failure Part

Using Pearson’s Chi-square test of homogeneity of proportions, we found an inequality in the overall failure pattern of 215 (25.5%) root, 453 (53.7%) trunk, and 176 (20.8%) branch failures (χ2 = 159.83; degree of freedom [DF] = 2; P < 0.0001; n = 844) (Table 2). The failure part of trunk failure had the highest prevalence, with more than 50% of all TCOs detected being attributed to the trunk of the tree’s failure.

View this table:
Table 2.

Number (n) and percentage (%) of tree failures by failure part. The omnibus test of homogeneity of proportions (TP, Chisquare) tested the null hypothesis (P0 = 1/3 for each category), meaning that all 3 failure parts were equally represented, such as 1/3 of all failure problems were root problems, 1/3 were trunk problems, and 1/3 were branches. The specific tests measured if the proportions of failure parts differed among the different levels of the examined variables, accessed by Chi-square or Cochran-Mantel-Haenszel tests. For instance, the significant p-value for genera and failure by tree part indicated that the proportion of failure parts depends on the genera. Specifically, while the majority of failures for Ash were trunks (77.1 %), the highest failure parts for Yellow poplar were the branches (51%). The highest overall frequency category within each variable was used as a reference (Ref) for the next step models in the proportional odds ratio analysis (Table 3). Ref (reference); DBH (diameter at breast height).

Tree Type

We observed significant nominal association between tree type and failure part (χ2 = 63.67; DF = 2; P-value < 0.0001; n = 844)(Figure 2). Softwood had about 20% higher chances of trunk failures resulting in TCO (66.9%, 170/254) compared to hardwood (48%, 283/590); in exchange, the hardwood were 7 times more likely to have branch failures that resulted in a TCO (28.1%, n = 166) than softwood trees (3.9%, n = 10). Tree-caused outages attributed to root failures occurred slightly more frequently in softwoods (29.1%) than hardwoods (23.9%).

Percentage of tree failures by location within the tree and general tree type (χ2 = 63.67; DF = 2; P < 0.001; n = 844). Width of the columns is proportional to tree type group size (n = 590 hardwood, n = 254 softwood). The vertical axis is the proportion of failure by root (brown), trunk (saffron), or branch (green). Percentage of failure by root, trunk, or branch sum to 100% for a given column.
Figure 2.

Percentage of tree failures by location within the tree and general tree type (χ2 = 63.67; DF = 2; P < 0.001; n = 844). Width of the columns is proportional to tree type group size (n = 590 hardwood, n = 254 softwood). The vertical axis is the proportion of failure by root (brown), trunk (saffron), or branch (green). Percentage of failure by root, trunk, or branch sum to 100% for a given column.

Tree Genera

The distribution of tree failure parts among the 5 genera and the “other” category varied. There was a significant nominal association between genus and tree failure part (χ2 = 118.68; DF = 10; P-value < 0.0001; n = 844)(Table 2). Ash (Fraxinus) had the highest proportion of trunk failures resulting in TCO (77.1%, n = 54) of all genera (Figure 3), followed by pine (Pinus)(66.9%). In contrast, only 20.4% (n = 10) of TCOs were attributed to yellow poplar (Liriodendron) trunk failures.

Percentage of tree failure part by tree genera (χ2 = 118.68; DF = 10; P < 0.001; n = 844). The width of the columns represent genus sample size. The vertical axis is the percentage of failure by root (R)(brown), trunk (T)(saffron), or branch (B)(green). The horizontal axis is the tree genera (ash, maple, oak, pine, yellow poplar, and other). Percentage of failure by root, trunk, or branch sum to 100% for a given column. Significant odds ratios (OR) from multiple nominal logistic regression (Model 2) include (pine is reference): ORB:T maple (10.952), oak (6.693), yellow poplar (26.013), and other (7.322); and ORR:T ash (0.352) and yellow poplar (2.832). See Table 3 for the sample size and full results.
Figure 3.

Percentage of tree failure part by tree genera (χ2 = 118.68; DF = 10; P < 0.001; n = 844). The width of the columns represent genus sample size. The vertical axis is the percentage of failure by root (R)(brown), trunk (T)(saffron), or branch (B)(green). The horizontal axis is the tree genera (ash, maple, oak, pine, yellow poplar, and other). Percentage of failure by root, trunk, or branch sum to 100% for a given column. Significant odds ratios (OR) from multiple nominal logistic regression (Model 2) include (pine is reference): ORB:T maple (10.952), oak (6.693), yellow poplar (26.013), and other (7.322); and ORR:T ash (0.352) and yellow poplar (2.832). See Table 3 for the sample size and full results.

For the branch category of tree failure part, the highest deviation from the expected value occurred in pine (3.9%, n = 10), where the proportion was much lower than in any other genus group, indicating that TCOs attributed to pine branch failures were rare. In addition, TCOs attributed to branch failures were higher for yellow poplar (51.0%, n = 25) and maple (35.7%, n = 20) than other species. The genus with the highest proportion of TCOs attributed to failure part (roots) were oaks (Quercus), in which 31.7% of TCOs attributed to oaks were due to roots. Conversely, ash had the lowest proportion of TCOs due to the tree failure part roots (12.9%)(Figure 3).

Tree Defects

Regarding distribution of the tree defects prior to failure among the tree failure parts, we observed that almost half of all the trees (46.6%, n = 389) had no visible defects, and 26.7% (n = 223) had unknown defects due to potential obstruction of view by the observer (Figure 4). Even so, when including all trees that were responsible for a TCO and possessed defect data (n = 835), there was an overall association of the tree defect category and tree failure part (χ2 = 50.05; DF = 10; P < 0.0001). Standing out was the tree defect category “no visible defects”, which had a higher rate of root failures (32.9%) and, in turn, a lower rate of failed trunks (42.67%) relative to the other defect categories. Cracks and wounds were the most common on the trunks (75% and 73%) from all 6 categories. Yet, when we examined the subset of data for 223 TCOs in which trees possessed visible defects (tree defect categories: codominant, cracks, lean, or wound), there was no significant statistical relationship between visible tree defects and tree failure part for trees responsible for a TCO (χ2 = 7.5; DF = 6; P = 0.2778). This subset followed the general pattern of most TCO occurred (54% to 75%) due to trunks, while the remainder ranged from 9% to 28% for tree failure part categories roots and branches.

Percentage of tree failure part by defects (χ2 = 50.05; DF = 10; P < 0.0001; n = 835). The width of the columns is proportional to the sample size of categories of tree defects which existed prior to failure. The vertical axis is the percentage of failure by root (R) (brown), trunk (T)(saffron), or branch (B)(green). Percentage of failure by root, trunk, or branch sums to 100% for a given column. Significant odds ratios (OR) from multiple nominal logistic regression (Model 1) include (none is reference): ORB:T lean (0 = 0.375), unknown (0.527), and wound (0.3); and ORR:T codominant (0.31), cracks (0.178), lean (0.319), unknown (0.48) and wound (0.229). See Table 3 for the sample size and full results.
Figure 4.

Percentage of tree failure part by defects (χ2 = 50.05; DF = 10; P < 0.0001; n = 835). The width of the columns is proportional to the sample size of categories of tree defects which existed prior to failure. The vertical axis is the percentage of failure by root (R) (brown), trunk (T)(saffron), or branch (B)(green). Percentage of failure by root, trunk, or branch sums to 100% for a given column. Significant odds ratios (OR) from multiple nominal logistic regression (Model 1) include (none is reference): ORB:T lean (0 = 0.375), unknown (0.527), and wound (0.3); and ORR:T codominant (0.31), cracks (0.178), lean (0.319), unknown (0.48) and wound (0.229). See Table 3 for the sample size and full results.

Tree Health

Overall, most of the TCOs were associated with trees in the ‘alive’ category (55.0%, n = 462), while dead trees accounted for 33.5% (n = 282) of the TCOs, and 11.43% (n = 96) of the trees were categorized as ‘in decline’. The CMH analysis which utilized tree health as an ordinal variable (alive, in decline, dead) detected a significant association of the tree health and tree failure part (CMH χ2 = 90.55; DF = 2; P < 0.0001). When examining tree health by tree failure part (Figure 5), the tree health category ‘dead’ had the highest proportion of failed trunks that resulted in a TCO (75.2%). However, of TCOs attributed to this tree health category, only 5.67% (n = 16) were from failed branches. The distribution of tree failure part for both of the remaining tree health categories, ‘alive’ and ‘in decline’, was more homogeneous.

Percentage of tree failures by part and tree health χ2 = 90.55; DF = 2; P < 0.0001; n = 840). Width of the columns depicts sample size in each health category. The vertical axis is the percentage of failure by root (R)(brown), trunk (T)(saffron), or branch (B)(green). Percentage of failure by root, trunk, or branch sum to 100% for a given column. Significant odds ratios (OR) include (alive is reference): ORB:T dead (0.116); and ORR:T dead (0.359). See Table 3 for the sample size and full results.
Figure 5.

Percentage of tree failures by part and tree health χ2 = 90.55; DF = 2; P < 0.0001; n = 840). Width of the columns depicts sample size in each health category. The vertical axis is the percentage of failure by root (R)(brown), trunk (T)(saffron), or branch (B)(green). Percentage of failure by root, trunk, or branch sum to 100% for a given column. Significant odds ratios (OR) include (alive is reference): ORB:T dead (0.116); and ORR:T dead (0.359). See Table 3 for the sample size and full results.

Tree Size

Diameter at breast height (DBH)(5 ordinal categories) was found to be significantly associated with the tree failure part (CMH χ2 = 56.02; DF = 4; P < 0.0001) (Table 2) for trees in which a TCO had been attributed. Specifically, in large diameter trees (> 61 cm) there was a decreased proportion of trunk failures that resulted in TCOs. Meanwhile, the proportion of branch failures that resulted in a TCO increased steadily with DBH. Tree-caused outages due to root and branch failures were more evenly distributed across the ordinal diameter categories.

Tree heights were initially collected in 3 categories: 0 m to 15.2 m; 15.3 m to 30.5 m; and > 30.5 m. Due to the low number of trees that were > 30.5 m (n = 21), these were combined with the middle category and named > 15.2 m (Table 2). Tree height was found to be significantly associated with the tree failure part (χ2 =11.49; DF = 2; P = 0.0032; n = 831). The shorter trees category (0 m to 15.2 m) exhibited a higher proportion of TCOs due to trunk failure (59.4%, n = 183) compared to the taller tree category (> 15.2 m), which showed a greater frequency of TCOs due to branch failures (24.1%, n = 126).

Factors

The majority (82%) of TCOs investigated were on slopes of 0% to 10%. The most frequent tree failure part was the trunk (53.6%, n = 446), whether on relatively flat ground or on a substantial slope (CMH χ2 = 8.57; DF = 2; P < 0.0138; n = 832)(Table 2). However, a higher proportion of root failures resulted in TCOs (30.41%) on steeper slopes, complemented by a smaller proportion of branch failures (12.16%), when compared to the flat ground (25.45% and 22.66%, respectively).

Seasonal Factors

Tree-caused outages occurred throughout the year during the entire 32-month study period (Table 2).More TCOs were associated with the months of August and November, while months least likely to have outages due to trees were May and December (χ2 = 49.33; DF = 22; P = 0.0007; n = 844)(Table 2). The highest percentage of TCOs due to root failures were in May and December. Branch failures occurred most often in January and September. Trunk failures occurred most often in February, March, and August. During the study period, June 2020 through February 2023, the months with the highest average rainfall were June, August, and October (Table 1). Logistic fit of failure part regressed on the monthly average precipitation was not significant (χ2 = 0.16; DF = 2; P = 0.925; n = 844)(data not shown). In analysis where months were grouped into 4 quarters, in general, about half of the failures, observed similarly in each quarter of the year, were due to trunk failure (48% to 56%), while branch failures ranged from 17% to 25% and root failures ranged from 21% to 34%, reflected in nonsignificant Chi-square statistics (χ2 = 12.04; DF = 6; P = 0.06)(data not shown).

Generalized Logits Models

Generalized logits (multiple nominal logistic regression) enabled us to predict the influence of more than one factor and produce odds ratios mutually adjusted to each other. The variable selection was based on earlier empirical steps and was also limited by sample size restriction for all of the combinations of the categories.

Model 1 included the tree health and defect factors, as they represented indirect estimates of tree wellness prior to failure. Both health and defects were significantly related to afflicted tree part (Table 3). Health was influential for both types of logits, as both branch and root failures from dead trees were less likely to cause outages when compared to trunk failures among the live, declined, and dead trees. Specifically, trees that were alive when they caused the outage had 1/0.116 = 8.6 times the odds of branch failing as the trunk, compared to dead trees. In addition, the alive trees had 1/0.359 = 2.79 times the odds of root failing as the trunk, compared to dead trees (Table 3). Defects such as lean, unknown, and wound make a difference for the branch to trunk failures when compared to trees without any detected defect prior TCO (Table 3). Similarly, codominant, cracks, lean, unknown, and wound were less likely to have root outages compared to trunk when that rate was set against the “none”. Figure 4 also demonstrated that the root, trunk, and branch failures were the most uniformly distributed in the “none” category. Predicted probability of each tree part failing, in addition to observed probability, frequency, and residuals for all 18 combinations of tree health and defects are in Appendix Table S1.

View this table:
Table 3.

Odds ratios from final nominal logistic regression models (Model 1: health and defects; Model 2: genera and DBH). Variable with an asterisk (*) was defined as the reference, and odds ratios marked with † were significant. DBH (diameter at breast height); CI (confidence interval).

Model 2 focused on taxonomy and morphology of trees. The semifinal model had factors of genera, DBH, and height. Since the effect of height was not significant when adjusted for genera and DBH, the final Model 2 included only genera and DBH, and both were significant predictors of the tree failure part for the TCO (Table 3). Outages due to branch failures were more likely than trunk failures in maple (10.9 times), oak (6.7 times), yellow poplar (26 times), and others (7 times) when compared to the branch to trunk odds in pines (Table 3). Outages due to ash and yellow poplar were more likely when compared to root vs. trunk failures of pines. Outages due to branch failures were more likely than trunk failures when DBH was 45.8 cm to 61 cm and 61.1 cm to 91.4 cm compared to the smallest DBH class (0 cm to 30.6 cm) (Table 3). Predicted probability of each tree part failing, in addition to observed probability, frequency, and corresponding residuals for all 28 available combinations of tree genera and DBH categories are in Appendix Table S2.

Discussion

Utility vegetation management programs aim to prevent outages due to trees growing or falling into distribution lines, yet numerous outages continue to be caused by tree failures (Guggenmoos 2009; Walker and Dahle 2022). Understanding the details of tree failures—the tree part, size, genus, health, defect, site factors, and, to a lesser extent, time of year—is crucial in helping utility urban forest managers prescribe the correct treatment for prevention of tree failures. The analysis of distribution of species locally did not occur because of the broad swath of Virginia that the study area encompassed. Site conditions had a bias towards flatter sites; it is unclear whether this was because these outages were the easiest to investigate for the utility urban foresters or because more outages happen on lower-sloped areas.

This study is an industry-focused approach that utilizes post outage examinations that can be replicated by utilities. A distinct limitation of this study is that it is focused on trees that had already failed, rather than trees that did not, and thus does not address likelihood of failure. While this study makes use of a convenience sample, as not every outage was examined, nor were the approximately 1.9 million trees that did not fail during the study period, utilizing convenience sampling can be considered a reasonable approach as utilities do not have the resources (financial, employees, or time) to conduct landscape-level research with laboratorylevel rigor. We believe observational studies such as this provide value to the utility and utility forestry industries despite the lack of a controlled lab setting or small urban or suburban study area with landscaping or planted trees. Rather, this observational study was conducted in forested mountain areas, in private landowners’ small wood lots, and in swampy, hard-to-reach areas often only accessible by foot, among others.

Tree Part

In our study of ROWs in the Mid-Atlantic, fall-ins (trunk and root failures) were the most prevalent source of TCOs in this study, which is similar to previous research (Guggenmoos 2003; Guggenmoos and Sullivan 2007; Guggenmoos 2011). Hence to reduce outages, utilities might consider investing additional resources in identifying trees likely to fail at the roots or trunk, as they accounted for the vast majority of TCOs. Or perhaps, given the difficulty in determining the likelihood of failure, utilities could prioritize trees for targeted crown reductions or removal by using a combination of tree failure profiles, likelihood of impact, and the consequences of impact. For example, given this study, a utility in the Mid-Atlantic could decide to prioritize pine, due to high number of trunk and root failures, and ash, due to high number of trunk failures, for removal and/or crown reduction. While removal obviously mitigates all risk of a fall-in TCO, a crown reduction can also accomplish the desired risk mitigation if tree height is reduced to the extent that it cannot come into contact with the conductor or other electrical infrastructure. However, it is also relevant to note that most trees subject to fall-in from outside of the utility’s easement are not owned by the utility or even included in easement language, as they are owned privately.

Trunk failures occurred more frequently than expected across all genera but particularly in pines (66.9%) and ash (77.1%), which had the highest incidence of trunk failures (Table 2). Ash was not surprising, as emerald ash borer (EAB) had been present in Virginia for several years prior, and although the utility had used significant resources to remove ash trees next to their facilities, a limited amount remained. Root failures (ORR:T = 0.352) were less likely than trunk failures in ash trees (Table 3), which is consistent with studies that look at failures due to EAB (Persad et al. 2013; Persad et al. 2019). The subsequent failure of these trees was captured as part of this study, most often in February, March, and August, which may be related to ice loading in the winter/early spring and summer storms with wind loading in August.

Tree Genera

Pine was the most common genus that caused outages (30.1%). Trees grouped as Other (26.8%) were the second most common genus. This was not surprising given the diversity (over 100 species) of hardwood species within the 22 counties across central Virginia that were part of the study area (Virginia Department of Forestry [date unknown]). Oaks were the third most frequent cause of outages with 22.4%, followed by ash trees (8.3%). The observation of ash trees was surprising based on the amount of proactive work that had been done to eliminate risks associated with ash trees (Persad et al. 2013) by the utility and the prevalence of EAB infested ash trees throughout Virginia in the years prior to this study. However, that work was largely centered in areas that had high percentages of ash within the forest cover type. What was not surprising was the proportion of ash trees that failed at the trunk (77.1%) vs. roots (12.9%) or branch failure (10.0%). Virginia first saw EAB impacts in 2003; it was eradicated but resurfaced in 2008 and then rapidly spread over the entirety of Virginia (Buck et al. 2017).

Tree Defect

Arboricultural training often focuses on identifying defects which are presumed to lead to elevated likelihoods of failure and thus increase tree risk (Dunster et al. 2017; Goodfellow 2020; Lilly et al. 2022). Yet published studies have not found a correlation between defects and post storm failures (Koeser et al. 2020; Nelson et al. 2022; Koeser et al. 2023). In this study, obvious defects (wounds, codominant stems, cracks, and lean) were also found to be infrequent. Only 26.7% of TCOs were from trees with identified defects, while 46.6% of the trees had no visible defects. The ORR:T for defects were significantly lower when comparing roots to trunk failure. It may be that root defects were more difficult for investigators to detect. When comparing branches to trunks failures, only leans, wounds, and unknown were found to differ from no defect (none), and they were all less likely. While the limited connection between obvious defects might be surprising, it may be difficult to identify defects once the tree has failed and landed on the ground. In our study, these were recorded as Unknown and accounted for 26.7%. We do not intend to suggest that defects are not important; rather we feel that these results demonstrate that there is much to learn about why trees fail around utilities corridors.

When a tree defect was present, it was likely to be a wound (10.7%), followed by lean, codominant stems, and cracks. Cracks would be prevalent after a tree failure, but it would be difficult to determine if cracks were pre-existing or simply a result of the wood separating as the tree failed. Wounds would be the easiest defect to spot once a tree is on the ground, especially if the failure was a trunk failure and the cavity was obvious at the failure point. Utility urban forest researchers and managers should continue to investigate ways to identify these trees, whether it be manual inspection off the ROW or incorporating remote sensing (Rust and Stoinski 2022; Walker and Dahle 2022). That codominant branches were only 6% of the overall failures is surprising given the amount of research that has been conducted on this type of defect (MacDaniels 1932; Miller 1958; Eisner et al. 2002; Gilman 2003; Smiley 2003; Kane 2008; Kane and Clouston 2008; Slater and Harbinson 2010; Slater and Ennos 2013; Slater et al. 2014; Dahle et al. 2022; Eckenrode et al. 2022), yet most of the studies utilized mechanical load testing to induce failure rather than postfailure reports.

Tree Health

Dead trees accounted for 33.5% of the outages, and only 5.7% were due to impacts from branches falling from dead trees. It is likely that this low number is a result of management activities removing offending branches, or the branch failures did not lead to an outage due to missing the conductor or bouncing off the conductor. A comparison of dead to live trees (ORB:T = 0.116 and ORR:T = 0.359) found that trunk failures (at 75% of the dead trees) were the most common. While these failures are likely due to breakdown of connective tissue in the wood, it could be argued that these trees should likely have been readily identified and removed prior to tree failure and the resulting outage. This demonstrates the difficulty utility urban foresters have in managing a vast landscape of trees. Further, it highlights the importance of implementing the Utility Tree Risk Assessment BMPs (Goodfellow 2020) and developing new methods to identify trees with either an elevated likelihood of failure and/or likelihood of impact.

Tree Size

Many utilities focus on “enhanced tree trimming” (Parent et al. 2019; Dumarevskaya and Parent 2023), which can also be called “ground to sky” or sometimes “system hardening”. System hardening often includes additional engineering solutions such as sturdier and taller poles, overcurrent devices, lightning arrestors, strategic undergrounding, and various other strategies in addition to tree pruning and removal. An often-held belief among practitioners and vendors is that branch failure causes most outages. Consequently, the focus of enhanced tree trimming work is to remove overhanging limbs from portions of, or the entirety of, highly import distribution or subtransmission lines.

Yet we found that overall outages caused by branches were only 20.9% in this study, although there were some genera (maples, oaks, yellow poplar, and other) with higher incidents of branch failures. The ORB:T for these 4 genera ranged between 7.3 and 26 compared to pines. These were likely due to species specific characteristics such as the propensity of oak and yellow poplar to self-prune lower branches as they age. Maple limbs may be more susceptible to ice and wind failure.

Midsized diameter trees caused more outages compared to the smallest trees (ORB:T = 1.867 for 45.8 cm to 61.0 cm and ORB:T = 3.92 for 61.1 cm to 91.4 cm). Branch failures were more prevalent in tall trees (25.0%), those greater than 15.3 m. Certainly, branch failures from trees that are shorter than the utility line (often 9 m to 14 m) are not likely to result in an outage. Taller trees will have larger branches that are above the powerlines and thus have a greater chance of impacting the powerlines. Indeed, this is the basis for specifications for utility vegetation management contracts; these may be species-specific and state that softwoods should have all overhanging branches removed from subtransmission or high impact feeders to reduce the impact from branch failures. Yet, the results from this study do not support the adoption of enhanced tree trimming for the region of this study, as only 5.7% of outages due to pine failure were caused by branches.

Seasonal Factors

More failure investigations occurred in January, August, and November. This may have been due to dry conditions enabling easy access to the failed trees but was probably more likely due to a lighter workload for the utility urban foresters and the ability to make time to conduct the assessments. We recommend that future researchers consider concentrating inspections to specific time frames such as seasons that match weather patterns. Furthermore, whole tree failures are more likely to occur after soils become saturated, which varies depending on study year and soil drainage class, so extending the time frame and geographic range of future studies may capture and account for weather, wind, and other meteorological events.

Tree-caused outage terminology is not standardized and varies from utility to utility, including how facilities are labeled. We have determined by talking to colleagues throughout the industry that terminology, collection methods, and practices vary significantly. In order to replicate research, a national effort should be undertaken to standardize terminology and reporting practices.

Conclusions

The main objective of vegetation management at electric utilities is to prevent outages. This study found that removing dead trees and identifying and mitigating live trees with a profile that lead to TCOs should be a key component of crafting a high-performing utility vegetation management programs. Additionally, utility foresters should develop prescriptive management that focuses on preventing outages from branches growing or blowing into the conductors. Utilities should consider including a focus on standing live trees in addition to the limited sample size of the trees that failed, perhaps utilizing a standardized model such as developed by Koeser et al. (2025). Additionally, the incorporation of machine learning may provide additional analytical power and assist utility urban forest managers in crafting an adaptive approach to program management (Walker and Dahle 2023; Salisbury et al. 2025).

This study highlights two factors as potentially significant changes for industry operations. First, the profile for many of the TCOs in this study included live trees failing at the trunk. Management of trunk failure will require a paradigm shift in how the industry approaches preventative vegetation maintenance. Second, the incidence of branch failure contrasted with the relatively high costs of “ground to sky” or “circuit hardening” tree pruning programs may not make sense. In the future, utilities that want to decrease TCO occurrences may need craft programs to aggressively target live tree failures in the zone between the substation and the first protective device. Furthermore, while regional studies of TCOs are useful, a national approach may provide added value to the utility urban forestry industry.

Conflicts of Interest

The authors reported no conflicts of interest.

Appendix

View this table:
Table S1.

Maximum likelihood predicted values for branch, root, and trunk failures from generalized logit model main effects using tree health and defects. SE (standard error).

View this table:
Table S2.

Maximum likelihood predicted values for branch, root, and trunk failures from the generalized logit model main effects using genera and DBH. DBH (diameter at breast height); SE (standard error).

Acknowledgements

The authors wish to thank the foresters of REC, the Analytics team of REC, John Crawford, and John Hewa for their enthusiasm and support of this project. This work was supported by the TREE Fund Utility Arborist Research Fund #20-500U, the USDA National Institute of Food and Agriculture McIntire Stennis project WVA00819, and the West Virginia Agricultural and Forestry Experiment Station.

Literature Cited

  1. ↵
    Apostolov A, Oke J, Suttle R, Arwade S, Kane B. 2023. Predicting tree failure likelihood for utility risk mitigation via a convolutional neural network. Sustainable and Resilient Infrastructure. 8(6):572–588. https://doi.org/10.1080/23789689.2023.2233759
    Google ScholarOpenURL
  2. ↵
    Buck J, Parra G, Lance D, Reardon R, Binion D. 2017. 2016 Emerald Ash Borer National Research and Technology Development Meeting. 2016 October 19-20. Wooster (OH, USA): Ohio Agricultural Research and Development Center. FHTET-2016-10. [Accessed 2025 December 31]. 126 p. https://www.fs.usda.gov/foresthealth/technology/pdfs/FHTET-2016-10_EAB_NRTDM.pdf
    Google Scholar
  3. ↵
    Coder KD. 2010. Root strength & tree anchorage. Athens (GA, USA): Warnell School of Forestry & Natural Resources, University of Georgia. [Accessed 2025 December 31]. 88 p. https://openscholar.uga.edu/record/24836
    Google Scholar
  4. ↵
    Coder KD. 2022a. Ice storm impacts on trees: Prioritized causes of damage. Athens (GA, USA): Warnell School of Forestry & Natural Resources, University of Georgia. 29 p. https://bugwoodcloud.org/resource/files/25237.pdf
    Google Scholar
  5. ↵
    Coder KD. 2022b. Trees, sites & ice storms: Attributes leading to tree damage, failure, & mortality. Athens (GA, USA): Warnell School of Forestry & Natural Resources, University of Georgia. 42 p. https://bugwoodcloud.org/resource/files/25367.pdf
    Google Scholar
  6. ↵
    Cucchi V, Meredieu C, Stokes A, Berthier S, Bert D, Mohamed N, Denis A, Lastennet R. 2004. Root anchorage of inner and edge trees in stands of maritime pine (Pinus pinaster Ait.) growing in different podzolic soil conditions. Trees. 18:460–466. https://doi.org/10.1007/s00468-004-0330-2
    Google ScholarOpenURL
  7. ↵
    Dahle G, Grabosky J, Kane B, Miesbauer J, Peterson W, Telewski FW, Koeser A, Watson GW. 2014. Tree biomechanics: A white paper from the 2012 international meeting and research summit at The Morton Arboretum (Lisle, Illinois, U.S.). Arboriculture & Urban Forestry. 40(6):309–318. https://doi.org/10.48044/jauf.2014.029
    Google ScholarOpenURL
  8. ↵
    Dahle GA, Eckenrode RT, Smiley ET, DeVallance D, Holásková I. 2022. Can mechanical strain and aspect ratio be used to determine codominance in red maple unions without included bark? Forests. 13(7):1007. https://doi.org/10.3390/f13071007
    Google ScholarOpenURL
  9. ↵
    Dahle GA, Holt HH, Chaney WR, Whalen TM, Cassens DL, Gazo R, McKenzie RL. 2006. Branch strength loss implications for silver maple (Acer saccharinum) converted from round-over to V-trim. Arboriculture & Urban Forestry. 32(4):148–154. https://doi.org/10.48044/jauf.2006.020
    Google ScholarOpenURL
  10. ↵
    Dorval AD, Meredieu C, Danjon F. 2016. Anchorage failure of young trees in sandy soils is prevented by a rigid central part of the root system with various designs. Annals of Botany. 118(4):747–762. https://doi.org/10.1093/aob/mcw098
    Google ScholarOpenURLCrossRefPubMed
  11. ↵
    Dumarevskaya L, Parent JR. 2023. Electric grid resilience: The effects of conductor coverings, enhanced tree trimming, and line characteristics on tree-related power outages. Electric Power System Resilience. 221:109454. https://doi.org/10.1016/j.epsr.2023.109454
    Google ScholarOpenURL
  12. ↵
    Dunster JA, Smiley ET, Matheny N, Lilly S. 2017. Tree risk assessment manual. 2nd Ed. Atlanta (GA, USA): International Society of Arboriculture. p. 194.
    Google Scholar
  13. ↵
    Dupuy L, Fourcaud T, Stokes A. 2005. A numerical investigation into the influence of soil type and root architecture on tree anchorage. Plant Soil. 278:119–134. https://doi.org/10.1007/s11104-005-7577-2
    Google ScholarOpenURLCrossRefWeb of Science
  14. ↵
    Eckenrode RT IV, Dahle GA, Smiley ET, DeVallance D, Cynk-Dahle I, Walker M. 2022. Examining strain propagation in the branch unions of Quercus alba L. and Quercus montana Willd. Urban Forestry & Urban Greening. 73:127591. https://doi.org/10.1016/j.ufug.2022.127591
    Google ScholarOpenURL
  15. ↵
    Eisner NJ, Gilman EF, Grabosky JC. 2002. Branch morphology impacts compartmentalization of pruning wounds. Journal of Arboriculture. 28(2):99–105. https://doi.org/10.48044/jauf.2002.013
    Google ScholarOpenURL
  16. ↵
    Ennos AR. 1993. The scaling of root anchorage. Journal of Theoretical Biology. 161(1):61–75. https://doi.org/10.1006/jtbi.1993.1040
    Google ScholarOpenURLCrossRefWeb of Science
  17. ↵
    Executive Office of the President. 2013. Economic benefits of increasing electric grid resilience to weather outages. Washington (DC, USA): President’s Council of Economic Advisers, US Department of Energy’s Office of Electricity Delivery and Energy Reliability, White House Office of Science and Technology. 28 p. https://www.energy.gov/sites/prod/files/2013/08/f2/Grid%20Resiliency%20Report_FINAL.pdf
    Google Scholar
  18. ↵
    Gibbons P, Cunningham RB, Lindenmayer DB. 2008. What factors influence the collapse of trees retained on logged sites?: A case-control study. Forest Ecology and Management. 255(1): 62–67. https://doi.org/10.1016/j.foreco.2007.08.020
    Google ScholarOpenURL
  19. ↵
    Gilman EF. 2003. Branch-to-stem diameter ratio affects strength of attachment. Journal of Arboriculture. 29(5):291–294. https://doi.org/10.48044/jauf.2003.033
    Google ScholarOpenURL
  20. ↵
    Goodfellow JW. 2020. Utility tree risk assessment. Atlanta (GA, USA). Best management practices. International Society of Arboriculture. 95 p.
    Google Scholar
  21. ↵
    Graziano M, Gunther P, Gallaher A, Carstensen FV, Becker B. 2020. The wider regional benefits of power grids improved resilience through tree-trimming operations evidences from Connecticut, USA. Energy Policy. 138:111293. https://doi.org/10.1016/j.enpol.2020.111293
    Google ScholarOpenURL
  22. ↵
    Guggenmoos S. 2003. Effects of tree mortality on power line security. Journal of Arboriculture. 29(4):181–196. https://doi.org/10.48044/jauf.2003.022
    Google ScholarOpenURL
  23. ↵
    Guggenmoos S. 2007. Storm hardening the electric system against tree-caused service interruptions. Utility Arborist Newsline. 2:1–5. [Accessed 2025 December 31]. https://www.ecosync.com/tdworld/PSE%20Storm%20Hardening.pdf
    Google ScholarOpenURL
  24. ↵
    Guggenmoos S. 2009. Investigation into storm hardening the electric transmission system for Puget Sound Energy. Produced from data from Puget Sound Energy. Bellevue (WA, USA): Puget Holdings LLC.
    Google Scholar
  25. ↵
    Guggenmoos S. 2011. Tree-related electric outages due to wind loading. Arboriculture & Urban Forestry. 37(4):147–151. https://doi.org/10.48044/jauf.2011.019
    Google ScholarOpenURL
  26. ↵
    Guggenmoos S, Sullivan TE. 2007. Outside right-of-way tree risk along electrical transmission lines. Utility Arborist Association. 6 p. https://www.ecosync.com/tdworld/SideTreeRisk.pdf
    Google Scholar
  27. ↵
    Guggenmoos S, Sullivan TE. 2012. Outside right-of-way tree risk along electrical transmission lines. In: Evans JM, Reinemann J, editors. 9th International Symposium: Environmental Concerns in Rights of-Way Management. 2009 September 27-30. 617 p.
    Google Scholar
  28. ↵
    Guikema SD, Davidson RA, Liu H. 2006. Statistical models of the effects of tree trimming on power system outages. IEEE Transactions on Power Delivery. 21(3):1549–1557. https://doi.org/10.1109/TPWRD.2005.860238
    Google ScholarOpenURL
  29. ↵
    Hamilton RS, Zaff D, Niziol T. 2006. A catastrophic lake effect snow storm over Buffalo, NY October 12-13, 2006. Buffalo (NY, USA): NOAA National Weather Service. 7 p.
    Google Scholar
  30. ↵
    Jodas DS, Brazolin S, Velasco GDN, de Lima RA, Yojo T, Papa JP. 2024. Urban tree failure probability prediction based on dendrometric aspects and machine learning models. Computers, Environment and Urban Systems. 108:102074. https://doi.org/10.1016/j.compenvurbsys.2024.102074
    Google ScholarOpenURL
  31. ↵
    Kabir E, Guikema S, Kane B. 2018. Statistical modeling of tree failures during storms. Reliability Engineering & System Safety. 177:68–79. https://doi.org/10.1016/j.ress.2018.04.026
    Google ScholarOpenURL
  32. ↵
    Kane B. 2008. Tree failure following a windstorm in Brewster, Massachusetts, USA. Urban Forestry & Urban Greening. 7(1):15–23. https://doi.org/10.1016/j.ufug.2007.11.001
    Google ScholarOpenURL
  33. ↵
    Kane B, Clouston P. 2008. Tree pulling tests of large shade trees in the genus Acer. Arboriculture & Urban Forestry. 34(2): 101–109. https://doi.org/10.48044/jauf.2008.013
    Google ScholarOpenURL
  34. ↵
    Kane B, Farrell R, Zedaker SM, Lofersky JR, Smith DW. 2008. Failure mode and prediction of the strength of branch attachments. Arboriculture & Urban Forestry. 34(5):308–316. https://doi.org/10.48044/jauf.2008.042
    Google ScholarOpenURL
  35. ↵
    Kane B, Finn JT. 2014. Factors affecting branch failures in open grown trees during a snowstorm in Massachusetts, USA. SpringerPlus. 3:720. https://doi.org/10.1186/2193-1801-3-720
    Google ScholarOpenURLPubMed
  36. ↵
    Kim Y, Rahardjo H, Tsen-Tieng DL. 2020. Stability analysis of laterally loaded trees based on tree-root-soil interaction. Urban Forestry & Urban Greening. 49:126639. https://doi.org/10.1016/j.ufug.2020.126639
    Google ScholarOpenURL
  37. ↵
    Klein RW, Koeser AK, Kane B, Landry SM, Shields H, Lloyd S, Hansen G. 2020. Evaluating the likelihood of tree failure in Naples, Florida (United States) following Hurricane Irma. Forests. 11(5):485. https://doi.org/10.3390/f11050485
    Google ScholarOpenURL
  38. ↵
    Koeser AK, Klein RW, Hauer RJ, Miesbauer JW, Freeman Z, Harchick C, Kane B. 2023. Defective or just different? Observed storm failure in four urban tree growth patterns. Forests. 14(5):988. https://doi.org/10.3390/f14050988
    Google ScholarOpenURL
  39. ↵
    Koeser AK, Salisbury A, Andreu M, Hauer RJ, Hilbert DR, Kane B, Klein R, Landry S, Martin AJF, Miesbauer JW, Rockwell CA, Staudhammer C. 2025. Standardizing pre-and post-storm data collection for urban forestry research. Arboriculture & Urban Forestry. 51(5):541–554. https://doi.org/10.48044/jauf.2025.024
    Google ScholarOpenURL
  40. ↵
    Koeser AK, Smiley ET, Hauer RJ, Kane B, Klein RW, Landry SM, Sherwood M. 2020. Can professionals gauge likelihood of failure?—Insights from tropical storm Matthew. Urban Forestry & Urban Greening. 52:126701. https://doi.org/10.1016/j.ufug.2020.126701
    Google ScholarOpenURL
  41. ↵
    Kurkela T, Aalto T, Varama M, Jalkanen R. 2005. Defoliation by the common pine sawfly (Diprion pini) and subsequent growth reduction in Scots pine: A retrospective approach. Silva Fennica. 39(4):360. https://doi.org/10.14214/sf.360
    Google ScholarOpenURL
  42. ↵
    Lilly S, Sydnor TD. 1995. Comparison of branch failure during static loading of silver and Norway maples. Journal of Arboriculture. 21(6):302–305. https://doi.org/10.48044/jauf.1995.049
    Google ScholarOpenURL
  43. ↵
    Lilly SJ, Bassett CG, Komen J, Purcell L. 2022. Arborists ‘certification study guide. 4th Ed. Atlanta (GA, USA): International Society of Arboriculture. 468 p.
    Google Scholar
  44. ↵
    Lott N, Ross T. 2005. Tracking and evaluating U.S. billion dollar weather disasters, 1980-2005. Asheville (NC, USA): NOAA’s National Climatic Data Center. 6 p. https://www.ncdc.noaa.gov/monitoring-content/billions/docs/lott-and-ross-2006.pdf
    Google Scholar
  45. ↵
    Lu W, Yu X, Jia G. 2020. Retrospective analysis of tree decline based on intrinsic water-use efficiency in semi-arid areas of North China. Atmosphere. 11(6):577. https://doi.org/10.3390/atmosll060577
    Google ScholarOpenURL
  46. ↵
    Lundström T, Jonas T, Stöckli V, Ammann W. 2007. Anchorage of mature conifers: Resistive turning moment, root-soil plate geometry and root growth orientation. Tree Physiology. 27(9): 1217–1227. https://doi.org/10.1093/treephys/27.9.1217
    Google ScholarOpenURLCrossRefPubMedWeb of Science
  47. ↵
    MacDaniels LH. 1932. Factors affecting the breaking strength of apple tree crotches. Journal of the American Society for Horticultural Science. 29:44.
    Google ScholarOpenURL
  48. ↵
    Miller RH. 1958. Morphology of Humulus lupulus. I. Developmental anatomy of the primary root. American Journal of Botany. 45(5):418–431. https://doi.org/10.2307/2439643
    Google ScholarOpenURLCrossRefWeb of Science
  49. ↵
    National Oceanic and Atmospheric Administration (NOAA). 2026. U.S. Climate Normals Quick Access. Ashland, VA, 2006-2020. Washington (DC, USA): National Centers for Environmental Information. [Accessed 2025 December 31]. https://www.ncei.noaa.gov/access/us-climate-normals/#dataset=normals-monthly&timeframe=15&location=VA&station=USC00440327
    Google Scholar
  50. ↵
    Nelson MF, Klein RW, Koeser AK, Landry SM, Kane B. 2022. The impact of visual defects and neighboring trees on wind-related tree failures. Forests. 13(7):978. https://doi.org/10.3390/f13070978
    Google ScholarOpenURL
  51. ↵
    Nowak CA, Ballard BD. 2005. A framework for applying integrated vegetation management on rights-of-way. Arboriculture & Urban Forestry. 31(1):28–37. https://doi.org/10.48044/jauf.2005.004
    Google ScholarOpenURL
  52. ↵
    Parent JR, Meyer TH, Volin JC, Fahey RT, Witharana C. 2019. An analysis of enhanced tree trimming effectiveness on reducing power outages. Journal of Environmental Management. 241:397–406. https://doi.org/10.1016/j.jenvman.2019.04.027
    Google ScholarOpenURLPubMed
  53. ↵
    Persad A, Dahle GA, DeVallance D, Rocha OJ, Grabosky J. 2019. Optical, acoustical, and fine root analyses of emerald ash borer infested ash trees. Arboriculture & Urban Forestry. 45(5):211–220. https://doi.org/10.48044/jauf.2019.017
    Google ScholarOpenURL
  54. ↵
    Persad AB, Siefer J, Montan R, Kirby S, Rocha OJ, Redding Mem, Ranger CM, Jones AW. 2013. Effects of emerald ash borer infestation on the structure and material properties of ash trees. Arboriculture & Urban Forestry. 39(1):11–16. https://doi.org/10.48044/jauf.2013.003
    Google ScholarOpenURL
  55. ↵
    Poulos HM, Camp AE. 2010. Decision support for mitigating the risk of tree induced transmission line failure in utility rights-of-way. Journal of Environmental Management. 45:217–226. https://doi.org/10.1007/s00267-009-9422-5
    Google ScholarOpenURL
  56. ↵
    Rachmadiyanto AN, Azzahra NF, Sudiana E, Wahyuni S. 2024. Tree health analysis of old Vatica pauciflora (Korth.) blume in Bogor Botanical Gardens using forest health monitoring method. IOP Conference Series: Earth and Environmental Science. 1433(1):012012. https://doi.org/10.1088/1755-1315/1433/1/012012
    Google ScholarOpenURL
  57. ↵
    Radmer DT, Kuntz PA, Christie RD, Venkata SS, Fletcher RH. 2002. Predicting vegetation-related failure rates for overhead distribution feeders. IEEE Transactions on Power Delivery. 17(4):1170–1175. https://doi.org/10.1109/TPWRD.2002.804006
    Google ScholarOpenURL
  58. ↵
    Reed DA. 2008. Electric utility distribution analysis for extreme winds. Journal of Wind Engineering and Industrial Aerodynamics. 96(1):123–140. https://doi.org/10.1016/j.jweia.2007.04.002
    Google ScholarOpenURL
  59. ↵
    Rust S, Stoinski B. 2022. Using artificial intelligence to assist tree risk assessment. Arboriculture & Urban Forestry. 48(2): 138–146. https://doi.org/10.48044/jauf.2022.011
    Google ScholarOpenURL
  60. ↵
    Saavedra-Romero LL, Alvarado-Rosales D. 2024. Tree health assessment in an institutional green space: INAOE case study. Agrociencia. 58–3. https://doi.org/10.47163/agrociencia.v58i3.3034
    Google Scholar
  61. ↵
    Sakals ME, Sidle RC. 2004. A spatial and temporal model of root cohesion in forest soils. Canadian Journal of Forest Research. 34(4):950–958. https://doi.org/10.1139/x03-268
    Google ScholarOpenURL
  62. ↵
    Salisbury AB, Koeser AK, Andreu MG, Chen Y, Freeman Z, Miesbauer JW, Herrera-Montes A, Kua CS, Nukina RH, Rockwell C, Shibata S, Thorn H, Wan B, Hauer RJ. 2025. Expanding a hurricane wind resistance rating system for tree species using machine learning. Arboriculture & Urban Forestry. 51(2):128–153. https://doi.org/10.48044/jauf.2025.002
    Google ScholarOpenURL
  63. ↵
    Santos AS, Faria LT, Boschi LS, Lopes MLM, Minussi CR. 2024. Estimation of vulnerable areas to faults caused by tree vegetation in power distribution systems. IEEE Acces s. 12:48472–48484. https://doi.org/10.1109/ACCESS.2024.3382943
    Google ScholarOpenURL
  64. ↵
    Simpson P, Van Bossuyt R. 1996. Tree-caused electric outages. Journal of Arboriculture. 22(3):117–121. https://doi.org/10.48044/jauf.1996.018
    Google ScholarOpenURL
  65. ↵
    Slater D, Bradley RS, Withers PJ, Ennos RA. 2014. The anatomy and grain pattern in forks of hazel (Corylus avellana L.) and other tree species. Trees. 28:1437–1448. https://doi.org/10.1007/s00468-014-1047-5
    Google ScholarOpenURL
  66. ↵
    Slater D, Ennos AR. 2013. Determining the mechanical properties of hazel forks by testing their component parts. Trees. 27: 1515–1524. https://doi.org/10.1007/s00468-013-0898-5
    Google ScholarOpenURL
  67. ↵
    Slater D, Harbinson C. 2010. Towards a new model of branch attachment. Arboricultural Journal. 33(2):95–105. https://doi.org/10.1080/03071375.2010.9747599
    Google ScholarOpenURL
  68. ↵
    Smiley ET. 2003. Does included bark reduce the strength of codominant stems? Journal of Arboriculture. 29(2):104–106. https://doi.org/10.48044/jauf.2003.013
    Google ScholarOpenURL
  69. ↵
    Stokes A, Drexhage M, Guitard D. 2000. A method for predicting the possible site of failure in trees during mechanical loading. In: Stokes A, editor. The supporting roots of trees and woody plants: Form, function and physiology. 1st Ed. Vol. 87. Dordrecht (Germany): Springer. p. 279–285. https://doi.org/10.1007/978-94-017-3469-1_27
    Google ScholarOpenURL
  70. ↵
    Stokes A, Ghani MA, Salin F, Danjon F, Jeannin H, Berthier S, Kokutse AD, Frochot H. 2007. Root morphology and strain distribution during tree failure on mountain slopes. In: Stokes A, Spanos I, Norris JE, Cammeraat E, editors. Eco-and ground bio-engineering: The use of vegetation to improve slope stability. Proceedings of the First International Conference on Eco-Engineering; 2004 September 13-17; Thessaloniki, Greece. Developments in Plant and Soil Sciences. Vol. 103. Dordrecht (Germany): Springer. p. 165–173. https://doi.org/10.1007/978-1-4020-5593-5_16
    Google ScholarOpenURL
  71. ↵
    Stokes A, Mine FX, Mao Z, Brancheriau L. 2012a. Multi-stemming and mechanical traits ensure persistence of subalpine woody plants exposed to a disturbance gradient. Journal of Vegetation Science. 23(2):325–338. https://doi.oıg/10.1111/j.1654-1103.2011.01349.x
    Google ScholarOpenURL
  72. ↵
    Stokes ME, Davis CS, Koch GG. 2012b. Categorical data analysis using SAS®. 3rd Ed. Cary (NC, USA): SAS Institute Inc. 590 p.
    Google Scholar
  73. ↵
    Suttle R, Kane B, Bloniarz D. 2022. Comparing the structure, function, value, and risk of managed and unmanaged trees along rights-of-way and streets in Massachusetts. Forests. 13(10):1602. https://doi.org/10.3390/f13101602
    Google ScholarOpenURL
  74. ↵
    Taylor WO, Watson PL, Cerrai D, Anagnostou E. 2022. A statistical framework for evaluating the effectiveness of vegetation management in reducing power outages caused during storms in distribution networks. Sustainability. 14(2):904. https://doi.org/10.3390/su14020904
    Google ScholarOpenURL
  75. ↵
    Tsen-Tieng DL, Rahardjo H, Choon LE, King FY. 2018. Anchorage and stability of tree rootsoil plates. Environmental Geotechnics. 7(5):330–337. https://doi.org/10.1680/jenge.17.00007
    Google ScholarOpenURL
  76. ↵
    University of Florida IFAS. 2015. Cracks in the trunk. Gainesville (FL, USA): University of Florida Institute of Food and Agricultural Sciences. https://hort.ifas.ufl.edu/woody/potential-failure.shtml
    Google Scholar
  77. ↵
    van Haaften M, Liu Y, Wang Y, Zhang Y, Gardebroek C, Heijman W, Meuwissen M. 2021. Understanding tree failure—A systematic review and meta-analysis. PLoS One. 16(2):e0246805. https://doi.org/10.1371/journal.pone.0246805
    Google ScholarOpenURLPubMed
  78. ↵
    Virginia Department of Forestry. [date unknown]. Virginia’s forest history. Charlottesville (VA, USA): Virginia Department of Forestry. [Accessed 2024 December 1]. https://dof.virginia.gov/forest-utilization-sustainability/learn-about-forest-utilization-sustainability/virginias-forest-history
    Google Scholar
  79. ↵
    Walker M, Dahle GA. 2022. Likelihood of failure of trees along electrical utility rights-of-way: A literature review. Arboriculture & Urban Forestry. 48(4):242–257. https://doi.org/10.48044/jauf.2022.018
    Google ScholarOpenURL
  80. ↵
    Walker M, Dahle GA. 2023. Literature review of unmanned aerial systems and LIDAR with application to distribution utility vegetation management. Arboriculture & Urban Forestry. 49(3):144–156. https://doi.org/10.48044/jauf.2023.011
    Google ScholarOpenURL
  81. ↵
    Wanik DW, Anagnostou EN, Astitha M, Hartman BM, Lackmann GM, Yang J, Cerrai D, He J, Frediani MEB. 2018. A case study on power outage impacts from future Hurricane Sandy scenarios. Journal of Applied Meteorology and Climatology. 57:51–79. https://doi.org/10.1175/JAMC-D-16-0408.1
    Google ScholarOpenURL
  82. ↵
    Wei Z, Halik Ü, Aishan T, Abliz A, Welp M. 2022. Spatial distribution patterns of trunk internal decay of Euphrates poplar riparian forest along the Tarim River, northwest China. Forest Ecology and Management. 522:120434. https://doi.org/10.1016/j.foreco.2022.120434
    Google ScholarOpenURL
Loading
Loading
Loading
PDF
Back to top

In this Issue

Arboriculture & Urban Forestry (AUF)

Arboriculture & Urban Forestry (AUF)

Vol. 52, Issue 4

July 2026

Table of Contents Table of Contents (PDF) Index by author

Jump to section

Keywords

Electric Utilities, Power Outages, Reliability, Tree-Caused Outages, Vegetation Management