Regional patterns of declining butternut (Juglans cinerea L.) suggest site characteristics for restoration

Abstract Butternut trees dying from a canker disease were first reported in southwestern Wisconsin in 1967. Since then, the disease has caused extensive mortality of butternut throughout its North American range. The objectives of this study were to quantify changes in butternut populations and density across its range and identify habitat characteristics of sites where butternut is surviving in order to locate regions for potential butternut restoration. The natural range of butternut (Juglans cinerea L.) extends over a large region of eastern N. America encompassing New Brunswick south to North Carolina, north to Minnesota, and southwest to Missouri. Despite the species’ large range, it is typically not a common tree, comprising a relatively minor component of several different forest types. We evaluated change in butternut abundance and volume from current and historic data from 21 states in the eastern United States. We related abundance and volume at two time periods to a suite of ecological and site factors in order to characterize site conditions where butternut survived. We also assessed the current level of butternut mortality across its range. Since the 1980s, the number of butternut trees and butternut volume have decreased by 58% and 44%, respectively, across its US range. Substantial relative decreases in tree numbers and volume occurred in most ecoregion sections. Five environmental variables were found to be significant predictors of butternut presence. The potential impacts of butternut canker are particularly acute as the canker pathogen invasion pushes a rare tree species toward extinction, at least at a local scale. Based on the results presented here, large‐diameter maple/beech/birch stands in dry, upland sites in eastern Minnesota, western Wisconsin, and upstate New York appear to offer the most favorable conditions for butternut growth and survival and thus may be the best stands for planting resistant butternut trees.

tify changes in butternut populations and density across its range and identify habitat characteristics of sites where butternut is surviving in order to locate regions for potential butternut restoration. The natural range of butternut (Juglans cinerea L.) extends over a large region of eastern N. America encompassing New Brunswick south to North Carolina, north to Minnesota, and southwest to Missouri. Despite the species' large range, it is typically not a common tree, comprising a relatively minor component of several different forest types. We evaluated change in butternut abundance and volume from current and historic data from 21 states in the eastern United States.
We related abundance and volume at two time periods to a suite of ecological and site factors in order to characterize site conditions where butternut survived. We also assessed the current level of butternut mortality across its range. Since the 1980s, the number of butternut trees and butternut volume have decreased by 58% and 44%, respectively, across its US range. Substantial relative decreases in tree numbers and volume occurred in most ecoregion sections. Five environmental variables were found to be significant predictors of butternut presence. The potential impacts of butternut canker are particularly acute as the canker pathogen invasion pushes a rare tree species toward extinction, at least at a local scale. Based on the results presented here, large-diameter maple/beech/birch stands in dry, upland sites in eastern Minnesota, western Wisconsin, and upstate New York appear to offer the most favorable conditions for butternut growth and survival and thus may be the best stands for planting resistant butternut trees.

K E Y W O R D S
butternut, Juglans cinerea, Ophiognomonia clavigignenti-juglandacearum, plant disease, restoration, species distributions

| INTRODUCTION
For over two centuries, North American butternut (Juglans cinerea L.) was cherished for its exceptional wood properties and was sought after for the manufacture of fine furniture, musical instruments, and boats (Woeste & Pijut, 2009). The species was also valued for its sweet, oily nuts that were desired by both Native Americans and European settlers and are also a source of large mast utilized by various wildlife species. The natural range of butternut extends over a large region encompassing New Brunswick south to North Carolina, north to Minnesota, and southwest to Missouri. Isolated, scattered butternut also occur in Arkansas, Mississippi, Alabama, Georgia, and South Carolina (Little, 1971;Figure 1a). Despite the species' large range, it is typically not a common tree, comprising a relatively minor component of several different forest types.
Butternut trees dying from a canker disease were first reported in southwestern Wisconsin in 1967. Since then, extensive mortality of butternut of all ages has been observed throughout its North American range (Ostry, 1998a;Ostry & Woeste, 2004). This disease is caused by the fungal pathogen Ophiognomonia clavigignenti-juglandacearum (syn. Sirococcus clavigignenti-juglandacearum -Broders & Boland, 2011) which is non-native in North America and possibly F I G U R E 1 (a) Butternut basal area per acre of forest land (Wilson, Lister, Riemann, & Griffith, 2013), historic range of butternut (Little, 1971), and ecoregion section boundaries in the eastern United States, (b) FIA plot locations with butternut present at time 1 and time 2 (plot locations are approximate) introduced from Asia (Furnier, Stolz, Mustaphi, & Ostry, 1999;Nair, 1998;Ostry, 1998b;Ostry, Mielke, & Skilling, 1994 Health Protection program estimated that 77% percent of the butternut trees in North Carolina and Virginia had been killed, and in the northeastern U.S., most of the monitored butternut trees were affected by butternut canker (U.S. Forest Service, 2005). A variety of different insect species are capable of carrying spores of the pathogen thus possibly explaining its widespread distribution (Halik & Bergdahl, 2002;Katovich & Ostry, 1998), and spores can remain viable for long periods (Moore & Ostry, 2015). The extent of the disease is so great that butternut is increasingly rare and is considered an imperiled species in many states (Woeste & Pijut, 2009).
Butternut is listed as a "species of concern" or a "sensitive species" in several states and is a Regional Forester Sensitive Species on 13 of the 16 National Forests in the Eastern Region of the US Forest Service. Butternut is listed as endangered in Canada.
Initial genetic analysis of O. clavigignenti-juglandacearum from several North American locations indicated that it may have been introduced into North America as a single strain (Furnier et al., 1999).
Recent evidence supports an introduction of three genetic clusters of the fungus and only asexual reproduction (Broders & Boland, 2011).
This finding provides some optimism that resistance in butternut might eventually be found and may not be quickly overcome by the development of increasingly pathogenic races of the pathogen.
Several studies have evaluated the genetic structure and diversity of butternut with results than can guide sound management of a declining species. For example, a study by Ross-Davis, Ostry, and Woeste (2008) reported that butternut retains a large amount of genetic diversity that is higher than previously estimated, and Boraks and Broders (2016) found significant gene flow among butternut populations in the northeast. Parks, Jenkins, Ostry, Zhao, and Woeste (2014) also found that genetic diversity in 19 watersheds in the Great Smoky Mountains National Park was evenly distributed with high mean heterozygosity although there was variability in some subpopulations due to hybridization with Japanese walnut (Juglans ailantifolia). Hybridization has been reported to occur across a large portion of the range of butternut (Hoban, McCleary, Schlarbaum, & Romero-Severson, 2009), and hybrids are virtually indistinguishable from true butternut, so it is possible that some trees recorded during the forest inventories reported here could be hybrids.
There has been good progress in the methods for identifying resistant butternut, collecting germplasm, and hybridizing butternut and Japanese walnut (Michler et al., 2005;Ostry & Moore, 2008;Woeste & Pijut, 2009). Given the progress in identifying potentially resistant trees, careful consideration is being given to the selection of locations where reintroductions are most likely to be successful (Woeste, Farlee, Ostry, McKenna, & Weeks, 2009). A predictive model was developed to identify potential sites for butternut restoration in Mammoth Cave National Park (Thompson, Van Manen, Schlarbaum, & DePoy, 2006), but butternut restoration has not been studied across its historic range.
The characteristics of the forest lands where surviving butternut trees remain can provide guidance for where restoration efforts should be focused. Butternut exhibits its best growth on well-drained soils and streambanks and is rarely found on infertile, compact, or dry soils. Additionally, the species is most often present in coves, stream benches, slopes, and other sites with good drainage up to elevations of about 1,500 meters. The most common associated tree species include basswood (Tilia spp.), black cherry (Prunus serotina Ehrh.), American beech (Fagus grandifolia Ehrh.), black walnut (Juglans nigra L.), eastern hemlock (Tsuga canadensis (L.) Carr.), hickory (Carya spp.), and oak (Quercus spp.). Butternut is considered to be a shade-intolerant species and cannot tolerate shade from competition above (Burns & Honkala, 1990).
The objectives of this research were to map the occurrence of surviving butternut in the United States using nationwide forest inventory data and quantify changes in butternut populations and density across its range in order to determine the potential impacts of butternut canker. An additional objective was to identify spatial trends in butternut changes and identify ecological characteristics of stands where butternut is surviving in order to suggest the potential characteristics of forests where butternut restoration is most likely to be successful.

| METHODS
The Forest Inventory and Analysis (FIA) program of the U.S. Prior to 1999, FIA collected data regionally using a periodic measurement system with sample designs that varied slightly by region.
Generally, inventories were conducted in each state every 6-18 years, depending on the state and region (Bechtold & Patterson, 2005). and total butternut volume (m 3 of live trees ≥2.54 cm d.b.h.) were generated for all states in the historic range of butternut (Little, 1971) at two time periods (Appendix S1) and by diameter class. Due to the variability in timing of the periodic inventories and implementation of annual inventories, the interval of time between surveys differs among states and was averaged for each ecoregion section (Appendix S2).
As the periodic and annual inventories are completely independent samples, it is not possible to directly attribute the death of individual trees. Therefore, long-term (multidecadal) changes in butternut abundance and density were estimated at regional levels instead of for individual trees.
Annualized change was calculated for each state by dividing the difference between estimates at time 1 and time 2 by the number of years between surveys, and relative change was computed by dividing the difference between time 1 and time 2 estimates by the time 1 estimate. In order to test for significant changes in butternut populations and volume across its range and within states, paired t tests were performed with ecoregion section-and plot-level estimates and times 1 and 2 (SAS Institute Inc, 2009) for ecoregion sections that had at least 10 plots containing butternut sampled at time 1. Relative and annualized change in butternut populations and volume were also mapped for all states.
In order to quantify the current level of tree mortality across the geographic range of the study, an annual mortality rate was estimated by state from remeasured annualized FIA data for butternut individually and for all species combined including butternut for comparison. and STDSZCD are likely to separate stands by recent management activity and site conditions related to competition, regeneration, and species composition. All included variables were categorical other than STAGE which was continuous. Before logistic models were run, ordinary linear regression, including tolerance and variance inflation diagnostics, was used to test for multicollinearity. None was observed because tolerance was above 0.4 in all cases (Allison, 1999).
Several other tree-level and stand-level attributes (elevation, slope, basal area, site index, and aspect) were included initially, but none were found to be statistically significant predictors (α = 0.05), so they were dropped from the models. Odds ratios were evaluated against a reference condition for each categorical variable where the reference condition was assigned as the category with the proportion of plots with butternut presence was closest to the average for the study area. Model goodness of fit was evaluated with the area under the receiver operating characteristic (ROC) curve, max-rescaled R-square, and percent accuracy of occurrence classification. The area under the ROC curve is provided for the classification models as an indicator of classification accuracy. An ROC value of 0.5 occurs when the classification is no better than random prediction; a value of 1.0 indicates perfect classification accuracy. A rough guide to interpretation is given by Fischer, Bachman, and Jaeschke (2003): ROC area greater than 0.9 ≈ high accuracy; 0.7-0.9 ≈ moderate accuracy; 0.5-0.7 ≈ low accuracy.
Mean annual change among all ecoregion sections was approximately −431,000 trees per year (0.2 per hectare; Table 1) and −87 million cubic meters of volume per year (0.001 cubic meters per hectare; Table 2). Substantial relative decreases in tree numbers and volume occurred in most ecoregion sections (Tables 1 and 2).
Three of the 19 ecoregion sections that were tested exhibited significant declines in butternut tree populations and volume. The largest significant decreases in butternut populations and volumes, computed as the difference in plot means between time 1 and time 2, were observed in ecoregion sections 212T, 222J, and 222L (Table 3) (Tables 1 and 2), but the decreases varied spatially (Figures 3 and 4). The largest annual decreases in numbers and volume occurred in section 222L (Figures 3c and 4c), but the largest relative decreases in numbers occurred in sections 222J and 212T and in volume in sections 222L, 222J, and 212T (Figures 3d and 4d).
Although butternut populations decreased in ecoregion sections 223E T A B L E 1 Estimates of numbers of butternut trees per hectare of timberland for times 1 and 2 with associated sampling errors, sample size (nonzero plots), annualized change, and relative change by ecoregion section   Table 6. ROC values for the models for both time periods indicated "moderate classification accuracy" ( Table 6). The chi-square statistics in Table 6 reveal that the most important variable for both time periods is ECOSECT. The second, third, and fourth most important variables in both models were FORTYPGRP, STDSZCD, and OWNGRPCD, respectively. STAGE and PHYSCL were the least important variables in both models.
Comparison of the odds ratios from both models indicated that butternut occurrence was significantly higher than the reference section, 221B, in ecoregion section 222L (Table 5), North-Central U.S.
Driftless and Escarpment, which is characterized as an unglaciated upland plateau with steep-sided bedrock ridges and mounds (McNab et al., 2007). The odds ratios for the time 2 model also indicated that butternut occurrence was significantly higher than the reference section in ecosection 212K (Table 5), Western Superior Highlands, which is characterized by uniform, undulating, poorly drained, level to rolling landscape of glacial drift plains (McNab et al., 2007). Similarly, odds ratios from both models indicated that butternut occurrence was significantly higher than the reference forest-type group, oak/pine, in the maple/beech/birch group and that occurrence was significantly higher on private lands than the reference ownership group, state and local governments. Finally, odds ratios from both models indicated that butternut occurrence was significantly lower than that of the reference stand size, large-diameter stands, in small-diameter stands, while odds ratios from the time 1 model indicated that occurrence was significantly lower in medium-diameter stands.

| DISCUSSION
Although butternut continues to be present across much of the range described by Little (1971; Figure 1), the abundance and volume of butternut trees have decreased across this range and in nearly all states (Table 1). It occurs at the highest densities in ecoregion sections 222L, 221H, 221E, and 222I (Figure 1), but the largest declines have also  Table 1 occurred in some of these sections, particularly 222L (Tables 1 and   3; Figures 3 and 4). Despite the apparent decrease in butternut tree abundance across its range and in most states, butternut volume has increased in 211E, 212K, 221H, and 222I although the increases were not statistically significant (Tables 2 and 3). Substantial decreases in the populations of butternut trees across nearly all diameter classes have occurred (Figure 2), but the loss of trees from the largest diameter classes has comparably larger ecological and economic consequences including reduced availability of mast, seed trees, den trees, carbon stores, and lumber. Mortality rate was computed by dividing the estimated volume of trees that died between measurement at time 1 and time 2 by the estimated volume of live trees present at time 1 for states with at least 10 remeasured plots containing butternut.

Volume of mortality trees (m 3 ) n
T A B L E 4 Estimates of butternut volume, mortality volume, sample size, and annual mortality rate for ecoregion sections with at least 10 remeasured plots F I G U R E 4 (a) Volume of butternut trees per acre, 1980s, (b) volume of butternut trees per acre, 2015, (c) annual change in butternut volume, (d) relative change in butternut volume, by ecoregion section for the time interval shown in Table 1 T A B L E 5 Estimated parameters for logistic regression occurrence models The spatial variation in changes in butternut tree abundance and volume and annual mortality rates (Figures 3 and 4; Table 3) may indicate differences in the virulence of the fungus causing butternut canker, the presence of disease-resistant butternut trees, or ecological site differences in the local environment that contribute to tree survival. Broders, Boraks, Barbison, Brown, and Boland (2015) reported T A B L E 5 (Continued) T A B L E 6 Performance of the logistic regression model of butternut occurrence by time period based on receiver operating characteristic (ROC) curve area, max-rescaled r-square value, and chi-square values for the parameter estimates in Table 5 Time period OWNGRPCD is ownership group, STAGE is stand age, STDSZCD is stand-size class, ECOSECT is ecoregion section, FORTYPGRP is forest-type group, and PHYSCL is physiographic class.
in Minnesota and Wisconsin in the 1960s and a less virulent strain that may have been present in the northeastern United States for far longer. The relative and annual decreases in numbers and volume reported here may reflect the difference between these two strains; the largest decreases in numbers and volume were concentrated in the Midwest and Lake States (Figures 3 and 4), but increases in butternut volume were concentrated in the 211E, 221H, and 223E ( Figure 4c,d).
Although there may be some variation in the heritability of re- where butternut has historically occurred. Hybrids may be best suited for planting at a different set of sites than would be appropriate for restoration using pure butternut genotypes (Crystal & Jacobs, 2014;Crystal, Lichti, Woeste, & Jacobs, 2016).
We report here a ca. 43% decline in numbers of butternut trees and 58% decline in butternut volume across the species range during a 30-to 40-year period. Although this is a large decline, it is substantially less than what has been reported elsewhere (U.S. Forest Service, 2005). During this same time period, the volume of black walnut (J. nigra L.), a species with a similar geographic distribution and shade tolerance classification, has increased by 125 percent. While we do not have any direct evidence of the cause of this change, the decline is most likely a result of tree mortality caused by the fungal pathogen O. clavigignenti-juglandacearum that may have been introduced from Asia many decades ago (Furnier et al., 1999;Nair, 1998;Ostry, 1998b;Ostry et al., 1994) or was a native fungal endophyte or minor pathogen that made a host jump to butternut (Broders, Boraks, Sanchez, & Boland, 2012). This conclusion is supported by focused surveys indicating infection by this pathogen as the major cause of butternut mortality (U.S. Forest Service, 2005).
The declining presence of butternut in North America thus represents another example of regional changes in forest composition driven by invasions of non-native insects and pathogens (Lovett et al., 2016), a problem that is particularly acute in eastern N. America (Liebhold et al., 2013). Other examples of massive declines of North American tree species caused by insect and pathogen invasions include the demise of American chestnut, Castanea dentata, caused by the exotic fungal pathogen Cryphonectria parasitica (Dalgleish, Nelson, Scrivani, & Jacobs, 2016), the regional decline in American beech, F. grandifolia, caused by beech bark disease (Morin & Liebhold, 2015) and the current wave of ash, Fraxinus spp., mortality caused by the emerald ash borer (Morin, Liebhold, Pugh, & Crocker, 2016). While the regional decline in abundance of butternut documented here is similar to these other examples in its regional scale, it is unique in that butternut was not a common tree, even before the invasion of the butternut canker pathogen. From a conservation perspective, the impacts of butternut canker are thus particularly acute as the pathogen invasion pushes a rare tree species toward extinction, at least at a local scale. Despite the rarity of butternut, its unique ecological characteristics including large mast may have a disproportionate impact on wildlife populations.
Butternut restoration offers a chance to reverse this trend, particularly if efforts are focused in forest stands most suitable for tree growth and recruitment.

ACKNOWLEDGMENTS
This study is dedicated to the late Kurt Gottschalk. Dr. Gottschalk was the lead author on this study prior to his passing in 2014. The authors are proud to finally complete the study in Dr. Gottschalk's honor. We

CONFLICT OF INTEREST
None declared.