Climate change winners and losers: The effects of climate change on five palm species in the Southeastern United States

Abstract Palms (Arecaceae) are a relatively speciose family and provide materials for food, construction, and handicraft, especially in the tropics. They are frequently used as paleo‐indicators for megathermal climates, and therefore, it is logical to predict that palms will benefit from predicted warmer temperatures under anthropogenic climate change. We created species distribution models to explore the projected ranges of five widespread southeastern North American palm species (Rhapidophyllum hystrix, Sabal etonia, Sabal minor, Sabal palmetto, and Serenoa repens) under four climate change scenarios through 2070. We project that the amount of habitat with >50% suitability for S. etonia will decline by a median of 50% by 2070, while the amount of habitat with >50% suitability S. minor will decline by a median of 97%. In contrast, the amount of suitable habitat for Rhapidophyllum hystrix will remain stable, while the amount of suitable habitat for Serenoa repens will slightly increase. The projected distribution for S. palmetto will increase substantially, by a median of approximately 21% across all scenarios. The centroid of the range of each species will shift generally north at a median rate of 23.5 km/decade. These five palm species have limited dispersal ability and require a relatively long time to mature and set fruit. Consequently, it is likely that the change in the distribution of these palms will lag behind the projected changes in climate. However, Arecaceae can modify physiological responses to heat and drought, which may permit these palms to persist as local conditions become increasingly inappropriate. Nonetheless, this plasticity is unlikely to indefinitely prevent local extinctions.

Palms are frequently employed as paleo-indicators for megathermal (i.e., tropical) climates (e.g., Greenwood & Wing, 1995;Pross et al., 2012). A combination of climate and dispersal ability appears to be the primary factors that determine palm species richness at both the continental and global scales (Blach-Overgaard, Kissling, Dransfield, Balslev, & Svenning, 2013;Eiserhardt et al., 2011). For instance, extinction rates in palms with megafaunal fruit in the western hemisphere have increased since the beginning of the Quaternary period, approximately 2.6 mya, due to a combination of climate oscillations and habitat fragmentation, as well as the loss of megafauna (Onstein et al., 2018).
There is some evidence that some palm species are expanding their range during recent decades. The dwarf palmetto, S. minor, has extended its range in Oklahoma (Butler, Curtis, McBride, Arbour, & Heck, 2011) and North Carolina (Tripp & Dexter, 2006), and individuals at the northwestern extreme of its range are undergoing a rapid population increase (Butler & Tran, 2017). The California fan palm Washingtonia filifera and the non-native Phoenix dactylifera have begun colonizing Death Valley Springs in California (Holmquest, Schmidt-Gengenbach, & Slaton, 2011). Chinese windmill palms (Trachycarpus fortunei) are gradually invading forests in Italy and Switzerland (Fehr & Burga, 2016;Walther et al., 2007).
Although there are a few studies documenting range shifts in palms, relatively little research has focused on the projected effect of climate change on palms. It has been suggested that African palms could be particularly vulnerable to anthropogenic climate change (Blach-Overgaard et al., 2009), with up to 87% of all species negatively impacted (Blach-Overgaard, Balslev, & Dransfield, 2015), although the near-term potential for extinction is considerably lower than for many other plant species (Cosiaux et al., 2018). In contrast, climatic changes are forecast to increase the extent of potentially suitable areas for commercially grown date palms (Phoenix dactylifera) in Iran, where as much as 61 million ha are projected to become suitable for date production by 2050 (Shabani, Kumar, & Taylor, 2014 Roystonia regia, Sabal etonia, S. mexicana, S. miamiensis, S. minor, S. palmetto, Serenoa repens, Thrinax morrisii, T. radiata, and Washingtonia filifera (Henderson et al., 1995). Five of these species are widespread in the southeastern United States, including Rhapidophyllum hystrix, S. etonia, S. minor, S. palmetto, and Serenoa repens. Large numbers of palms are commercially grown for ornamental horticulture in Florida and Texas (Broschat, Meerow, & Elliott, 2017) and four of these palm species are widely planted outside their native range, although S. etonia is seldom observed at commercial nurseries (pers. obs.).
One species, Serenoa repens, is particularly commercially valuable, as it is one of the top three herbaceous dietary supplements in the United States (Jaiswal et al., 2019), generating sales of approximately $23 million USD during 2015 (Gafner & Baggett, 2017). Additionally, Serenoa repens is considered a keystone species (Carrington & Mullahey, 2006), with more than 200 vertebrate using it for foraging, cover, or nesting (Maehr & Layne, 1996).
Despite the importance of palms to the ecology and economy of the southeastern United States, the effects of anthropogenic climate change on the distribution of these species have not yet been investigated. Our goal was to identify the bioclimatic variables that determine the niches of these five widespread palm species in the southeastern United States. We then projected the spatial extent of these variables under multiple climate change scenarios for 2050 and 2070, in order explore how the distribution of these species might be affected by anthropogenic climate change.

| MATERIAL S AND ME THODS
We modeled the current and projected ranges of five palm species: Rhapidophyllum hystrix, Sabal etonia S. minor, S. palmetto, and Serenoa repens (Phillips, Anderson, & Schapire, 2006;Phillips, Dudik, & Schapire, 2004; Figure 1). We downloaded records of these five species from the Global Biodiversity Information Facility (https:// www.gbif.org/) and combined them with undigitized herbarium records from Cornell and the New York Botanical Gardens. We followed the procedures outlined in Butler, Stanila, and Iverson (2016) for data processing and model building. We eliminated duplicates and records from outside the native range and resampled the locality data to one record per 25 km 2 . We downloaded elevation and 19 bioclimatic variables from WorldClim (Hijmans, Cameron, Parra, Jones, & Jarvis, 2005; http://www.world clim.org/) at a resolution of 2.5 arc-minutes (25 km 2 ; Table 1). We downloaded rasters of organic matter, pH, and available water content from the STATSGO2 dataset (http://webso ilsur vey.nrcs.usda.gov). The spatial extent of the analysis can influence several aspects of the modeling process (Barve et al., 2011;Merow, Smith, & Silander, 2013), and it is recommended that the ecology and the dispersal abilities of the organisms be considered when building models. Since many of these palms are planted far outside their native range (e.g., Sabal palmetto will grow unprotected in Oklahoma City, Oklahoma, and Sabal minor will apparently survive in Manhattan, NY; pers. obs.), the spatial extent of the variables was set to the area from extreme southern Texas north to the southern third of Canada and east to the Atlantic Ocean. We followed the procedure outlined by Butler et al. (2016) and only included the variables with the most useful predictive information (i.e., the highest gain when used in isolation), as well as the variables that provided unique predictive information. As regularization multipliers (β) are an important component of model prediction and complexity (Moreno-Amat et al., 2015), we used the regularization approach implemented in ENMtools (Warren, Matzke, & Cardillo, 2019) and small sample corrected variant of Akaike's information criterion (AICc) scores were used to evaluate models (Warren & Seifert, 2011) using all possible combinations of the variables that did not exhibit high multicollinearity (e.g., |r| < .8). We used 10,000 background points, with 70% of occurrence records used for training, and 30% used for model validation. We plotted sensitivity versus 1 -specificity to created receiver operating characteristic (ROC), and 10-fold crossvalidation AUC (area under the curve) scores were used to evaluate the accuracy of the resulting model. We used AICc scores and model weights in conjunction with AUC scores to determine the models that best describe the current distributions of the five palm species.
We projected the potential future distribution of Rhapidophyllum hystrix, Sabal etonia, S. minor, S. palmetto, and Serenoa repens at 2.5 arc-minutes (25 km 2 ) using the model that best predicted the current distribution of each species in conjunction with future climate conditions for 2050 and 2070 using the IPCC 5 data from WorldClim (Hijmans et al., 2005). Four IPCC scenarios were evaluated, including RCP 2.6, RCP 4.5, RCP 6.0, and RCP 8.5, which differed in the amount of carbon dioxide added to the atmosphere over the 21st

| RE SULTS
The best model for Rhapidophyllum hystrix (i.e., with the lowest AICc score) included the variables elevation, mean temperature of coldest quarter (BIO 11), precipitation of wettest month (BIO 13), and precipitation of warmest quarter (BIO 18; Table 2). The AUC for this model was 0.978 ± 0.003. Areas with suitability >50% had an F I G U R E 1 The five palm species whose distribution we examined included Rhapidophyllum hystrix (a), Sabal etonia (b), Sabal minor (c), Sabal palmetto (d), and Serenoa repens (e) Note: The natural log of probability of the data present in the model is given by the log likelihood. AICc is a small-sampled corrected AIC score; only models that are within four units of the top AICc model are shown. Delta AICc is the difference between the AICc score for a model and the lowest AICc score. The model weight (wAICc) is the relative likelihood for each model, divided by the total relative likelihood for all models that were considered. AUC (area under the curve) is a metric for evaluating the accuracy of the model. The regularization multiplier is given by β.  Table 2). The AUC for this model was 0.992 ± 0.001.

TA B L E 1 The bioclimatic and soil variables examined in this study
There was also some model support for minimum temperature of coldest month (BIO 6) and mean temperature of coldest quarter (BIO 11). Areas with suitability >50% had a mean temperature of the wettest quarter of 26.8-27.7°C, moderate precipitation seasonality (the coefficient of variation ranged from 39 to 50), precipitation of the wettest quarter of 50.6-59.0 cm, and precipitation of the warmest quarter of 50.3-57.8 cm. Areas that are currently shown as >50% suitability are restricted to Florida ( Figure 3).
The best model for Sabal minor included temperature annual range (BIO 7), mean temperature of warmest quarter (BIO 10), mean temperature of coldest quarter (BIO 11), and elevation ( There was also some model support for mean temperature of coldest F I G U R E 3 The modeled current and future distributions for Sabal etonia. The legend shows the probability of occurrence, with the darkest shade representing >0.5 probability. Blue circles represent sites where S. etonia (n = 86) were located quarter (BIO 11) and precipitation of wettest quarter (BIO 16). Areas that were predicted to have suitability >50% had a minimum temperature of coldest month above 6.4°C, with a mean temperature of wettest quarter ranging from 26.5 to 27.8°C, precipitation of the warmest quarter above 46.7 cm, and in soils with a pH below 5.6.
Areas that are currently shown as >50% suitability extended from southern Louisiana east to southern Georgia ( Figure 6).
The median projected change in highly suitable conditions (i.e., those >50% suitability) for all five species by 2070 was −2% (range −99% to 30%), although there was considerable variation among species (  Figure 7). There was also a decline in projected precipitation of wettest quarter and precipitation of warmest quarter, although precipitation seasonality remained largely the same (Figure 7).  Figure 8).
Under nearly all scenarios, highly suitable habitat for S. palmetto expanded and shifted slightly northwest ( Figure 5) 6.0 scenario, the amount of highly suitable habitat decreased by 2% to 131,568 km 2 , of which 86% was shared with the current range (Table 3).
Under all scenarios, suitable conditions for Serenoa repens shifted slightly to the northeast and generally increased in extent (Figure 6).  (Table 3).

The change in distribution for
Centroids shifted generally northward for each of the five species (Figure 9), at a median rate of 23.5 km/decade. However, the response rate varied substantially among species. Under all scenarios, the rate of change for Rhapidophyllum hystrix was 11-24 km/decade (Table 4).
Centroids for Serenoa repens and S. palmetto shifted at a moderate rate of 13-34 km and 12-43 km/decade, respectively. However, centroids for S. minor shifted much faster (68-160 km/decade) than other species. The rapid shift to the northeast for S. minor centroids should be interpreted cautiously, however, because a pronounced range contraction is forecast resulting in greater weights being assigned to locations currently beyond the natural range of this species.
Palms are commonly used as indicators for megathermal climates (e.g., Pross et al., 2012;Reichgelt, West, & Greenwood, 2018) and therefore should be especially responsive to climate change. Our models suggest that highly suitable habitat for S. etonia and S. minor will decline substantially in extent during the 21st century while the amount of highly suitable habitat for Rhapidophyllum hystrix will stay largely constant. Highly suitable habitat is projected to slightly increase for Serenoa repens and substantially increase for S. palmetto. These results broadly mirror the results published on other taxa in the southeastern United States, which show some species increasing in extent while other species decline under anthropogenic climate change (e.g., Butler et al., 2016;McKenney, Pedlar, Lawrence, Campbell, & Hutchinson, 2007;Osland, Enwright, Day, & Doyle, 2013).
Despite the potential for the range of some palm species to increase in extent, these five species may be unable to enlarge their ranges as rapidly as the habitat becomes potentially suitable. For example, by 2050, the extent of highly suitable habitat for S. palmetto is projected to increase by 18%-50%. However, the most frequent dispersal method of S. palmetto seeds is by raccoon (Procyon lotor), gopher tortoise (Gopherus polyphemus), white-tailed deer (Odocoileus virginianus), and feral hog (Sus scrofa; Abrahamson & Abrahamson, 1989), none of which typically disperse very far (Gehrt & Fritzell, 1998;Kilgo, Labisky, & Fritzen, 1996;McRae, Landers, & Garner, 1981;Truvé & Lemel, 2003). Although birds may also occasionally feed on S. palmetto seeds, fruit set is during October when many bird species are migrating south (Stiles, 1980), which makes it unlikely that avian frugivory will facilitate northward dispersal.
Additionally, in northern Florida, it takes a minimum of 14 years for wild Sabal palmetto to begin growing a trunk and 59 years for half F I G U R E 8 Boxplots of BIO 7 (temperature annual range), BIO 10 (mean temperature of warmest quarter), and BIO 11 (mean temperature of coldest quarter) for the area that currently has >0.5 probability of occurrence for S. minor, showing the changes in bioclimatic variables in that area by 2070 under the RCP 8.5 scenario of all individuals to develop a trunk (McPherson & Williams, 1996).
Since S. palmetto will not fruit until it has developed a trunk (Fox & Andreu, 2019), expansion of S. palmetto by animal dispersal outside of its current range is likely to be slow.
Likewise, the potential for Serenoa repens to rapidly colonize new suitable habitat appears to be limited. While Serenoa repens produces viable seeds and seedlings, in Florida it appears to spread primarily by vegetative sprouts, with some genets speculated to have been present for millennia (Takahashi, Horner, Kubota, Keller, & Abrahamson, 2011). Although seedlings exhibit relatively high survivorship, with 35%-57% surviving over a 19-year study, average growth rate is very slow and was generally <0.5 cm per year (Abrahamson & Abrahamson, 2009), although the growth rate of some individuals may be higher in the absence of exotic grasses (Foster & Schmalzer, 2012). The combination of primarily vegetative spread and a very slow growth rate suggests that the ability of this species to expand its range in concordance with the changing climate is probably extraordinarily low.
However, the models for the projected distributions of Serenoa repens should be interpreted cautiously. Although Serenoa repens is endemic to the United States, the native range extends to the southern tip of Florida (Henderson et al., 1995 We projected that the extent of highly suitable habitat for S. minor will exhibit a dramatic decline during the 21st century. The current distribution of the dwarf palmetto, S. minor, extends from Oklahoma to Texas and east to North Carolina and Florida (Butler & Tran, 2017). Globally, it is listed as a secure species, and at the state level, it is not considered to be a species of special concern across most of its range, with the exception of North Carolina where it is listed as S3 (Vulnerable) species and Oklahoma where it is listed as a S2 (Imperiled) (NatureServe, 2019;ONHI, 2017). However, we project that the amount of highly suitable habitat for S. minor will decline by 87%-93% by 2050, driven primarily by an increase in the F I G U R E 9 The centroids are the geometric center of the range of each species under each scenario. A black star represents the current centroid, while blue stars show projected centroids by 2050 and red stars show projected centroids by 2070. Due to the concave distribution of Sabal etonia and S. palmetto, centroids for these species are present in the Gulf of Mexico, where neither are expected to occur mean temperature of the warmest quarter across its current range.
It is conceivable that S. minor may be able to withstand warmer temperatures than current conditions. For example, Goldman (1999) documented an isolated population of S. minor south of the main range in Nuevo León, Mexico (Goldman, 1999). However, this population shows introgression with S. mexicana (Goldman et al., 2011), a species that is widespread in Central America, and it is possible that the tolerance of S. minor for the climate in this location could be partly genetic.
The median projected centroid shift for each species was 23.5 km/ decade and ranged from 11 to 160 km/decade. However, palms typically exhibit low dispersal ability (Bacon et al., 2013), and it may not TA B L E 4 The distance from the centroid for each scenario to the current centroid as well as the rate per decade be possible for these species to expand their range at this rate. Animal seed dispersers are an important component of palm reproduction (Zona & Henderson, 1989), and seed dispersal for the five study species is primarily by animals (Zona, 2006) although information on seed dispersal in some of the five palm species is very limited. For example, the only documented animal dispersing seeds of Rhapidophyllum hystrix are the black bear (Ursus americanus; Maehr, 1984). Additionally, anthropogenic climate change may affect plant recruitment and could potentially enhance, delay or even preclude seed regeneration (Walck, Hidayati, Dixon, Thompson, & Poschlod, 2011). Furthermore, we did not incorporate the sea level rise of 0.5-1.4 m above 1990 levels, projected to occur by the end of the 21st century (Rahmstorf, 2007), which may potentially reduce the extent of suitable habitat for all five palms. Finally, we did not attempt to incorporate changes in land use in our models, which may affect the prevalence of palms in the future.
For example, both S. etonia and Serenoa repens exhibit strong flowering responses after episodic fires (Abrahamson, 1999;Carrington & Mullahey, 2013). Efforts to suppress fires, therefore, could potentially restrict the persistence of these species on the landscape during the coming decades.
However, some species may disperse in a fashion that leads to isolated founder plants that can establish new populations, if local environmental conditions are suitable (Shapcock et al., 2020). Given that four of the five species considered here are common in the nursery trade, it is possible that individuals planted in gardens outside of the native range may act facilitate naturalization for future generations, similar to the pattern observed for Trachycarpus fortunei in Switzerland (Fehr & Burga, 2016) and eight invasive palm species in Panama (Svenning, 2002).
In addition to changes in temperature, precipitation, and seasonality, ongoing increases in atmospheric greenhouse gases are affecting growth and physiology in plants (Thompson, Gamage, Hirotsu, Martin, & Seneweera, 2017). Increasing levels of CO 2 have increased growing season leaf area, particularly in the tropics (Zhu et al., 2016).
Overall, however, the ability of these five palm species to take advantage of suitable conditions outside of their native range appears to be limited. Additionally, Lavergne, Mouquet, Thuiller, and Ronce (2010) suggested the long-lived species with low rates of reproduction and dispersal may not be able to keep pace with environmental changes wrought by anthropogenic climate change.
Native palm species in the southeastern United States appear to fit this mold, as they exhibit high adult survivorship coupled with a low dispersal ability. Sabal minor, for example, may reach up 400 years of age (Ramp, 1989) and individual stems of Serenoa repens may live to 700 years with near-zero annual mortality (Abrahamson, 1995).
Consequently, while conditions in current native range may become increasingly unsuitable for some species, these palms may temporarily avoid local extinction, particularly if they are able to take advantage of refugia (Ashcroft, Chisholm, & French, 2009;McLaughlin et al., 2017). Nonetheless, these responses will likely be insufficient to prevent local extinction over the long term.

ACK N OWLED G M ENTS
We thank A. Stalter and A. Kirchgessner for access to herbarium records and D. Goldman for assistance with identifying valid localities for specimens. Funding for this project was provided by the UCO Office of Research & Grants and CURE-STEM.

CO N FLI C T O F I NTE R E S T
There are no competing interests.