Climate change both facilitates and inhibits invasive plant ranges in New England

Significance

Invasive species are often expected to benefit from novel conditions encountered with global change. Our range models based on demography show that invasive Alliaria petiolata (garlic mustard) may have much lower establishment in New England under future climate, despite prolific success under current climate, whereas other invasive and native plants may expand their ranges. Forecasts suggest that management should focus on inhibiting northward spread of A. petiolata into unoccupied areas and understanding source–sink population dynamics and how community dynamics might respond to loss of A. petiolata (it modifies soil properties). Our methods illustrate inadequacy of current approaches to forecasting invasions in progress, which are based on correlations between species’ occurrence and environment and illustrate critical need for mechanistic studies.

Abstract

Forecasting ecological responses to climate change, invasion, and their interaction must rely on understanding underlying mechanisms. However, such forecasts require extrapolation into new locations and environments. We linked demography and environment using experimental biogeography to forecast invasive and native species’ potential ranges under present and future climate in New England, United States to overcome issues of extrapolation in novel environments. We studied two potentially nonequilibrium invasive plants’ distributions, Alliaria petiolata (garlic mustard) and Berberis thunbergii (Japanese barberry), each paired with their native ecological analogs to better understand demographic drivers of invasions. Our models predict that climate change will considerably reduce establishment of a currently prolific invader (A. petiolata) throughout New England driven by poor demographic performance in warmer climates. In contrast, invasion of B. thunbergii will be facilitated because of higher growth and germination in warmer climates, with higher likelihood to establish farther north and in closed canopy habitats in the south. Invasion success is in high fecundity for both invasive species and demographic compensation for A. petiolata relative to native analogs. For A. petiolata, simulations suggest that eradication efforts would require unrealistic efficiency; hence, management should focus on inhibiting spread into colder, currently unoccupied areas, understanding source–sink dynamics, and understanding community dynamics should A. petiolata (which is allelopathic) decline. Our results—based on considerable differences with correlative occurrence models typically used for such biogeographic forecasts—suggest the urgency of incorporating mechanism into range forecasting and invasion management to understand how climate change may alter current invasion patterns.

Invasions and climate change are two of the primary factors that alter ecological systems. Forecasting ecological responses to these dynamic, potentially no-analog scenarios requires biologists to understand the fundamental processes that regulate change. The interaction of climate change and invasion remains a mystery, although it has been argued that climate change may foster invasions in many cases, whereas inhibition is less likely (1, 2). Studies have focused on such positive interactions (3), because they are readily observed; it is difficult to recognize when climate change has mitigated an invasion simply because there may not be an invasion to study. Only with a mechanistic understanding of how climate regulates life history transitions to mitigate or accelerate invasions can we improve the efficiency of management plans.

Links between global change and invasion are complex and idiosyncratic, although in general, climate change, land use change, and increased resource availability seem to favor invasive species over natives (examples are reviewed in ref. 1). Although the advantages offered to invasive species by disturbed habitat and increased CO₂ and N are evident, the effect of climate change per se on invasion is less clear (ref. 4 discusses the influence of extreme weather events). Existing studies have relied primarily on correlative range models (species distribution/niche/climate envelope models) to forecast how habitat suitability might change with climate (5–8), often because mechanistic approaches are more data hungry. However, forecasting invasions and other nonequilibrium scenarios necessarily requires predictions in locations that are either geographically or environmentally different from where the species has been observed, which is precisely where occurrence datasets are lacking. Few studies have linked demography to climate change to understand invasion (9, 10), but these studies will be critical to understand drivers underlying unanticipated shifts in invasion risk.

Incorporating mechanism is key to the extrapolation necessary to forecast invasions. To detect mechanisms, we need experiments. Manipulation of factors likely to change with climate change—temperature, precipitation, nitrogen, and carbon dioxide—and measurement of organismal response under field conditions integrate the biotic and abiotic factors affecting individuals. Species can also be manipulated as we do here in this “experimental biogeography” context. Individuals were transplanted to span geographic and/or environmental gradients larger than those delimited by current occurrence data. Experimental bioassays of species performance in natural environments can provide critical insights into the factors limiting current distributions and predicting further spread (11, 12).

Here, we present a mechanistic approach to assessing establishment risk for two invasive plants in New England (NE; in the northeastern United States) by combining experimental biogeography with demographically based population modeling. We studied Alliaria petiolata (garlic mustard), a monocarpic biennial mustard (Brassicaseae) native to Eurasia that is now prevalent throughout southern NE and much of the eastern United States (13), as well as Berberis thunbergii (Japanese barberry), a woody shrub native to Japan that is distributed throughout the United States as an ornamental species and now found in natural areas throughout southern NE and much of the eastern United States (13). We paired each invasive species with a native ecological analog [Turritis (formerly Arabis) glabra and Lindera benzoin, respectively] to gain insight into the life history attributes that made these species invasive, compared them with native species responses to climate change, and determined whether our demographic approach could accurately reconstruct known distributions of native species. We combined vital rates in Bayesian integral projection models (IPMs) (cf. 14, 15) for each species and projected the models across the NE landscape to predict their potential present and future geographic distributions based on the ability to establish a population. We used these models to address the following questions.

• i)Which uninvaded areas have suitable climate and/or habitat and risk invasion under current and future climate conditions?
• ii)What are the demographic drivers of invasion, how do they differ from those of native species, and how can they inform control strategies?
• iii)Do insights gained from demographically driven distribution models result in qualitatively different predictions from occurrence-based approaches to predicting species’ distributions?

Results

Range Predictions and Climatic Niches.

By following individual plants transplanted across the environmental temperature and precipitation gradients in NE, we were able to infer demographic responses of both invasive species and their native analogs under conditions where they are already both present and absent. Our demographically based range predictions broadly captured the known distribution of both native species based on predicting positive population growth rate (λ > 1) during the 20 y after introduction at presence locations (89.5% correct presences for T. glabra and 100% correct for L. benzoin) (Fig. 1). Notably, the spatial pattern of presences for both natives closely matches the predicted suitable habitat (Fig. 1) and significantly differentiates locations where each species is present compared with locations where it has not been detected (SI Appendix, Fig. S3). Furthermore, viable populations (λ > 1) of T. glabra and L. benzoin were only predicted in high-light (open canopy) and low-light (closed canopy) habitats, respectively, where they are known to occur (16) (SI Appendix, Figs. S17 and S33). Hence, our experimental approach is capable of accurately predicting potential geographic distributions. The invasive species were able to survive and grow even in the northernmost plots outside current known distributions. Native analogs survived in some of the northern plots as well, but overall, λ was lower across sites (Fig. 1), and they were generally more sensitive to light and soil conditions (SI Appendix, sections C–F). Predicted midcentury potential distributions all showed considerable shifts compared with current distributions [Fig. 1 and SI Appendix, Figs. S8, S19, S26, and S34, all species responses to a full range of projected general circulation model (GCM) scenarios]. Although A. petiolata was predicted to have suitable habitat throughout NE under current climate, future warming is predicted to render much of southern NE unsuitable or just marginally viable under all six climate models that we considered (SI Appendix, Fig. S8). Parts of northern NE, where A. petiolata likely has not reached, remain at risk. In contrast, northern parts of NE that are currently marginally suitable or unsuitable for B. thunbergii and both native species were predicted to become much more suitable.

Fig. 1.

(A, C, E, and G) Current and (B, D, F, and H) future (2041–2050) predicted distributions based on population growth at low density. Future projections were based on an intermediate GCM among six that we considered [General Fluid Dynamics Laboratory Earth System Model version 2M (GFDL.ESM2M)] under RCP4.5 (SI Appendix). White circles indicate all known presence points. Although absence data are not available to formally validate demography predictions, we note that, for (C and G) both native species, presences are distributed in suitable habitat (λ > 1) with a higher density of records in places with higher λ (population growth during establishment) than expected by chance (SI Appendix, Fig. S3), confirming that our study approach can reasonably predict geographic distributions. Both invasive species have the potential to spread considerably farther north based on current climate. In the future, models predict that demographic performance of (B) A. petiolata will be mitigated by climate warming in southern NE but will be improved for (D) B. thunbergii. Predictions were made based on the assumption of “good” local habitat values of N, pH, and PAR (z score = ±1.5 as appropriate for each species).

Comparing Demographic with Occurrence Models.

For B. thunbergii, the responses of λ and relative occurrence rate (ROR; relative probability that a presence was observed in each cell) to climate resulted in very similar predictions of suitable habitat, except at the edges of the climate domain where there are few data (Figs. 1E and 2B). These similarities lead to high correlation between rankings of habitat suitability under present climate (r = 0.85 between λ and ROR). However, under the future climate scenario, ROR is relatively even across the region (Fig. 2D), whereas λ-values are considerably higher but show the same spatial pattern as under present climate (r = 0.34 between λ and ROR). The occurrence and demographic models confirm one another; because λ > 1 everywhere, all cells are roughly equally likely to contain a presence. However, the demographic model contains more information on variation in habitat suitability based on the predicted variation in λ. In contrast, nearly opposite climate responses are fitted between λ and ROR for A. petiolata (Fig. 3 A and B vs. Fig. 2 E and G). These responses lead to negative correlations between λ and ROR and highlight the risk associated with inferring invasion potential based only on occurrence (r = −0.24 under present climate; r = −0.26 under future climate) (compare Fig. 1 A and B with Fig. 2 A and C).

Fig. 2.

Predicted distributions and fitted response curves from occurrence models for comparison with demographically based predictions in Figs. 1 and 3. (A, C, E, and G) A. petiolata. (B, D, F, and H) B. thunbergii. (E–H) Fitted response curves are shown in red that reflect a ratio between used habitat (blue densities) and available sampled habitat (gray densities). Both the current and future distributions [from GCM General Fluid Dynamics Laboratory Earth System Model version 2M (GFDL.ESM2M) under RCP4.5] for both invasive species differ substantially compared with our demographic models. Notably, some abrupt jumps in response curves occur in parts of climate space with very few data.

Fig. 3.

Predicted response of λ to climate gradients. Rows indicate λ for different species to temperature (A, C, E, and G) and precipitation (B, D, F, and H). Gray circles indicate predictions in each cell across NE. Red lines indicate general trends fitted with generalized additive models. These response curves synthesize the effect of climate across all vital rates (which can be different for growth, survival, etc.). Scatter around the lines is driven by differences among vital rates in their responses to climate. Predictions are made based on the assumption of good local habitat values of N, pH, and PAR (z score = ±1.5 as appropriate for each species).

Native/Invasive Comparative Demography.

A. petiolata performed best in cooler, wetter climates (Fig. 3), with all vital rates responding similarly (Fig. 4 and SI Appendix, section C). In stark contrast, the microhabitat responses differed for different vital rates; lower light levels supported A. petiolata survival and growth, whereas seed number increased with increasing light levels (Fig. 5). This tradeoff between seed production and growth/survival provides an insight into the persistence of A. petiolata; the tradeoff results in very similar λ in both closed and open canopy (Fig. 5 D and H). For the native T. glabra, the growth and survival had a similar climatic response to A. petiolata; however, seed production and germination were higher in warmer, drier plots (Fig. 4 and SI Appendix, section D). The net effect at the population level was dominated by seed production and germination, resulting in a nearly opposite climatic response of λ compared with A. petiolata (Fig. 3).

Fig. 4.

Comparison of fitted vital rates between (invasive) A. petiolata (A–F) and (native) T. glabra (G–L) at each of 21 experimental locations. Dashed lines/bars indicate open canopy habitat, whereas solid lines/bars indicate closed canopy habitat. Notably, A. petiolata has superior performance across all vital rates on average; however, the primary difference that explained its success was its substantially higher seed production compared with T. glabra. A. petiolata exhibits poorer performance in warm environments than cooler ones (note lower survival, seed production, germination, and germinant survival).

Fig. 5.

Comparison of predictions for A. petiolata in (A–D) open and (E–H) closed habitats. Similarly, high population growth in both (D) open and (H) closed habitats is driven by demographic compensation, wherein vital rates respond differently to environmental gradients.

B. thunbergii had higher growth, juvenile survival, and germination in warmer climates, whereas only juvenile survival showed a positive response to May precipitation (SI Appendix, section E). The net effect at the population level was a strong positive response to warmer climate and a modal response to precipitation (Fig. 3). The native, L. benzoin, had a variable response to warmer temperatures and a negative response to wetter climates across all vital rates (SI Appendix, section F). At the population level, λ showed a strong positive response to temperature driven by growth and a strong negative response to May precipitation driven by all vital rates in concert (SI Appendix, Table S9). B. thunbergii growth and survival were higher in warmer conditions than for L. benzoin, and seed production was orders of magnitude greater for B. thunbergii, thus resulting in higher overall fecundity for the invasive (SI Appendix, Figs. S21 and S30).

Discussion

Our results show how interactions between the demographic drivers of invasions and climate can both facilitate and inhibit invasion success. Future increases in population growth, as predicted for B. thunbergii, as well as vacancies left if a dominant invasive disappears, as predicted for A. petiolata, may both have dramatic effects on communities and ecosystems. Only by understanding the demographic drivers of invasion success can we begin to anticipate systematic shifts in invasion risk and ensure that management resources are used efficiently.

Present and Future Distributions.

Biennials.

Under current climate, invasion risk is extremely high for A. petiolata throughout NE (Fig. 1A). The high suitability predicted in the north suggests that A. petiolata has not yet dispersed there (17). A. petiolata’s success is driven by apparent demographic compensation (vital rates have opposite responses to environmental gradients) (18), wherein vital rates respond differently to open and closed habitats (Fig. 5). Although A. petiolata is known to flourish in forest understory (19), its high reproductive output in open habitat implies that it can bridge gaps in its distribution by producing many seeds, giving greater opportunity to reach closed canopy habitat where growth and survival are high.

However, projected future climate change may mitigate A. petiolata’s invasion in southern NE while reducing otherwise prolific population growth in some parts of northern NE. Better performance in cooler climates (Figs. 3A and 4) indicates that expected regional warming may render southern NE unsuitable for establishment of A. petiolata. A. petiolata’s negative response of all vital rates to temperature contrasts with the variable environmental responses of all other species in our study, suggesting that this negative response is not idiosyncratic (Fig. 4 A–F and regressions in SI Appendix, section C). These future forecasts require extrapolation to warmer conditions than observed in our experiment and consequently, should be interpreted with caution. That is, the functional form of climatic responses may be more complex than the simple linear functions that our samples permitted, or other factors other than climate may limit distribution. Although the precise climate dependence of λ may be imperfect when extrapolating, the qualitative implications of a negative response to warmer temperatures seem robust in this study (see below). In contrast, suitability will increase regionally for T. glabra under future climate projections.

Because of the suggestive nature of A. petiolata’s temperature response, we sought other evidence for its generality. We projected our demographic model across the eastern United States (following the same methods used in NE) and compared this with known occurrence points. For this comparison, we calculated the posterior probability that λ > 1 and interpreted this as the probability that a population could establish. The southern range boundary seems well-predicted by our model (Fig. 6). Hence, although A. petiolata is known to perform well in the central United States, our models predict the southern limit of known populations quite accurately. Of course, some inaccuracies persist; for example, Pardini et al. (20) found populations in Missouri on both sides of our predicted range boundary (Fig. 6). Given that deviations between our predictions and observed presences might be driven by more differences in local habitat conditions than our model can accommodate (i.e., source–sink dynamics or overly simple environmental responses in our fitted models among many other possibilities), our predictive accuracy is reassuring. Regardless, the clear implication of our projection across the eastern United States is that, in warmer climates, the local habitat conditions need to be more favorable to offset the negative temperature response. Extrapolation in environmental space across the eastern United States is comparable with that required to forecast future climate in NE; hence, reliable predictions across the eastern United States support the reliability of our NE future forecasts.

Fig. 6.

Known occurrence records (black dots) of A. petiolata overlaid on predictions of establishment probability (posterior probability that population size can increase) in closed canopy (A) and open canopy (B) habitat. Predictions are made for “optimal” local habitat (pH, N, light; z score = ±3) conditions. The correspondence between known presences and our predictions suggests that our demographic model has identified a reliable factor (primarily warmer summer temperatures) limiting the distribution of A. petiolata and lends support to our future extrapolations in NE (Fig. 1). Note that these predictions for the eastern United States involve significant extrapolation in both environmental and geographic space, and future studies should aim to evaluate their generality.

To further test the generality of the fitted demographic responses of A. petiolata, we performed a similar exploration in its native range (Europe) and correctly predicted 99.9% of presences there (SI Appendix, Fig. S11). Hence, even when extrapolating outside the study region, our demographic models capture a dominant factor—warm summer temperatures—shaping distributional limits. Despite this likely distributional limit, we caution that A. petiolata’s prolific seed production (e.g., 6 of 133 flowering adults produced >2,300 seeds) (similar results are in ref. 20) would enable it to readily establish after local extinction if it were able to persist in particularly favorable microclimates or during a sequence of favorable years.

Woody shrubs.

Under current climate, our models predict that B. thunbergii can establish in all but coastal Maine and the northernmost parts of Maine, New Hampshire, and Vermont (Fig. 1E). Even in areas predicted to be unsuitable (λ < 1), λ is close to one, and some occurrence records in these areas suggest that populations are viable, likely in particularly favorable local habitat or because of establishment after a few years of favorable weather. In any case, based on B. thunbergii’s positive response to warm temperatures (Fig. 3), it is apparent that these northern habitats will only become more suitable under climate change, allowing prolific population growth throughout the region (Fig. 1F and SI Appendix, Fig. S26). Although B. thunbergii may establish in open habitat throughout much of NE, our models predict that it can also establish in closed canopy habitats in southern NE (SI Appendix, Figs. S24 and S25) (21, 22). Establishment in such suboptimal habitats may enable B. thunbergii to spread more rapidly across a fragmented landscape, although we did not see evidence for demographic compensation as with A. petiolata. The native, L. benzoin, shows very similar patterns to those of the invasive, suggesting that they will continue to co-occur in a warming climate.

Comparison with Occurrence Models.

To advance invasion risk forecasts, we need to understand the relationships between more intensive demographic studies like this and more commonly available occurrence-based estimates. Why do λ and ROR agree for B. thunbergii but not for A. petiolata? The correlation of the spatial pattern of occurrence data for B. thunbergii with λ in Fig. 1C suggests that B. thunbergii’s geographic distribution may be closer to equilibrium than that of A. petiolata. The possibility of equilibrium is further supported by noting that, across their global ranges, B. thunbergii occurs in colder locations than observed in NE in only 4 of 3,635 records, whereas A. petiolata commonly occurs in colder locations (SI Appendix, Figs. S12 and S28). B. thunbergii is an agricultural land abandonment specialist, and with little in the way of this land use change happening today in NE, there is likely very little expansion of current populations under current climate (21).

Understanding how the species distribution model generates the response curves that drive their predictions provides insight into the different temperature responses by A. petiolata between the occurrence- and demography-based models. The occurrence-based model predicts an increasing response to temperature between 18 °C and 21 °C (in contrast to the decline by the demography-based model). This disparity derives from a fundamental difference between the processes generating occurrence and demographic patterns. A species does not necessarily occur most frequently at locations where its demographic performance is highest (23). High dispersal rates or availability of disturbed habitat can increase occurrence rates compared with less accessible locations where demographic performance is higher. This difference highlights the critical need for disentangling processes from patterns; one could errantly infer a positive response to temperature for A. petiolata in some parts of environmental space when, in fact, this response may be caused by dispersal or land use patterns. By identifying the correct demographic response, management plans for A. petiolata can be better designed for less accessible but at-risk regions of northern NE that would have been interpreted as unsuitable from the occurrence-based model alone (Fig. 2A). Finally, the occurrence-based model draws background points from cold northern locations where few samples have been recorded—which leads to the declining response to temperature between 21 °C and 23 °C in Fig. 2E.

Three observations support the use of the demographic model over the occurrence model for A. petiolata. First, all vital rates showed a negative response to warmer temperature (Fig. 4 and SI Appendix, section C), meaning that predictions of λ are unlikely to be driven by idiosyncrasies in our experiment. Superior vital rates in the north suggest that the generally positive response of occurrence to temperature is driven by dispersal limitation, additional sampling bias that we have not accounted for, or both. Second, the mismatch between current occurrence records in NE and high-suitability habitat predicted by the demographic model suggests dispersal limitation. The observation of a handful of presence records in southern Canada suggests that some suitable habitat exists in the north. Third, the current distribution map of A. petiolata across the eastern United States and Europe suggests that it is limited by warmer temperatures (Fig. 6 and SI Appendix, Fig. S11).

Management Recommendations.

Management of A. petiolata is particularly challenging, often providing only short-term solutions or requiring frequent culling (17). A common approach is to remove flowers before seed set (17) (e.g., with timely mowing or using herbivores that target inflorescences, so that individuals neither survive nor reproduce). Our models readily enable us to quantify the efficacy of this approach. By multiplying the predicted flowering probability by a proportion less than one, we explored how variation in the percentage of inflorescences eliminated by a herbivore will affect population growth, while retaining the observed responses to habitat and climate. In our simulations, on average across the region, A. petiolata responded similarly to inflorescence predators in both open and closed habitat. Broadly, one would need to eliminate ∼99.5% of inflorescences in open habitat and ∼98% in closed habitat to achieve λ < 1 under current climate. Our herbivory simulation results show the difficulty of managing A. petiolata with biocontrol. These simulations corroborate similarly unrealistic efficiency determined using a density-dependent model (20). We, therefore, conclude that early detection provides the best chance of limiting invasions at small scales. At the regional scale, because populations are likely to persist on unmanaged lands, climate warming may provide the only reprieve from A. petiolata in southern NE.

Management of B. thunbergii, an ornamental species, has recently focused on recommending low-fecundity cultivars (24). To evaluate the efficacy of such introductions, we used published data on large adult seed production for two popular cultivars (lime glow and crimson pigmy) with similar size dependence of vital rates but lower seed production compared with the wild populations studied here (24). Notably, 15 cultivars (with atropurpurea and crimson velvet among the most widely planted) in that study had higher seed production at comparable sizes to WT B. thunbergii and would lead to greater population growth than currently observed if all other vital rates were equal (which we do not have data to evaluate). Only dwarf cultivars (with likely different size dependence of vital rates) and the two cultivars that we compare here had lower seed production than WT B. thunbergii in equal-aged individuals. Lime glow and crimson pigmy cultivars had seed production that was ∼58% of WT B. thunbergii (24); hence, for simulations, we multiplied our fitted seed production models by 58%. This reduction led to virtually no change in population growth across the region because of the extremely high seed production of B. thunbergii (SI Appendix, Fig. S27) (ref. 25 discusses similar conclusions). In fact, seed production in open habitat would require a reduction of 99.9% for WT B. thunbergii to achieve λ < 1. Although this precise number is contingent on a number of other model assumptions, including consistent growth, survival, and germination across cultivars, it is clear that only a dramatic reduction in seed output can reduce establishment risk.

A more practical approach would be to focus on further developing cultivars with simultaneously reduced seed production, germination, and seedling survival. If we treat these three vital rates as multiplicative (approximately true in our IPMs), an “effective” reduction of seed output of 99.9% could be achieved with, for example, cultivars with 58% seed production and 4% germination and seedling survival of WT B. thunbergii (or similarly, 10% for each of seed production, germination, and seedling survival). Currently, the greatest reductions across 21 cultivars of all growth habits 5 y after planting are 22% (seed production), 50% (germination), and 14% (seedling survival), which would achieve the required reduction in effective seed output to avoid population establishment (calculated from data in ref. 24). However, these reductions occur for different cultivars. All vital rates were not measured for all cultivars in ref. 24, and therefore, we cannot evaluate whether any single cultivar exhibits the necessary effective reduction in fecundity.

Conclusions.

An unexpected finding from our study was the predicted decline of A. petiolata in southern NE under future climate warming. Our illustration shows that climate change may mitigate invasions (2, 8); however, this mitigation is not an apparent benefit but rather, a red flag. Although A. petiolata may decline considerably in NE, invasion risk remains high in northern NE and southern Canada. We have not dodged the invasion, only shifted the focus. Furthermore, if a robust herb, such as A. petiolata, responds poorly to warming, native competitors could respond similarly (e.g., T. glabra is also expected to shift northward but without declining in the southern portion of its range) (Fig. 1 and SI Appendix, Fig. S19), providing many opportunities for other invaders to fill vacated niches or A. petiolata to exploit fluctuating resources should local adaptation arise or refugia persist. A. petiolata also modifies the soil where it has colonized; hence, understory communities seem to be at risk even after its disappearance (17, 26, 27).

Our demographic approach to predicting species distributions provides a number of improvements over existing approaches to forecasting invasions.

• i)We linked demographic variation to environmental variation and illustrated a number of insights even with a modest number of populations and individuals. Existing demographic studies rarely describe the environmental drivers underlying population change (28), perhaps because too few populations are studied to detect trends.
• ii)Our models can identify suitable habitat beyond current distribution boundaries, enabling targeted early detection. During potential invasions or when species are still likely spreading (e.g., A. petiolata) and management is most critical, occurrence models may underpredict suitable habitat or suffer from issues of small sample size.
• iii)Vital rates respond differently to climatic or habitat gradients (Fig. 5); hence, studying only a portion of the lifecycle can be misleading when demographic compensation exists.
• iv)Our models of potential distribution can be readily combined with popular simulation models of spatial dynamics (29), while avoiding the tenuous relationships between occurrence probability and demography (Fig. 2) on which existing versions of these simulation models rely (23).

Although our demographic experimental biogeography approach has provided a number of insights into the invasions studied here, modifications may be required for other types of predictions or other systems. For example, our demographic models are appropriate for predicting potential distributions that provide a first-order assessment of species’ niches and invasion risk, whereas realized niches will depend critically on other processes, such as dispersal or biotic interactions. Notably, we also focus on population establishment, and the persistence and impacts of such populations will likely depend on density dependence and competition. Adding combinations of these processes relevant for other questions would require embedding our demographic model in a simulation framework, such as that in refs. 30 and 31. Dispersal and competition may also be critical to understanding spatial patterns at higher spatial resolution, although it is worth noting that competition might reasonably be omitted from models where a species invades an apparently vacant niche, such as with A. petiolata here.

Using experimental biogeography to parameterize population models is a general approach with many practical advantages for exploring dynamic responses to global change for both invasive and native species. For potentially invasive species, establishment is a key to invasion, which motivates a focus on early portions of the lifecycle. Pairing experimental approaches with field observations of older/larger individuals allows one to efficiently characterize the full lifecycle across environmental gradients and evaluate the influence of management activities targeting different life stages. A short-term focus on establishment also allows projects that are practical for typical funding cycles or postgraduate programs, while offering relatively rapid assessment of invasion risk that can influence management tactics early in an invasion. Our approach, therefore, provides the practical links to understand the drivers of invasion, forecast potential distributions, and evaluate management strategies.

Materials and Methods

Despite recent attention to process-based range modeling (32, 33), few biogeographic studies have incorporated demography (34, 35). When species’ ranges are shifting (e.g., in response to climate change or during invasion), population growth rate is key to understanding establishment risk and local abundance variation. Using our experimental biogeography approach, we developed structured population models to predict the establishment dynamics of an invasive monocarpic biennial mustard (A. petiolata), a woody shrub (B. thunbergii), and their respective native ecological analogs. Using a series of transplant plots, we collected demographic and environmental data over three growing seasons to infer individual-level vital rates. For the long-lived shrubs, we supplemented those with additional field observations of adults (SI Appendix, section A). The invasive species were chosen, because they are pervasive, widespread, problematic invasive species throughout NE that contrast in growth form. Both are predicted to spread across NE* (36) but so far, have spread to a fraction of the projected area. The paired native species were chosen as the closest ecological analogs available in our study area, but we note that the match was imperfect and discuss below how native/invasive differences drive invasion. By transplanting the invasive species into locations that span the regional environmental gradients, we were able to build demographic models that relate vital rates to environmental conditions. We divided environmental factors into climatic (temperature and precipitation) and microsite (light, N, and pH) variation (details are in SI Appendix, section A) to differentiate regional- from local-scale invasion patterns (2). From these inferred demographic mechanisms, we predicted invasion risk beyond the current invasion fronts.

To quantify invasion risk, we focused on establishment, because it is the most important process in determining invasion success (37, 38). Focusing on establishment allows rapid risk assessments, which are particularly critical for managing invasions when a species is already known to be an invasion threat elsewhere. A focus on population establishment facilitates botanical experiments, because results become available rapidly and individuals can be culled before they risk reproduction. For long-lived species, experimental work can be paired with observational data from established populations to capture later life history stages.

Study Species.

A. petiolata (garlic mustard) is a monocarpic biennial mustard (Brassicaseae) that performs best in forest understories and dominates the herbaceous understory layer, where it invades (39). As a monocarpic biennial, it produces a rosette in its first year and typically reaches reproduction in its second year, producing copious seeds.

T. glabra (syn. A. glabra; tower mustard) is a monocarpic biennial mustard (Brassicaseae) native to most of North America and the closest native ecological analog that we could find to garlic mustard. Although it has been found growing alongside A. petiolata in NE at forest edges and disturbed meadows, T. glabra generally performs best in dry fields, open woodlands, and disturbed soils (16).

B. thunbergii (Japanese barberry) is a woody shrub common in NE, where it is often found in dense populations in forest understories and tolerates a wide range of soil conditions (21, 22).

L. benzoin (northern spicebush) is a medium to tall deciduous, native, woody, shade-tolerant shrub that can be found across a variety of different habitats in eastern North America (40). L. benzoin is common in moist forest understories (16) and was chosen as an ecological analog, because it co-occurs with B. thunbergii in forest understories and rich bottomlands (16) (although only B. thunbergii occurs in drier edge habitat). These woody species provide a useful contrast to the biennials for forecasting invasions with our experimental demographic approach, because their entire life history is not readily captured within the experimental design over timeframes that are likely useful for mitigating invasions.

Demography Data.

Twenty-one experimental study sites encompassed a climate and land use gradient in five regions across NE (SI Appendix, Fig. S1 and Table S1). The study region spanned a climatic gradient in mean temperature in the warmest month and mean May (spring) precipitation (SI Appendix, Fig. S1), which were the least correlated (0.22) measures of temperature and precipitation that we considered. Given our sample size, including more correlated predictors would have led to parameter identifiability issues. Sites within regions were chosen to characterize variation in microsite conditions, including light [from full sunlight to <1% transmission characterized by photosynthetically active radiation (PAR)] and soil nutrients (from circumneutral calcareous soils to poor acidic or sandy soils characterized by low soil pH and N) (12, 41) (data collection details are in SI Appendix, section A). Plot locations included deciduous, mixed deciduous, and coniferous forests; open agricultural landscapes; open sandy disturbed soils; and forest canopy gaps.

Seeds were collected from multiple wild populations throughout NE. Seeds were germinated in the greenhouse, and 6-wk old seedlings were planted out in late spring of 2009 (147 individuals per species total). A second round of planting for the biennials occurred the following spring (147 individuals per species). Dead woody seedlings were also replanted the following spring (65 L. benzoin and 80 B. thunbergii). Seedlings were planted 1 m apart, with seven individuals per species per plot. Plants were allowed to experience natural biotic conditions of the site after an initial clearing with minimal disturbance (which allowed transplant).

We combined our experimental data with three additional datasets comprising field observations and additional plantings (SI Appendix, sections E and F) for the long-lived woody species, because the length of our study did not allow us to estimate seed production, flowering probability, or adult growth/survival. The first dataset described vital rates for larger individuals grown from cuttings based on an additional transplant experiment conducted in central Connecticut and northern Vermont (42). Because these individuals were also culled before reaching maximum size, we included a second dataset of field observations of wild populations to estimate size-dependent flowering probability and seed production. Environmental covariates were omitted for seed production and flowering probability models (see below), because we had insufficient data spanning the environmental gradients [note that little variation in seed production for ornamental cultivars of Japanese barberry across the climatic gradients in our study region has been found (43)]. Finally, we used a third dataset collected over 21 y at five locations in central Connecticut to estimate the size dependence of adult survival of L. benzoin (1,647 year × individual observations). We used these to set an upper bound on adult mortality for both woody shrubs (details are in SI Appendix, section E). Because data were not available for adult mortality of B. thunbergii, we used the same upper bound based on L. benzoin data as a very conservative representative of large, robust shrubs. We note that this may be an underestimate of survival, because the US Department of Agriculture reports that B. thunbergii is expected to live 100–250 y and that mortality has been observed to be negligible after individuals reach a size of three stems (24, 44).

We established a field germination experiment alongside each transplant plot location by initially clearing litter and allowing seeds to experience otherwise natural conditions. Twenty-five seeds of mixed populations of each species were planted at each plot location in 2010 and 2011, and the number germinating was recorded. For biennials, we also recorded survival of germinants into summer to differentiate summer mortality from mortality over winter.

Landscape-Scale Data.

To characterize regional climate, we summarized 4-km resolution PRISM data (45) to describe current (2003–2012) and future (2041–2050) climate scenarios under representative concentration pathway 4.5 (RCP4.5) (considering six climate models) (SI Appendix, Fig. S1). These climate data were used as covariates in vital rate regressions (SI Appendix, section A), which allowed us to infer/extrapolate demographic predictions across the landscape to map species’ distributions (see below). For comparison with demographic predictions, we used occurrence data from GBIF, BISON, EDDMAPS, and iNaturalist obtained with the R package spocc (46) as well as IPANE (13), a database of invasive plants in NE.

Demography Models.

We used individual-level measurements of survival, growth, flowering, seed production, germination, and offspring size to build regressions that describe how each varies as a function of individual size (when relevant), microsite conditions, and climatic conditions (available as Datasets S1 and S2). For each regression, a model of the form vital rate ∼ individual size + climate + local habitat was fit to individual year observations. Models for germination and offspring size did not include the individual size predictor (because it was not relevant), whereas flowering and seed production models for the woody shrubs did not include climate and habitat predictors, because those data were not available. Climate predictors included mean temperature during the warmest month and mean May precipitation. Habitat predictors included PAR, soil N, and soil pH (data collection details are in SI Appendix, section A). Models implicitly include the assumption that species-level responses to each of these predictors dominate population-level (ecotypic) variation in responses. Linear terms were included for each predictor; our sample size did not permit exploration of more complex functional relationships (one exception is the flowering model in SI Appendix, Fig. S4).

The vital rate models were combined in IPMs (cf. 14, 15) to predict population growth rate after arrival of seeds. IPMs are stage-classified population projections models, analogous to matrix population models (47), where the state variables defining individuals are described on a continuous scale (details are in SI Appendix, section B). In our study, the state variable was a measure of individual size based on log-transformed basal rosette area for the biennials (millimeters²) and log-transformed cylindrical area of the shrubs (millimeters³). For the biennials, we included age (years) as a state variable to impose a 2-y lifecycle. We used deterministic simulations, because environmental covariates represented average climate. Note that, although our vital rate models correlate individual state with environment, mechanistic insights emerge from combining these demographic components and studying emergent properties at the population level.

Our predictions focus on population growth during establishment, because λ at low density is expected to predict potential distribution well (48). The invasive species in our study are known to exhibit prolific population growth in much of our study area (22, 49); hence, we expect that population establishment is the bottleneck limiting invasion. To build maps of λ, we supplied the environmental covariates in 4 × 4-km grid cells across NE to the vital rate regressions and used these predictions as the basis for an IPM in each grid cell (following methods in ref. 35). Microsite conditions were simulated under favorable conditions of PAR, N, and pH depending on species’ performance (at standardized values of ±1.5; henceforth referred to as “favorable habitat”) (SI Appendix, section A). Hence, our maps should be interpreted as conditional on availability of favorable local habitat. For biennials, our explorations found that transient dynamics lasted less than 10 y, and therefore, we used the asymptotic population growth rate. For shrubs, we simulated dynamics for 20 y after the introduction of 20 seeds as a representative way to characterize transient dynamics after arrival. We report the average value of λ over this simulation and interpret cells with λ > 1 as part of the species’ potential distribution (cf. 50).

Occurrence Models.

A key to justifying our experimental approach is validation and comparison with less costly alternatives. Although ecologists generally agree that mechanistic models are preferable to correlative ones, they represent a double-edged sword. The increased understanding gained from a mechanistic model comes at the expense that one must correctly quantify all of the dominant mechanisms to develop a reliable forecasting model.

To evaluate how our demographic predictions differ from the more traditional range modeling approach, we built occurrence models using all available online occurrence data for A. petiolata and B. thunbergii. We used Maxent (51), because absence data are not available at the scale of our study, and we included the same two climatic predictors, accounted for sampling bias, and chose settings to allow flexible but smooth models (using only hinge features) that would have the potential to identify response curves similar to those in Fig. 3 should they exist (additional modeling details are in SI Appendix, section G). Flexible response curves offer the ability to describe more complex responses than we could study with demographic models (because of identifiability) while enforcing smoothness (and avoiding overfitting) and better characterizing the climatic niche (52). To make comparisons, it is important to note that occurrence model predictions are shown in terms of ROR (i.e., given a presence, the relative probability that it was drawn from each cell) (53, 54). ROR does not allow one to predict where a species occurs, only which locations are more likely than others (53, 54). Furthermore, limited comparisons of occurrence predictions with demographic predictions at biogeographic scales have showed very weak correlations (23); hence, differences are expected from these modeling approaches. Therefore, we relied on qualitative comparisons of the shapes of the fitted response curves and the geographic pattern of habitat suitability predicted by each modeling approach.