In a Clamshell
Ecological niche modeling is a recently developed tool for predicting the distribution of organisms based on their environmental limitations. Its value in invasive species research is the ability to forecast the potential geographic distribution of a species in the introduced range based on occurrence points (and absence points for some methods) and appropriate environmental data layers. Such predictions are important for understanding the invasion process and for developing effective intervention strategies to reduce the ecological and economic impacts of invasions. Due to the variety of little tested techniques for ecological niche modeling, we provide a review of the currently available techniques and their strengths and weaknesses. One focus of this chapter is the set of unique challenges of environmental niche models for freshwater species. We provide an in-depth description of one of the most commonly used methods (genetic algorithm for rule-set prediction, or GARP), including studies that evaluate the approach. We also list potential data sources for researchers.
Ecological niche modeling seeks to address a fundamental question in ecology: What is the potential geographic distribution of an organism? The answer is central to a wide range of environmental questions (Guisan and Thuiller 2005). It can identify potential conservation zones or guide selection of areas for the reintroduction of threatened and endangered species (Araujo et al. 2004) based on their potential range. Similarly, the modeling of suitable habitat can reveal at which unsurveyed sites rare species are most likely to occur (Elith and Burgman 2002; Raxworthy et al. 2003). Based on scenarios of future climate change, land use, and habitat availability, ecological niche modeling can predict potential future shifts in species distributions (Thuiller et al. 2004). It can also provide background information to sharpen the focus of management efforts and risk analyses (Leung et al. 2002; Drake and Bossenbroek 2004).
Ecological niche modeling can also be used to predict suitable ranges for invasive species in an introduced habitat (Peterson 2003; Peterson et al. 2003; Drake and Bossenbroek 2004; Iguchi et al. 2004; Herborg et al. 2007c). With the increasing availability of high-resolution spatial environmental data sets, more accessible species distribution data, and faster computers, an increasing number of scientists, conservationists, and managers are using ecological niche models to predict invasions. One controversial, but potentially advantageous, feature of ecological niche modeling is its ability to predict distributions of little-studied species, based solely on georeferenced presence points (and absence points for some approaches) and spatially explicit environmental data for a range of parameters deemed important in determining the species range.
A good example of the scope of ecological niche models is a recent study of the invasive Chinese mitten crab, which has a long history of invasion in European rivers. Besides its catadromous life cycle (it grows and matures in freshwater but has to return to the sea to reproduce), little else is known of its environmental tolerances and ecological requirements. Since the recent colonization of San Francisco Bay by the species, there is growing concern for its spread throughout North America. The initial steps for developing an ecological niche model included the collection of presence data for Chinese mitten crab in its native Asian range and its introduced European range, as well as climate data (mean, minimum, and maximum air temperature, frost frequency, precipitation, wet day index), hydrological data (river discharge and river temperature), and oceanographic data (spring ocean temperature) (see table 4.1 for details). Since the available data for the native range was limited and the invaded range in Europe is still expanding, a widely tested ecological niche modeling approach (genetic algorithm for rule-set prediction, or GARP) using presence-only data was chosen. After identifying which environmental variables contributed significantly to model performance, the model was validated by successfully predicting the invaded European range based on the native range (Herborg et al. 2007c). To identify ports in the United States at the highest risk of mitten crab establishment, two separate predictions, one based on the native Asian range and one on the invaded European range, were combined with the amount of ballast water released into each major U.S. port (see figure 4.1; Herborg et al. 2007a). The models identified Chesapeake Bay as the location with the highest risk in the United States, a prediction that has recently been substantiated by the discovery of several mitten crabs in that area (Ruiz et al. 2006). This example highlights how combining biological factors (species potential distribution) and economic activities (maritime trade) can cause particular locations (ports and their associated rivers) to be at particular risk of introduction. Once high-risk locations are identified, economic analysis of the effect particular invaders can have on valuable industries can be initiated.
Our objective here is to highlight the importance of ecological niche modeling as a tool for understanding the invasion process and for developing effective intervention strategies to reduce the impact of invasions. Because niche modeling is a rapidly developing methodology, there is ongoing discussion about the most appropriate and accurate methodology (Elith et al. 2006). We provide a mini-review of the available techniques and their strengths and weaknesses. The methods discussed here can be applied to terrestrial or aquatic environments, but most of our examples are based on aquatic organisms due to the expertise of the authors and the particular challenges researchers face when developing an ecological niche model for an aquatic species.
TABLE 4.1. List of environmental niche model methods and their data requirements (modified from Elith et al. 2006).
FIGURE 4.1.
Environmental suitability, on a scale from 0 to 100, combined with inland migration distances for the Chinese mitten crab. Reprinted from Herborg et al. (2007a), with permission of the Ecological Society of America.
Environmental niche modeling is a rapidly developing area of research, and the variety of available methods is overwhelming. While there is no recipe for selecting the most appropriate method for a particular problem, there are several considerations that can provide guidance to a potential user. An important first step is to determine if the model will include only presence data, presence and absence data, or biological community information in addition to presence/absence data (see table 4.1). Clearly, presence data is prerequisite for any prediction and either can be obtained from observations or, in the absence of observation points, can be randomly sub-sampled from the range of a species using geographic information systems (Herborg et al. 2007b). While predictions based on this second approach will likely lead to less accurate results, in some cases it may be the only option given the available information. Reliable absence data are more difficult to obtain, and their availability depends on the type of species, its habitat, and the amount of research conducted. Additionally, species can be absent from areas with suitable environments because of local extirpation, historical events such as glaciation or plate tectonics, and dispersal constraints (Peterson et al. 1999). Absence data for species in their invaded range is very problematic, since the species is unlikely to have spread through its potential range.
The second step is selecting an environmental niche model method that has been tested successfully given the type of data available. Two recent studies have tested and compared the performance of a large number of environmental niche modeling methods (Seguardo and Auraujo 2004; Elith et al. 2006) in detail. Elith et al. (2006) compared widely used approaches, such as GARP, BIOCLIM, and generalized linear models (GLMs), with more recent or rarely used models with respect to their ability to model current distributions of species accurately. The models were validated based on independent observations that were compared with model predictions. Although all modeling techniques performed generally well, several methods, such as MAXENT, the open modeler version of GARP, and regression-based approaches (multivariate adaptive regression splines, GLMs, general dissimilarity models) stood out from the rest. Another study looked at the performance of eight ecological niche modeling approaches (Seguardo and Auraujo 2004), but only three were in common with Elith et al. (2006). In that study, neural networks performed the best, but not significantly better than more traditional methods like GLMs and generalized additive models (GAMs). The worst performers were DOMAIN and BIOMAP.
Considering the wide variety of methods available, we discuss only a subset here, beginning with statistical approaches to environmental niche models, including GLMs, GAMs, and nonparametric multiplicative regression (NPMR). GLMs are a class of regression models in which the mean response is a nonlinear “link” function of the predictors, and the data may belong to one of several distributions, that is, need not be normally distributed. Logistic regression, a model for binary data in which covariates linearly affect the logit-transformed data, is a type of GLM that is based on the logit link function and binomial sampling distribution (McCullagh and Nelder 1989), resulting in a sigmoidal model. In practice, logistic regression is the main GLM used for niche modeling (Guisan et al. 2002), but others are available, for example, log-linear models. Since the simple logistic regression assumes independence, a version called the autologistic model is available to account for spatial autocorrelation (Augustin et al. 1996). GAMs are a generalization of GLMs in which the link function is replaced with a nonparametric estimator (Hastie and Tibshirani 1999; Wood 2006), making the underlying model considerably more flexible. As with GLMs, the logistic regression version is most commonly used for niche modeling (Wood and Augustin 2002). Spatial autocorrelation of the dependent variables is readily incorporated into GAMs, in which case they are sometimes called geoadditive models (Wood and Augustin 2002; Kammann and Wand 2003). NPMR takes this approach a step further by modeling the interaction of environmental covariates multiplicatively (McCune 2006). Finally, a new nonparametric approach, MAXENT, has been introduced based on the maximum entropy principle (Phillips et al. 2006). Like statistical density estimation, MAXENT aims to identify a probability density of species occurrence.
Nonstatistical methods use computation to find rule-based or model-based criteria for discriminating suitable from unsuitable habitat (e.g., GARP, CLIMEX, artificial neural networks [ANNs]). GARP requires input of binary presence data and pseudo-absence data. Pseudo-absence data are generated by GARP from a random selection of locations without confirmed species presence (Stockwell 1999), so they are commonly simulated, and much of the discussion of GARP has focused on this feature. GARP represents a species niche according to a collection of logical rules that “evolve” over the course of model iterations to provide an optimal fit. In contrast with GARP’s rule-based approach, CLIMEX is a model-based approach (Sutherst and Maywald 1985). The model is a biologically motivated ecological index that supposes climate factors are the most important determinants of species fitness and distribution. The aim of the interactive CLIMEX software is to identify the parameters of submodels (e.g., thresholds, slopes) that best match observed and predicted species distributions. CLIMEX is much less automated and depends more heavily on user-supplied guidance (expert knowledge) than other methods. Like GARP, the classification mode uses a genetic algorithm to estimate the parameters of the stress index component of the CLIMEX model. Estimation is accomplished by iterative model tuning and evaluation directed entirely by the user.
ANNs are a class of algorithms for pattern recognition in data. ANNs may be applied to classification or boundary estimation problems. For ecological niche modeling applications, classification has been the more common approach. ANNs are extremely flexible and may be applied to data structures that would be awkward to address with other methods. The downside is that ANNs can be very sensitive to model specification and user-supplied optimization parameters, thus requiring considerable expertise to be used effectively.
BIOCLIM is another nonstatistical approach, which is based on a multidimensional bounding box defined by the empirical quantiles of a data set. A problem with the BIOCLIM approach is that for most species there will be physiological trade-offs such that the corners of the box are generally uninhabitable; that is, an organism might tolerate extreme conditions in one variable if all the other variables are ideal, but the organism cannot tolerate extremes in all environments simultaneously. This problem is exacerbated when the dimensionality of the data is increased (i.e., as niche axes are added), but this can be addressed in a number of ways. One method introduced by Guo et al. (2005) to address this problem is the support vector machine (SVM). SVMs were developed for automated classification and belong to a class of machine learning algorithms called kernel methods. The basic idea is that the raw data are presumed to exhibit some property that is very complicated in the natural dimensions in which the data were collected. However, by transforming the data with a functional relation called a kernel into a new, higher dimensional space, these properties appear much simpler.
Independent of the environmental niche model selected, environmental data relevant to the species in question is essential for modeling ecological niches. While an increasing amount of environmental data are available at broad extents or even globally, the freshwater environment probably has the least data available. Global coverages for watersheds of various resolutions, flow directions, slope, aspect, and other elevation-derived parameters have been generated (HYDRO1k database; table 4.2); only limited data on water temperature, flow rates and volumes, water chemistry, benthic substrate, and so on, are available on a broader scale (see table 4.2). Many of the data sources only have limited value in smaller scale analysis and/or require lengthy processing into data layers. Small-scale hydrological data might be more readily available through government agencies, in particular, in the United States.
As with environmental data, species distribution data can be gathered from a range of sources. Clearly, museum records and scientific publications are a primary source of distribution data, but native distributions in particular are rarely published in the recent primary literature due to the perceived limited scientific value of such reports. Several national and international databases on the current distribution of invasive species are available online (table 4.3). In the case of recent invasions, absence data are of limited value because in most cases it will be impossible to determine if the species is absent due to environmental conditions or lack of colonization opportunities.
Ecological niche modeling has been applied to several aquatic invasive species. Zebra mussels (Dreissena polymorpha) have caused large economical (Leung et al. 2002) and ecological (Ricciardi et al. 1998) impacts since their introduction into the Laurentian Great Lakes in North America. A GARP model was applied to the invaded distribution in the United States to predict the potential distribution of the species in North America (Drake and Bossenbroek 2004). The contribution of an initial set of geological, hydrological, and terrestrial climate layers (bedrock geology, elevation, flow accumulation, frost frequency, maximum temperature, precipitation, slope, solar radiation, and surface geology; see table 4.2) on the numbers of false positives (omission error) and false negatives (commission error) was utilized to select a subset of layers with higher predictive accuracy. Further combinations of the subset of environmental layers highlighted a strong effect of elevation in some predictions. The model (figure 4.2) concluded that while large parts of the Colorado, Columbia, and Missouri rivers are unsuitable environments, coastal areas along the West Coast, including the San Joaquin and Sacramento rivers and the lower reaches of the Columbia, are suitable for establishment (Drake and Bossenbroek 2004).
TABLE 4.2. Sources of environmental layers relevant for aquatic ecological niche modeling.
TABLE 4.3. Sources of species distribution data.
FIGURE 4.2.
Prediction of suitable environment for the zebra mussel in United States based on its current invaded range. Reprinted from Drake and Bossenbroek (2004), with permission of the American Institute of Biological Sciences.
Largemouth bass (Micropterus salmoides) and smallmouth bass (M. dolomieu) are two North American predatory fishes with a high impact on the native fauna of Japan. Topographical, hydrological, land-use, land-cover, and terrestrial climate data (elevation, slope, aspect, topographic index, land use, land cover, percent tree cover, flow accumulation, and flow direction; table 4.2) were utilized to predict the potential range of both species using GARP (Iguchi et al. 2004). Due to the widespread invasion of these two game fish species, a validation of the predictions was possible (see the following section). A different approach (artificial neural network) was taken to predict the vulnerability of lakes in Ontario to smallmouth bass colonization, based on a wide range of climate, habitat, and biological factors (Vander Zanden et al. 2004).
Eurasian ruffe (Gymnocephalus cernuus) and rainbow smelt (Osmerus mordax) are both recent invaders in North America, and ecological niche modeling based on ecological and terrestrial climate data (annual precipitation, elevation, ground frost frequency, slope, wet day frequency, and maximum, mean, and minimum air temperature; table 4.2) was used to predict their potential range (Drake and Lodge 2006). A subset of layers was selected for each species according to Drake and Bossenbroek (2004), and final predictions indicated large areas of suitable environments in North America (figure 4.3), underlying the importance of rapid management actions (Drake and Lodge 2006).
FIGURE 4.3.
The potential distribution of the Eurasian ruffe in North America. Reprinted from Drake and Lodge (2006), with permission of the American Fisheries Society.
The combination of ecological niche modeling and estimates of propagule pressure can provide information on locations where a species can survive and where it is likely to arrive. Predictions of suitable environments (figure 4.1) and the volume of ballast water released were used to quantify the relative risk of establishment of the Chinese mitten crab (Eriocheir sinensis) in U.S. ports (Herborg et al. 2007a).
One challenge in using ecological niche modeling to forecast potential geographic range of invasive species is that experimental validations of model predictions are not ethically acceptable. For example, releasing zebra mussels into a lake where they are absent, but where models predict they can persist, for the sake of model validation is not permissible. More creative approaches are required to surmount the problem of model validation. Several validation approaches have been employed, including (1) generating models based on a few known point localities of a cryptic species and then searching for the species at sites predicted suitable by the model, (2) dividing highly resolved and detailed data sets into training and validation subsets to test the ability of the model to accurately predict what is already known but has been reserved from model training, and (3) the generation of ecological niche modeling based on simulated point occurrence data according to investigator specified rule-sets and then determining how accurately the model recovers the rule-sets.
While several studies have compared different modeling approaches (see “Data Sources and Approaches,” above), validating predictions of invasive species ranges is inherently difficult, and only limited studies are available. Iguchi et al. (2004) validated their GARP-generated ecological niche modeling of smallmouth and large-mouth bass in their native North American range using independent point occurrence data (i.e., occurrence points not used during model development). Their analyses showed that the predictive ability of the models is statistically significant (p < 0.001) (Iguchi et al. 2004). This model, validated for the native range of the bass, is then projected on the environmental conditions in Japan to predict the geographic range of both bass species there, where bass are not native. While limited data were available for the validation of model projections, the models predicted Japanese occurrences accurately.
Herborg et al. (2007c) used native occurrences of the Chinese mitten crab (Eriocheir sinensis) to validate a GARP prediction of environmental suitability based on the well-documented invasion in Europe. The comparison of the model prediction and the observed occurrences indicated high accuracy in the prediction of occurrence points: 84% of all reported occurrences were in areas predicted suitable by >80% of the models. Also, watersheds with established populations of mitten crabs had significantly higher environmental suitability in the model than did uninvaded watersheds.
Species distribution modeling is conceptually connected to Hutchinson’s (1957) ideas on fundamental and realized niches, where the fundamental (potential) niche represents all environments under which the species can persist, whereas the realized (occupied) niche is the area where the species is actually present. The discrepancy between the occupied and the potential niches can occur because of biogeographic dispersal barriers, glacial refugias (e.g., Nekola 1999), and biotic interactions (e.g., competition, predation) that result in range limitation. These differences can be a source of error associated with ecological niche modeling because the current range of a species, whose environmental characteristics are used to calibrate the model of its potential range, may not represent the potential niche.
Freshwater species are particularly likely to be restricted by dispersal limitations between watersheds and hydrological barriers. Thus, for many aquatic species, numerous unoccupied sites may exist that are environmentally different from presently inhabited sites but that would provide habitat sufficient to maintain a positive net population growth rate. When this is the case, any estimates of potential species distribution based on point occurrence observations will be overly conservative. Human-aided long-range dispersal, which often accompanies invasion events, can overcome this type of range limitation and could lead to establishments of species in locations not predicted as suitable by ecological niche models.
Another challenge associated with ecological niche modeling is that models are generally constructed using only presence data. True absence data are extremely difficult to collect and are almost never available. These data limitations may obscure the signal of the ecological requirements of a species and lead to inaccuracies in model predictions (Brotons et al. 2004). Also of concern is that an ecological niche model would ideally be based on data for each cell of a hypothetical grid overlaid on the current range of the species. It would also be desirable to have the frequency of points in the data set be representative of the frequency of species occurrence in the various regions and types of environments being used to fit the model. These idealistic requirements pose a particular challenge for aquatic species because such species are difficult, relative to many terrestrial species, to survey. Having a stratified random sample of species occurrence is critical to producing accurate ecological niche models because most modeling techniques are sensitive to sample size and how evenly the data are distributed (Stockwell and Peterson 2002). Even invasive aquatic species, which typically attain large population sizes as they reach nuisance levels at a particular site, may persist as small populations, unnoticed for many years, at other sites. Thus, initially cryptic populations can limit the quality of ecological niche models for aquatic invasive species because a subset of suitable sites will not be included in the point occurrence data set that is used to train and validate the model. In cases where substantial and relatively representative point occurrence data are available, this is a minor concern, but this is not so when the known sites of established populations are few, as is often the case early in invasions.
Training and projecting environmental niche models for aquatic species are also limited by the paucity of aquatic environmental data of suitable extent or resolution. Broad, continental-scale, digitized data are available for climatic and terrestrial parameters. While freshwater habitat data might be available for smaller scale studies, forecasts of potential distribution for aquatic species on a larger scale have relied on climatic and terrestrial parameters as surrogates for aquatic conditions (Drake and Bossenbroek 2004; Iguchi et al. 2004). However, a comparison of predictions of fish distributions based on environmental factors at broad spatial scales and small-scale aquatic habitat data revealed that broad-scale variables can predict fish distribution successfully (Marsh-Matthews and Matthews 2000).
Another concern is the selection of environmental parameters included at the outset of the ecological niche modeling process. The inclusion of some parameters can have a nontrivial influence on the model’s final predictions. For example, in predicting the potential distribution of zebra mussels in the United States, Drake and Bossenbroek (2004) included elevation as a predictor variable. In the original model results, the pattern of potential habitat was clearly related to elevation, even though there is no reason to believe that zebra mussels actually respond to elevation. In other words, these models are only as good as the data with which they are developed. Removal or addition of a single parameter as a potential predictor of species distribution at the outset of a modeling exercise can result in substantial discrepancies in modeled predictions of potential distribution (Guisan and Zimmerman 2000).
Some ecological niche modeling approaches are computationally intensive, which, despite ever-increasing processor speeds and ever-growing RAM banks, can limit their usefulness. Managing and interpreting the copious output of some environmental niche modeling methods can be burdensome. Finding a balance between model complexity and interpretability will require continued effort by ecological niche modelers.
Ecological niche modeling has the potential to be a valuable tool for the management of invasive species, due to its ability to predict the potential geographic distribution of a species before an introduction. Such predictions allow a prioritization of management effort for a range of invaders based on their potential distribution in an introduced range. For example, the predictions of potential distributions of a number of species of snakeheads and Asian carps helped identify those fish species that pose the highest risk for Canadian freshwater systems (Herborg et al. 2007b). Another important application of environmental niche models for managers and bioeconomic analysis is the possibility to incorporate future scenarios for climate change, altered land use, or management, species, or habitat modification into predictions of potential distributions of invaders. A lake-based environmental niche model incorporating a range of different climate change scenarios predicted the range expansion of smallmouth bass in Canada, highlighting its potential impact on northern fish communities in Canada (Sharma et al. 2007). Other scenarios that could be incorporated in future environmental niche models could include a range of potential management strategies that would alter habitat suitability and availability. This would allow a quantitative test of different management strategies, crucial for a cost-benefit analysis. As is noted often throughout these chapters, the spread of invasive species has and will continue to have a large economic impact on society. The potential spread of invasive species, such as zebra mussels and the emerald ash borer, have resulted in large, multistate (if not international) education programs to prevent the further spread of these organisms (e.g., the 100th Meridian Initiative [www.100thmeridian.org/] and the Emerald Ash Borer website [www.emeraldashborer.info/]). Having accurate predictions of the potential habitat of these species would allow focused outreach and prevention efforts to areas considered most at risk—there is no need to spend money to educate people in a pine forest about the risks of the emerald ash borer.
Predicting the potential economic benefits of slowing the spread of an invasive species requires the integration of forecasts of potential range, consideration of species-specific dispersal capabilities, and evaluation of potential economic impacts and the costs and benefits of different control and prevention strategies. The integration of these issues is a general goal of this book and has been specifically addressed with regard to the spread of zebra mussels to the western United States (see chapter 12) and the emerald ash borer in Michigan and Ohio (see chapter 6). The general objectives of these research projects are twofold: (1) to provide estimates of the regional economic impact that invasive species will inflict upon the region of concern and (2) to provide policy makers with quantitative guidance for cost-effective alternative strategies to control, prevent, or slow the spread of these species.
Achieving these objectives requires that we (1) determine the potential geographic range of the invasive species, (2) predict its spread via natural and human-mediated dispersal, (3) estimate the economic impact of the invasion, and (4) determine its regional economic consequences. Determining the potential geographic range involves generating an environmental niche model. In the case of the emerald ash borer, it is currently assumed that the potential habitat of the borer is equivalent to the distribution of ash trees, which can be determined by U.S. Forest Service data (Iverson et al. 2006). As the emerald ash borer spreads, or as more information is discovered about its native range, environmental niche models can be used to make refined predictions of potential habitat. Estimating the economic impact of the invasive species involves estimating the regional labor and capital employment in a spatially explicit manner. For the emerald ash borer, one assessment suggests that their damage to the urban communities of Ohio, based on losses in landscape value and the costs of tree removal and replacement, could reach $2 billion to $8 billion (Sydnor et al. 2007). Determining the regional economic consequences as the invasive species spreads can be done through the development of a regional computable general equilibrium model (see chapter 12 for details). The emerald ash borer and zebra mussel projects thus provide clear examples of the need for accurate assessments of the ecological niche of an invasive species for management purposes.
The ecological niche modeling techniques discussed above have not yet been directly linked to economic analyses, but they have been used to make recommendations about the management of invasive species. An ecological niche model of the Asian longhorn beetle suggested that most areas of the Pacific Coast of the United States are unsuitable habitat, which would therefore allow a focus of all management efforts on the Atlantic Coast (Peterson and Vieglais 2001). A model combining ecological niche modeling and ballast water discharge calculated the relative invasion risk of Chinese mitten crab for major ports in the United States (figure 4.1). Herborg et al. (2007a) determined Norfolk, Baltimore, and Portland as having the highest risk, while other major ports such as Los Angeles–Long Beach have a very low risk, thus identifying key areas for monitoring and risk management. An ecological niche model of potential zebra mussel habitat in the United States found less suitable habitat west of the 100th meridian than previously predicted (figure 4.3; but see Whittier et al. 2008) but concluded that the Columbia, San Joaquin, and Sacramento river drainages were at risk. These results had a direct effect on the management strategy of the 100th Meridian program, which aimed to stop the spread of zebra mussels in the United States (Drake and Bossenbroek 2004). Ecological niche modeling was used to produce a potential habitat map as a template for a vector model of the spread of West Nile virus by birds and mosquitoes. This approach advanced understanding of how the pathogen is spreading throughout the United States (Peterson et al. 2003). Arriaga et al. (2004) identified the risk posed by the buffelgrass invasions of desert scrub in the arid and semiarid regions of the northern Mexican Sonoran Desert. These examples demonstrate the usefulness of ecological niche modeling for informing management strategies for specific species.
Ecological niche models have also been used to make management recommendations for multiple species or species groups. Peterson et al. (2003) demonstrated that native ranges of four plant species could be used to predict their potential ranges in North America. Each of the four species analyzed by Peterson et al. (2003) were currently in North America, and the niche model predictions contained the current distribution. Two of the species, garlic mustard and Russian olive, were predicted to have a potential distribution much larger than currently exists in North America, suggesting that continued spread is likely. The high predictive ability of the Peterson et al. (2003) results suggest that niche modeling can be successful in estimating the potential habitat of species that have not yet been introduced. Likewise, Thuiller et al. (2005) found that for 96 plant taxa (species and subspecies) the distribution data in the native country enabled accurate predictions of the climate in regions where these taxa are invasive. The ability to accurately model plant niches to predict the invasive range of these taxa highlights the potential of ecological niche modeling as an important management tool for other taxonomic groups, including fishes (Kolar 2004).
While there are currently no formal applications linking environmental niche models and economic analysis, there is clearly opportunity for future research that would greatly benefit policy makers, managers, and the public. First, the economic impact of most invasive species is correlated with population abundance. Thus, if ecological niche modeling could be enhanced to estimate population abundance (i.e., the potential severity of impacts), economic predictions would have a range of potential impacts as opposed to merely presence or absence. For example, damage functions linking predicted population size with financial losses could be generated and incorporated in economic analyses. Second, risk assessments are often (if not inherently) subjective, and such subjectivity could affect the application of ecological niche models. Further involvement of economists in the interpretation of environmental niche models and resulting policy suggestions may help us to better understand the consequences of such subjectivity. Predictions of environmental niche models can vary substantially based on the variables chosen for the model (for an example, see Drake and Bossenbroek 2004). Therefore, if different analysts include different parameters when modeling the potential distribution of the same species, disagreements in range forecasts are possible. Moreover, even if potential range maps are identical, different analysts may draw different conclusions because their interpretations are based on their own inherent biases, of which they may be unaware. For example, property owners or natural resource managers with a vested interest in the consequences of environmental impacts of invasive species may overestimate the future risk posed by such threats (Burgman 2005). On the contrary, an industry that engages in activities that may introduce nonindigenous species or that may be affected in the future by an invasive species may be more conservative in their predictions, not wanting to invest in prevention or control earlier than is necessary. Economists are equipped to analyze the relative benefits of such diverse approaches to risk management (Shogren 2000).
Generally speaking, certain types of economic models can provide guidance on the timing and magnitude of many kinds of investments, from the allocation of capital in the stock market to public spending on HIV intervention programs (Zaric and Brandeau 2002). In the context of invasion biology, economic models could be used to address similar questions such as when preventative measures should be put in place, how much should be spent on outreach and education versus quarantine efforts or population control, and whether management activities produce a net economic benefit.
The ability of a model to accurately describe or predict nature can be limited by the quality of empirical observations used to generate the model. This is the case for ecological niche models. High-quality data sets of species presence, not to mention absence, for large geographic regions are desirable in producing and validating environmental niche models. While the majority of available occurrence data, particularly from the native range of a species, is maintained by museums, acquiring such data can be prohibitively expensive. Economists could help ecologists and the agencies that fund them determine how much to invest in obtaining accurate point locality data. Analyses of this sort could be based on the projected financial losses associated with a potential invader, the cost of data collection, and the prospects for model improvement with each additional data point available.
Improving the accuracy and reliability of ecological niche modeling will require continued effort from researchers in the field—both those that develop the theory of ecological niche modeling and those that test the validity of its predictions. This is a field of research with ample opportunities for collaboration among biologists, ecological modelers, and statisticians. Opportunities to collaborate with economists would also be valuable because further considerations of spatial predictions of the potential range of a nonnative species may enhance forecasts of the economic impacts of invasive species. In other words, if the abiotic conditions that predict the potential for species presence can be delimited on maps, then current risk and future change in risk should be predictable. This would allow economists to generate dynamic models of financial impacts. Furthermore, understanding the relationship between species presence and anthropogenic habitat change could lead to dynamic models of the locations of the suitable habitat conditions themselves, which may change in response to social trends. One example of this is irrigation in the Desert Southwest of the United States that has allowed the establishment of the red imported fire ant (Solenopsis invicta Buren) in that region, which would otherwise be too dry for this species (Morrison et al. 2004).
Additionally, several procedural measures can help to improve the accuracy and robustness of ecological niche modeling predictions. Given the effects of discrepancies between the observed and potential niche on model predictions, it is essential to be aware of the quality of data incorporated and its potential limitation. In the case of invasive species, models should be regenerated as the geographic extent of the invasion grows and more presence data become available.
Other areas in need of further investigation include the importance of variable selection and the effects of the relative size of geographic area used for model training. Also, the question of whether to include invaded range point localities in training data sets, as opposed to only native range data, remains open. The answer to this interesting question may depend on whether there has been local adaptation in the invaded range.
The areas needing the most attention in species distribution modeling, unsurprisingly, reflect the limitations of ecological niche modeling already discussed in this chapter. Araujo and Guisan (2006) discusses five areas needing attention, each corresponding to one or more of the current limitations of ecological niche modeling: (1) niche concept clarification, (2) sampling designs for the data used in model construction, (3) parameterization improvements, (4) techniques for understanding parameter contribution and selecting the best models, and (5) methods for evaluating models (Araujo and Guisan 2006).
Ecological niche modeling is an area of active research as well as a useful tool for understanding and dealing with biological invasions. The ecological modeling community continues to develop new techniques that will enhance the accuracy of ecological niche modeling (Elith et al. 2006), but it is already valuable for predicting the regions most at risk for invasion by particular species. Ecological niche modeling helps ecologists understand what factors determine the distribution of species, native or otherwise, and, among other things, test hypotheses about species evolution (Peterson et al. 1999). Managers and policy makers can also benefit from the use of environmental niche models. Management actions targeting different stages in the invasion process (see chapter 1, figure 1.1) can be informed by potential range forecasts. Identifying potential habitat for nonindigenous species can focus prevention and control efforts and can inform early detection/rapid response activities. Moreover, environmental niche models can guide these efforts at a variety of spatial scales, from global to regional. The formal incorporation of potential geographic ranges of invasive species in economic analyses is an exciting prospect. Some early efforts in this direction have been described here, but many important advances await further collaboration between ecologists and economists. Such advances will surely increase our capacity to efficiently reduce the negative effects of invasive species on society.
Acknowledgments This chapter was substantially improved by the editors of this book and reviews by Joanna McNulty, Jim Muirhead, and an anonymous reviewer. This material is based on work supported by the Integrated Systems for Invasive Species project (D. M. Lodge, principal investigator) funded by the National Science Foundation (DEB 02-13698). This is publication no. 2009-01 from the University of Toledo Lake Erie Center.
Araujo, M. B., P. J. Densham, and P. H. Williams. 2004. Representing species in reserves from patterns of assemblage diversity. Journal of Biogeography 31:1037–1050.
Araujo, M. B., and A. Guisan. 2006. Five (or so) challenges for species distribution modelling. Journal of Biogeography 33:1677–1688.
Arriaga, L., A. E. Castellanos, E. Moreno, and J. Alarcon. 2004. Potential ecological distribution of alien invasive species and risk assessment: a case study of buffel grass in arid regions of Mexico. Conservation Biology 18:1504–1514.
Augustin, N. H., M. A. Mugglestone, and S. T. Buckland. 1996. An autologistic model for the spatial distribution of wildlife. Journal of Applied Ecology 33:339–347.
Brotons, L., W. Thuiller, M. B. Araujo, and A. H. Hirzel. 2004. Presence-absence versus presence-only modelling methods for predicting bird habitat suitability. Ecography 27:437–448.
Burgman, M. A. 2005. Risks and decisions for conservation and environmental management. Cambridge University Press, Cambridge, UK.
Drake, J. M., and J. M. Bossenbroek. 2004. The potential distribution of zebra mussels in the United States. BioScience 54:931–941.
Drake, J. M., and D. M. Lodge. 2006. Forecasting potential distributions of nonindigenous species with a genetic algorithm. Fisheries 31:9–16.
Elith, J., and M. A. Burgman. 2002. Predictions and their validation: rare plants in the Central Highlands, Victoria, Australia. Pages 303–314 in J. M. Scott, P. J. Heglund, M. L. Morrisonet, J. B. Haufler, M. G. Raphael, W. A. Wall, and F. B. Samson, editors. Predicting species occurrences: issues of accuracy and scale. Island Press, Covelo, CA.
Elith, J., C. H. Graham, R. P. Anderson, M. Dudik, S. Ferrier, A. Guisan, R. J. Hijmans, F. Huettmann, J. R. Leathwick, A. Lehmann, J. Li, L. G. Lohman, B. A. Loiselle, G. Manion, C. Moritz, M. Nakamura, Y. Nakazawa, J. M. Overton, A. T. Peterson, S. J. Phillips, K. Richardson, R. Scachetti-Pereira, R. E. Schapire, J. Soberon, S. Williams, M. S. Wisz, and N. E. Zimmermann. 2006. Novel methods improve prediction of species’distributions from occurrence data. Ecography 29:129–151.
Guisan, A., T. C. Edwards, and T. Hastie. 2002. Generalized linear and generalized additive models in studies of species distributions: setting the scene. Ecological Modelling 157:89–100.
Guisan, A., and W. Thuiller. 2005. Predicting species distribution: offering more than simple habitat models. Ecology Letters 8:993–1009.
Guisan, A., and N. E. Zimmermann. 2000. Predictive habitat distribution models in ecology. Ecological Modelling 135:147–186.
Guo, Q., M. Kelly, and C. H. Graham. 2005. Support vector machines for predicting distribution of Sudden Oak Death in California. Ecological Modelling 182:75–90.
Hastie, T. J., and R. J. Tibshirani, R. J. 1999. Generalized additive models. Chapman and Hall/CRC, Boca Raton, FL.
Herborg, L. M., C. J. Jerde, D. M. Lodge, G. M. Ruiz, and H. J. MacIsaac. 2007a. Predicting invasion risk using measures of introduction effort and environmental niche model. Ecological Applications 17:663–674.
Herborg, L. M., N. E. Mandrak, B. Cudmore, and H. J. MacIsaac. 2007b. Comparative distribution and invasion risk of snakehead and Asian carp species in North America. Canadian Journal of Fisheries and Aquatic Sciences 64:1723–1735.
Herborg, L. M., D. A. Rudnick, Y. Siliang, D. M. Lodge, and H. J. MacIsaac. 2007c. Predicting the range of Chinese mitten crabs (Eriocheir sinensis) in Europe. Conservation Biology 21:1316–1323.
Hutchinson, G. E. 1957. Concluding remarks. Population studies: animal ecology and demography. Cold Spring Harbor Symposia on Quantitative Biology 22:415–427.
Iguchi, K., K. Matsuura, K. M. McNyset, A. T. Peterson, R. Scachetti-Pereira, K. A. Powers, D. A. Vieglais, E. O. Wiley, and T. Yodo. 2004. Predicting invasions of North American basses in Japan using native range data and a genetic algorithm. Transactions of the North American Fisheries Society 133:845–854.
Iverson, L. R., A. M. Prasad, D. Sydnor, J. M. Bossenbroek, and M. W. Schwartz. 2006. Modeling potential Emerald Ash Borer spread through GIS/cell-based/gravity models with data bolstered by web-based inputs. Pages 12–13 in V. Mastro, R. Reardon, and G. Para, editors. Emerald ash borer research and technology development meeting, Pittsburgh, PA. U.S. Department of Agriculture, Forest Service, Animal and Plant Health Inspection Service, Otis, MA.
Kammann, E. E., and M. P. Wand. 2003. Geoadditive models. Journal of the Royal Statistical Society Series C 52:1–18.
Kolar, C. 2004. Risk assessment and screening for potentially invasive fishes. New Zealand Journal of Marine and Freshwater Research 38:391–397.
Leung, B., D. M. Lodge, D. Finnoff, J. F. Shogren, M. A. Lewis, and G. Lamberti, G. 2002. An ounce of prevention or a pound of cure: bioeconomic risk analysis of invasive species. Proceedings of the Royal Society of London Series B 269:2407–2413.
Marsh-Matthews, E., and W. J. Matthews. 2000. Geographic, terrestrial and aquatic factors: which most influence the structure of stream fish assemblages in the midwestern United States? Ecology of Freshwater Fish 9:9–21.
McCullagh, P., and J. A. Nelder. 1989. Generalized linear models. Chapman and Hall/CRC, Boca Raton, FL.
McCune, B. 2006. Non-parametric habitat models with automatic interactions. Journal of Vegetation Science 17:819–830.
Morrison, L. W., S. D. Porter, E. Daniels, and M. D. Korzukhin. 2004. Potential global range expansion of the invasive fire ant, Solenopsis invicta. Biological Invasions 6:183–191.
Nekola, J. C. 1999. Paleorefugia and neorefugia: the influence of colonization history on community pattern and process. Ecology 80:2459–2473.
Peterson, A. T. 2003. Predicting the geography of species’ invasions via ecological niche modeling. Quarterly Review of Biology 78:419–433.
Peterson, A. T., M. Papes, and D. A. Kluza, D. A. 2003. Predicting the potential invasive distributions of four alien plant species in North America. Weed Science 51:863–868.
Peterson, A. T., J. Soberon, and V. Sanchez-Cordero. 1999. Conservatism of ecological niches in evolutionary time. Science 285:1265–1267.
Peterson, A. T., and D. A. Vieglais. 2001. Predicting species invasions using ecological niche modeling: new approaches from bioinformatics attack a pressing problem. BioScience 515:363–371.
Phillips, S. J., R. P. Anderson, and R. E. Schapire. 2006. Maximum entropy models of species geographic distribution. Ecological Modelling 190:231–259.
Raxworthy, C. J., E. Martinez-Meyer, N. Horning, R. A. Nussbaum, G. E. Schneider, M. A. Ortega-Huerta, and A. T. Peterson, A. T. 2003. Predicting distributions of known and unknown reptile species in Madagascar. Nature 426:837–841.
Ricciardi, A., R. J. Neves, and J. B. Rasmussen. 1998. Impending extinctions of North American freshwater mussels (Unionida) following the zebra mussel (Dreissena polymorpha) invasion. Journal of Animal Ecology 67:613–619.
Ruiz, G. M., L. Fegley, P. Fofonoff, Y. Cheng, and R. Lemaitre. 2006. First records of Eriocheir sinensis H. Milne Edwards, 1853 (Crustacea: brachyura: varunidae) for Chesapeake Bay and the mid-Atlantic coast of North America. Aquatic Invasions 1:137–142.
Seguardo, P., and M. G. Auraujo. 2004. An evaluation of methods for modelling species distributions. Journal of Biogeography 31:1555–1568.
Sharma, S., D. A. Jackson, C. K. Minns, and B. J. Shuter. 2007. Will northern fish populations be in hot water because of climate change? Global Change Biology 13:2052–2064.
Shogren, J. F. 2000. Risk reduction strategies against the “explosive invader.” Pages 56–69 in C. Perrings, M. H. Williamson, and S. Dalmazzone, editors. The economics of biological invasions. Edward Elgar, Cheltenham, UK.
Stockwell, D. 1999. The GARP modelling system: problems and solutions to automated spatial prediction. International Journal of Geographic Information Science 13:143–158.
Stockwell, D. R. B., and A. T. Peterson. 2002. Effects of sample size on accuracy of species distribution models. Ecological Modelling 148:1–13.
Sutherst, R. W., and G. F. A. Maywald. 1985. Computerized system for matching climates in ecology. Agriculture, Ecosystems and Environment 13:281–299.
Sydnor, T. D., M. Bumgardner, and A. Todd. 2007. The potential economic impacts of emerald ash borer (Agrilus planipennis) on Ohio, U.S., communities. Arboriculture and Urban Forestry 33:48–54.
Thuiller, W., L. Brotons, M. B. Araujo, and S. Lavorel. 2004. Effects of restricting environmental range of data to project current and future species distributions. Ecography 27:165–172.
Thuiller, W., D. M. Richardson, P. Pysek, G. F. Midgley, G. O. Hughes, and M. Rouget. 2005. Niche-based modelling as a tool for predicting the risk of alien plant invasions at a global scale. Global Change Biology 11:2234–2250.
Vander Zanden, M. J., J. D. Olden, J. H. Thorne, and N. E. Mandrak. 2004. Predicting occurrences and impacts of smallmouth bass introductions in north temperate lakes. Ecological Applications 14:132–148.
Whittier, T. R., P. L. Ringold, A. T. Herlihy, and S. M. Pierson. 2008. A calcium-based invasion risk assessment zebra and quagga mussels (Dreissena spp). Frontiers in Ecology and the Environment 6:180–184.
Wood, S. N. 2006. Generalized linear models: an introduction with R. Chapman and Hall/CRC, Boca Raton, FL.
Wood, S. N., and N. H. Augustin. 2002. GAMs with integrated model selection using penalized regression splines and applications to environmental monitoring. Ecological Modelling 157: 157–177.
Zaric, G. S., and M. L. Brandeau. 2002. Dynamic resource allocation for epidemic control in multiple populations. IMA Journal of Mathematics Applied in Medicine and Biology 19:235–255.
