Protocols and Models for Inventory, Monitoring, and Management of Threatened and Endangered Plants: Part 2; Models for Inventory, Monitoring, and Management of Threatened and Endangered Plant Species

Part II: Models for Inventory, Monitoring, and Management
of Threatened and Endangered Plant Species

(1)Colorado State University Department of Rangeland Ecosystem Science
Fort Collins, CO; (2)US Army Corps of Engineers, Waterways Experiment Station, 3909 Halls Ferry Rd., Vicksburg, MS 39180-6133

ON THIS PAGE
Section I: Executive Summary of Models
Section II: Introduction and Background
2.1: Introduction
2.2: Statutory background
2.3: Characterization and Evaluation of Habitat Relationships
2.4: Habitat and Plants
Section III: Habitat-Based Models
3.1: Uses of Models
3.2: Construction of Models
3.3: General Constraints to Modeling
Section IV: The Validity and Applicability of Habitat-Based Models
for Management
4.1: Sources of Error in Habitat-Based Modeling
4.2: Model Validation
4.3: Current Agency Procedures for TES Inventory, Monitoring, and Management
4.4: Recommendations to Agencies Addressing the Survey, Inventory, and Monitoring of Plant TES
4.4.1: Survey
4.4.2: Inventory
4.4.3: Population and habitat monitoring
4.4.4: Research
4.4.5:Summary
References Cited: Part I; Part II
NEXT PAGE
Section V: Descriptions of Models
Section VI: Summary, Conclusions, and Future Directions
Part II Table: Summary of Model Characteristics

Habitat-based modeling has been used extensively to indicate potential wildlife distributions and to assess potential effects of habitat alterations on wildlife species. This report examines possible applications of these and other kind of models to the inventory, monitoring, and management of threatened and endangered plant (TES) species. First, general considerations in habitat-based modeling are addressed, including potential sources of error, model validation issues, statutory requirements, and background for TES conservation. It is noted that models in general must be viewed as having limited value for extrapolation from one habitat to another; variation in natural systems limits the validity of models for such uses. In addition, the limitations of using habitats as an indicator of population status for plant TES are discussed.

The second part of the report provides a review of various kinds of models with applications for habitat characterization. Included are approaches that have been used for wildlife and plants in basic and applied research applications, and the limitations and strengths of each approach. The potential applicability of these models to plant TES is also addressed.

In a final section, two lines of reasoning are reviewed that suggest that models currently designed to assess wildlife habitats may provide generally better results for plant TES. Included are examples of several feasible applications of habitat-based models to TES plants: for locating a potential habitat for introductions of populations, and for guiding management in initial protection measures.

Habitat-based modeling has been seen most prominently in applications to wildlife, including threatened and endangered species (TES). Few examples exist of species-specific applications of habitat-based models to plants in general or to plant TES in particular. However, many studies (e.g., ordination studies and gradient analysis) have found correlations between vegetation types and a number of habitat and environmental variables. Designing models to predict the distribution of individual plant species based on their habitat requirements would therefore seem to be feasible. If accurate for prediction, such models would be useful in addressing the inventory and monitoring needs of federal agencies that must manage their lands for multiple uses, including the conservation of plant TES. This report is a review of the most widely used and yet, current modeling techniques available for habitat characterization and predictive uses with respect to survey, inventory, monitoring, and management of animal and plant TES.

One of the primary goals of modeling in species-habitat management is prediction of the responses of species to habitat alteration (Van Horne and Wiens 1991), because agencies must reconcile multiple land-use needs with species conservation. In such applications, models have often been expected to describe, simplify, predict, and extrapolate data about the habitat relationships of species, and to be reasonably accurate in these tasks. However, models are always limited in various ways: by their assumptions, by the kinds of data used in models, by the functions used in their construction, and by the inherent variation that exists in the systems to which models are applied. These limitations are attributes of models, not separate concepts, and must always be considered in applying model results to real-world problems. Limitations are partly by-products of the intrinsic simplification that takes place in the development of a model.

Despite these limitations to models, they have been used extensively for prediction since the 1980's in applications to wildlife species and habitat management by federal agencies. This is in part because resource managers tend to view models as useful even where predictivity is low or where models have not been tested. The reasoning is that some degree of guidance is better than none at all (Thomas 1986). This attitude is useful in general resource management where alternative, more detailed research approaches are not practical or timely, and where the large scale of resource issues demands an extrapolative and predictive approach.

The success of this approach requires, of course, that resource managers heed the numerous cautions about model limitations found in the literature (discussed in detail in our sections on validation and constraints to modeling). While unvalidated model results are in fact used (Block et al. 1994), they are to be used carefully and in conjunction with other information. Thus, a statement that habitat models are used in animal TES studies and management does not imply that they are ever the sole criteria on which management decisions are based. Habitat models should and do serve a supplemental function in guiding management of animal TES.

Certainly in applications to animal TES, numerous sources of information in addition to habitat models are provided to managers. For the endangered Florida panther (Felis concolor coryi), for example, in addition to habitat studies, research has been conducted on population status (McBride 1985), population viability analysis (Ballou et al. 1989), genetic issues (O'Brien et al. 1990, Seal 1991), and captive breeding and reintroduction (Eisenburg 1985).

Despite the fact that managers may often make use of untested models, in this report we emphasize that habitat models are scientific tools that should be subject to rigorous testing. All models, as the end products of numerous assumptions, are tentative hypotheses that must be tested repeatedly, in the locations to which they will be applied, if they are to be considered valid. Acceptance of the risks of using poorly predictive models may be justified in general land-use management applications, where margin for error is relatively high (e.g., where populations and mitigating habitat are relatively abundant), and where alternative approaches are lacking or not feasible. However, as Morrison et al. (1992), in reference to wildlife species, have argued, ". . . population trends of species of high concern, especially species on state and federal threatened and endangered lists, should be monitored directly in the field, rather than inferred through habitat relationships models". We make these qualifications to emphasize both the scientific constraints of modeling, and the potential risks associated with uncritical applications of habitat modeling to TES management.

The consideration of nongame wildlife species and plants in federal land management is, with few exceptions, a relatively recent phenomenon. Rising public concern in the 1960's over environmental quality and species declines culminated in a flurry of legislation relating to wildlife and plant species, their habitats, and the environment. Much of this legislation indicated a need for inventory, monitoring, and planning on federal lands. The National Environmental Policy Act (NEPA), of 1969 (42 U.S.C. 4321-4347), for example, requires environmental impact assessment (including impact on wildlife habitat) of federally-funded projects, creating a need for inventory of existing habitat resources. Other legislation mandates formal land-use planning of federal lands, for example the Forest and Rangelands Renewable Resources Act (FRRRA) of 1974 (16 U.S.C. 1601-1610), for National Forests; the Federal Land Policy and Management Act (FLPMA) of 1976 (Public Law 94-579), for public domain lands administered by the Bureau of Land Management (BLM); and the National Forest Management Act of 1976 (NFMA) (16 U.S.C. 1600-1614).

The Endangered Species Act of 1973 (ESA) (16 U.S.C. 1531-1544), amended most recently in 1988 (USFWS 1992), focused attention on species in danger of extinction. ESA provides various protections to species federally listed as threatened or endangered, stating as one of its primary goals, ". . . to provide a means whereby the ecosystems upon which endangered species and threatened species depend may be conserved" [Section 2(b)]. Listed species are classified as "endangered" (species judged to be in immediate danger of extinction throughout all or a major part of their range); or "threatened" (species likely to become endangered in the near future). Species that have recovered from risk may be delisted. The ESA also provides protection for animal species proposed for endangered or threatened status (candidate species), but that have not been formally listed. Candidate plant species, however, are not protected on federal lands under the ESA.

Among the ESA's protections is a prohibition against federal agencies engaging in any action ". . . likely to jeopardize the continued existence of any endangered species or threatened species or result in the destruction or adverse modification of habitat of such species" (ESA Section 7(a)(2)). Also required under Section 7 is biological assessment to determine whether proposed agency actions are likely to cause harm to listed species, animal species proposed for listing, or to critical habitat of threatened or endangered species. Consultation is required with the U.S. Fish and Wildlife Service (USFWS) or the National Marine Fisheries Service (NMFS). These services also design and implement regulations to enforce the provisions of ESA. However, agencies themselves are responsible for meeting the requirements of Section 7 (Clark 1994).

Statutes other than the ESA also contain provisions for protecting listed species, notably Conservation Programs on Military Land (1988) and the National Forest Management Act (NFMA) of 1976. In addition, federal agencies may implement internal policy of their own relating to species and habitat conservation (Clark 1994). For example, National Park Service (NPS) and BLM policies are to regard candidate plant species as if they were listed (Tamara Nauman, Dinosaur National Monument, NPS, personal communication).

As amended in 1978, Section 4 of the ESA requires that recovery plans be developed and implemented for listed species. USFWS (1990) defines recovery as ". . . the process by which the decline of an endangered or threatened species is arrested or reversed, and threats to its survival are neutralized, so that its long-term survival in nature can be ensured." Recovery plans, produced by or for USFWS or NMFS, describe the species, its life history, ecology, habitat relationships, reasons for listing, and prior conservation measures. Recovery plans also outline the objectives and criteria for recovery, and a proposed plan of action for recovery (USFWS 1990). While the ESA does not specify objective standards either for the determination of a species' listing status or for removal from the federal list (delisting) (Rolf 1991), USFWS (1990) requires recovery plan authors to state a realistic recovery objective (e.g., delisting, downlisting, ongoing protection), to estimate the number of viable, distinct populations of the species that would justify the objective, and to list the actions needed to improve the species' status. For each threatened or endangered species a recovery plan is required that ". . . delineates, justifies, and schedules the research and management actions necessary to support the recovery of a species" (ESA Section 4(f)). However, USFWS and the ESA have become notorious for having an extensive backlog on recovery plans for listed species (Clark 1994).

In addition to management of species and natural communities in accord with federal legislation, agencies must provide for other mandated land uses that may impact species habitat. For example, under the NFMA, USFS must manage its lands for timber production, watershed quality, recreational uses, and livestock grazing, while also providing for the needs of native species and their habitats. Department of Defense lands are used for military training and testing but managers must abide by federal regulations pertaining to species. Evaluation of the effects of habitat perturbation brought about by other uses must therefore be made frequently and on a relatively short time scale, and must encompass large areas.

For the purposes of this paper, habitat is defined simply as a place where an organism is able to live, either temporarily or permanently (Krebs 1994, Allaby 1994). Specific resources are provided by habitat, for example, food or nutrients, water, and shelter (Morrison et al. 1992). The habitat has a characteristic range of environmental conditions including a climate and moisture regime (Cooperrider et al. 1986). In addition, the habitat has a structure, composed of vegetation (MacArthur and MacArthur 1961, James 1971, James and Wamer 1982) and other physical components of the environment such as rocks or soil structure (Cooperrider et al. 1986). The concept of habitat may also include ecological interactions within and between species (Cody 1981).

Habitat characterization is the description of those elements or variables of the environment considered to be important to a species for its survival under natural conditions. Characterization of habitat includes simple narratives of habitat observations as found in species descriptions in field guides and floras, the use of statistical methods to establish habitat relationships to species, and quantitative predictive techniques used to predict the suitability of a given habitat for a particular species. Our use of the term "characterization" is therefore inclusive of habitat modeling. Any characterization is a model, in that it proceeds from observed or statistically-derived relationships between species and their habitats, and provides predictions about the preferences or requirements or distributions of species. For example, if the statement is made that a species of frog is found in riparian areas, we have both performed a characterization and constructed a simple model of the habitat relationships of the species, providing some indication of where to look for the species and what to expect of the habitat in which we find it. Conversely, however, a model is not necessarily a characterization because of the modeling constraints related to biological variability, model simplification, or lack of available data (see Section 3.3, General Constraints to Modeling).

We define habitat evaluation as a procedure that quantifies the quality of habitat. Habitat evaluation systems and habitat relationships models often assume that habitats have a range of quality or suitability for a given organism, and that this range can be quantified as a measure of quality (Morrison et al. 1992). At one end of this continuum, marginal habitats occur where food, water, and shelter are minimally adequate for survival of individuals in the short-term. At the other extreme are habitats that are optimal, providing the resources for long-term survival and reproduction of the species. Maintenance of viable populations depends on the presence of a sufficient number of habitats to provide for all life stages, including sufficient habitat in which the populations can reproduce. Defining the suitability of habitats, however, raises problems. For example, animal populations may irrupt periodically into unsuitable habitat (Van Horne 1983). Only site-specific monitoring can reveal whether populations of animals found in a habitat actually are viable.

For plants, habitat components of potential importance include nutrient and moisture availability, moisture regime, light, soil characteristics, presence of pollinators, seed source, and presence of herbivores and competitors (Bazzaz 1991). Research on plant habitat has focused on "environmental factors" (comparable to our use of the terms "habitat variable" or "habitat factor"), such as light or water, often with the goal of delineating factors limiting to a species' distribution (Liebig 1940). This is true for single-species approaches (e.g., in physiological ecology), as well as for studies of vegetation distribution and association, where environmental factors are correlated with plant community distributions (Greig-Smith 1982). Indeed the basic approach of wildlife biologists to habitat relationships is similar in concept to that of early plant biogeographers such as Warming (1909) and Schimper (1903), who described plant distributions in relation to environmental factors; and to the ordinations used in vegetation-environment studies of plant synecologists (Austin 1985, Bray and Curtis 1957, Whittaker 1967).

Predictions provided by modeling efforts are useful in several general ways: 1) As a means of testing the validity of the original assumptions of the model. Predictive modeling allows the formalization and testing of hypotheses, and has a long tradition in basic theoretical research in ecology, for example in optimal foraging theory (MacArthur and Pianka 1966). Such uses allow the study of systems and relationships too complex for experimental manipulation. 2) As an inferential component of the scientific process. As an inferential tool, statistical analysis for the identification of significant variables is a form of modeling, and is used extensively in many fields. 3) As an aid to decision-making. Management decisions affecting habitat often must encompass geographic areas too large for detailed study. Correlations determined to exist between species and habitat variables using small-scale studies, formalized as a model, can be potentially extrapolated to larger areas. This last approach may provide the greatest potential for use of habitat evaluation models in plant TES conservation, including inventory, monitoring, and management.

Modeling approaches have been developed to address basic problems that confront agencies charged with habitat and species management on federal lands. Given the variety and extent of federal lands and the large number of species on them, individual study of all species dependent on federal lands is not economically feasible. Habitat-based approaches may, for example, attempt to relate habitat quality to population status for use as an indirect indictor of the status of populations. In addition, the apparent potential for models also to extrapolate data across geographic space is a basic appeal of the modeling approach for use in management applications. While intensive site-, species-, and population-specific studies are always more accurate than the inferences of predictive models, collection of data at these levels is often and justifiably regarded as impractical for general resource inventory.

In descriptive and assessment applications, models can help to interpret large and complex data sets, as is seen in traditional applications of statistical models for vegetation research (for example, ordination or multiple regression), and in studies of habitat relationships in animals. Models can also help to synthesize habitat and species information from disparate data sources, providing a simplified summary of available habitat information that is accessible to managers (Laymon et al. 1985).

A model is a representation of reality (Allaby 1994). Any model is based on some perception of correlation, whether the correlation is established statistically or not (i.e., models are often based on qualitative information). The value of a model will depend in part on the selection of variables that will yield relevant information and that are suited to the habitat and species in question; that is, on the quality and validity of the data. Dueser and Shugart (1978) and Whitmore (1981) have outlined general criteria for the selection of variables to be measured in a study of species-habitat relationships. They emphasize that variables chosen should be easily and precisely measurable and should have biological meaning and relevance for the species in question.

In construction of a formal habitat-based model, habitat variables are measured in the field, related to species (in some cases the relationship is assumed), and used to construct a predictive function that relates the species in question to habitat attributes. For example, the relationship between habitat variables and the abundance of a species may be assumed to be linear, as is often the case for habitat suitability index (HSI) models (see Section 5.1.1), or an estimate of the nature of the relationship may be explored using regression fitting techniques to establish a function. The relationship may be viewed a probability function, as in pattern recognition (PATREC) models. More complex methods are often used, as discussed in our sections on individual model types.

The precision of a model is its capability "to replicate particular system parameters"; generality is its potential to "represent a broad range of similar systems" (Marcot et al. 1983). Sampling theory shows us that we can estimate the error of a sampling procedure statistically, given that we have sampled adequately. A model developed from data collected at a particular site (using an appropriate sampling regime) is likely to be more precise for that site than for another.

However, a model that attempts to extrapolate information from one location to other locations requires generality, in that its predictions must be accurate for the range of variation over which it is extrapolated. Because models are based on the variables and relationships measured at a particular site or group of sites (or even on unsampled, assumed relationships), which may not be representative of the full range of possible variation in habitat use, their application to other locations is questionable.

Precision must therefore be sacrificed to achieve generality, but models used to encompass too much variation in order to achieve generality become uninformative. Models intended for extrapolation of information from the location at which the data were collected must therefore be viewed as constrained by the difficulty of achieving simultaneous generality and precision. Krebs (1980) and Peters (1991) have discussed this problem for theories in general.

Holling (1984) provided a classification of problemsthat can be addressed by modeling approaches based on the level of understanding and data that are available (Figure 1). We usually have some understanding of the problem in habitat relationships modeling, in that we reasonably expect a relationship to exist between attributes of habitat and the distribution or abundance of species. However, even high-quality data taken at one site, will necessarily only be speculative if extrapolated to another site. Habitat-based modeling therefore operates in region 2 or 4 of Figure 1. Analytical modeling used in basic research may increase understanding by being limited to analysis of the processes in a site, working primarily in region 1, for example, using statistical analysis. Holling's point is that modeling may be applied to many levels of information, but that limitations must be recognized as to the accuracy of model predictions, depending on the level of information on which the model is based.

Accurate prediction is often considered to be beyond the capability of models of ecological relationships, because a model is by necessity a gross simplification of a complex system (Krebs 1994). Unlike the variables of engineering or chemistry (for which sufficient understanding and reliable data exist, i.e., region 3 of Figure 1), those variables used in the study of organism-environment relationships are not based entirely on invariable physical laws. Like all ecological relationships they are complex, and linked by behavioral, genetic, and ecological factors at multiple scales (Ricklefs 1991). As Allen et al. (1984) pointed out in reference to the large-scale simulation modeling efforts of the International Biological Program, which ended in 1975:

Section IV: The Validity and Appplicability of Habitat-Based Models for Management

Uncertainty in modeling results arises primarily from three sources: 1) from problems in data collection and extrapolation, 2) from the inherent variability of natural systems, and 3) from the assumptions and construction of the model itself. Morrison et al. (1992) point out that the implications of the level of uncertainty of a model is different for models used for purely analytical purposes than for models used in guidance for irreversible management decisions.

First, error can arise when extrapolations of models are expected to provide management guidance by predicting the effects of habitat perturbation on species population dynamics, via various levels of assumptions and data manipulation. In order to identify variables useful in predicting species abundance, a researcher must determine the suitability of the habitat in which the variables are measured. Often, high density of a species in a habitat is assumed to indicate quality habitat. This assumption is basic to pattern recognition (PATREC) methods and has often been used in other model types (e.g., in habitat suitability index (HSI) and habitat capability (HC) models).

Second, variation in biological systems can be a source of error. Habitat relationships modeling assumes a degree of correlation between species and their habitats (Cody 1981) because the physical and behavioral traits of species have developed, through natural selection over evolutionary time, in response to environmental conditions. The distributions of species are thus viewed as constrained by the range of habitat conditions that they are able to use effectively. However, with respect to habitat relationships, we can expect variation in habitat requirements among species, in the degree of habitat specialization among species, and among ecotypes or geographic variants within species; we also expect genetic variation among individuals and populations in phenotype, i.e., in morphology, behavior and physiology. For these reasons we can expect geographic variation in the habitats used or with potential for use by species.

Thus, even where optimal habitat conditions are identified with some degree of confidence, both habitat and populations, and therefore their degree of correlation, vary spatially and temporally. Rotenberry (1986) and Wiens (1977) have argued, for migratory birds, that population densities may thus become "uncoupled" from habitat variables. Such an effect obviously limits the generality of models.

Lastly, the assumptions and construction of the model itself may prove to be a source of error. If a key environmental variable that has a strong effect, for example, on population demographics, is not included, then the model will necessarily yield invalid predictions or conclusions. Or, the model may not be designed to the correct scale for the problem at hand. For example, a broad-scale succession model (see Section 5.1.4) would probably not be sensitive enough to detect short-term or small-scale changes in plant TES habitat and population dynamics.

Any model should be subject to rigorous testing if it is to be used for predicting the consequences of habitat alteration on the distribution or abundance of species. Van Horne and Wiens (1991) describe two steps in testing of model predictions: verification, which tests the internal consistency of the model, usually by comparing model predictions to the data set used to derive the model; and validation, which tests a model against a new set of data. The second step thus tests the extrapolative value of a model, or its generality. The issue of the predictivity of models used in basic research, because of their tentative, hypothetical nature, is a theoretical matter: while predictivity may affect the conclusions of a research study, results are not usually applied. For this reason, and because of the complexity of the systems modeled, validation per se is not normally a part of basic research. Generality is assessed by comparison with existing data or as new data becomes available. However, in management applications, predictivity and accuracy are practical concerns.

Seldom, if ever, could examination of a model or testing of its internal consistency provide a useful indication of its value for extrapolation, unless the model's data encompassed the full range of variation that exists in the field in locations to which it is to be applied. That is to say that while a model may have been developed from an adequate sample of one or several sites, it probably will not encompass all variation. Thus validation of a model is distinct from estimates of sampling error or confidence intervals in statistical sampling procedures, which apply to the range of variation in the sampled population, usually an individual biological population or group of populations and associated habitat. A model becomes extrapolative if it is applied to areas other than those sampled; i.e., to the statistical population composed of all locations in which a species actually could occur in the area to which the model is to be applied. Thus validation tests the application of a model against the real-world variation that may limit its value.

There is no standard method for validation. However, some estimate of the accuracy of predictions is generally provided in validation studies. Validation results are often reported as an analysis of the amount of error produced by the predictions of the model (e.g., Flather and King 1992, Block et al. 1994). Marcot et al. (1983) listed 23 criteria to evaluate validity of wildlife-habitat models. These include such attributes as realism, referring to the relevance of selected variables and relationships, and generality, as well as accuracy of prediction. Verification and validation can refer to checking of all issues affecting a model and its predictions, including data collection, model structure, and the geographic range of its accuracy (Van Horne and Wiens 1991).

Morrison et al. (1992) have made a useful distinction in this regard between two classes of predictions, hindcasting and forecasting. Hindcasting often uses correlation, regression, or multivariate statistics to elucidate observed patterns and strictly applies only to the time and place where the original data were gathered. Forecasting predicts species conditions at a time and place different from the area where field data was originally collected. Morrison et al. (1992) caution that often workers erroneously use the results of hindcasting to make predictions about species conditions. Without model validation and proper design, predictions from techniques suited for hindcasting can be incorrect.

Habitat-based models are often low in predictivity, by which we mean that they have high rates of error in their predictions. Laymon and Barrett (1986) evaluated a model for the northern spotted owl by Laymon et al. (1985) and unpublished models for marten and Douglas' squirrel. They found these models to be poorly predictive even though they were based on apparently reliable information. Flather and King (1992) found their regional model for three species to have high error rates, as did Block et al. (1994) for a wildlife-habitat relationships model. Rotenberry (1986) reported poor prediction results for a model of habitat relationships of shrub-steppe birds.

The step of validation is recognized as a crucial component of habitat modeling for wildlife management (Morrison et al. 1992, Cooperrider 1986, Flather and King 1992, Van Horne and Wiens 1991, Block et al. 1994), though it has often been skipped prior to management use. Indeed cessation of the development of the habitat suitability index (HSI) series of habitat models published by USFWS has been attributed to the lack of field validation of existing models (see our section on HSI models).

It is often implied in the literature, particularly for HSI-type models, that field testing will identify any problems with a model (Cooperrider 1986, Morrison et al. 1992), which can then be "fixed." This assumption may be tenuous at best. For instance, Starfield and Bleloch (1986), Block et al. (1994), and Stormer and Johnson (1986) describe the testing of models as an iterative process of successive tests leading to new models, which in turn are tested. The process of field testing, then ". . . is very different from the clear-cut recipe of first building a model, then validating it, and finally using it" (Starfield and Bleloch 1986). Stormer and Johnson (1986) and Block et al. (1994) add that the testing process is complete only when the model performs at acceptable levels of accuracy. Stormer and Johnson (1986) also stress that "like theories . . . models are always tentative and never absolutely proven."

Manipulative experiments have been suggested as the most accurate means of validating the habitat variable assumptions of models, and population monitoring as the best method of field validation of model predictions (Van Horne and Wiens 1991). In model validation, field testing often refers to some level of site-specific population monitoring (e.g., Synge 1981, Palmer 1987, Cropper 1993, Given 1994).

It can be argued that if predictive modeling needs to be validated on a site-specific basis prior to each management use at a new location, then much of the utility of modeling is lost. In many cases (i.e. for TES with very limited ranges or few populations), an apparent need for a model may be negated by the time necessary to validate its predictions. Site-specific monitoring could ultimately prove more efficient in such cases.

4.3 Current Agency Procedures for TES Inventory, Monitoring, and Management

Currently federal agencies are required under ESA to inventory and monitor plant TES. Description of habitat is also required under USFWS guidelines for recovery plans (USFWS 1990). However, USFWS does not specify methodology to be used in the inventory and monitoring of plant TES or their habitat. Further, no federal agency, to our knowledge, has developed guidelines for methodology specific to the survey, inventory, and monitoring of plant TES species and habitat (based on literature searches and telephone contacts with agency personnel at BLM, NPS, USFS, USFWS, NBS in Colorado, Utah, Montana, and Washington, DC).

Methodological guidelines have not been developed because plant TES are considered as unique problems and are thus each treated on a species-specific basis by USFWS and NBS (Tom Muir, NBS Inventory and Monitoring Office, Washington, DC, personal communication). Individual management units of agencies may develop their own guidelines, though we are aware of none in existence at present. For example, Dinosaur National Monument (NPS, Colorado/Utah) is developing a protocol for monitoring of threatened and endangered plant species (Tamara Nauman, NPS, personal communication). The U.S. Army has developed a statement of strategy and philosophy for treatment of TES species on Army lands (Tazik and Martin 1994) that does not include methodological guidance.

The Nature Conservancy (TNC) and affiliated state Natural Heritage Programs are organizations that have been contracted by federal agencies to perform assessment of plant TES. TNC, a private organization that works in biodiversity conservation, is often consulted or contracted by federal and state agencies for the inventory stage of habitat assessment where plant TES are involved. Their methods integrate species and habitat using an "elements of diversity" approach. TNC methods are pragmatic and designed for guidance in acquisition of important habitat areas: the goal is to locate areas that merit preservation or special consideration. Elements include any scale of consideration for locating and inventorying diversity, for example vegetation types, plant community types, species, geographic features, soil types, and other specific habitat attributes. Elements are described as part of an inventory process designed to locate elements at narrowing scales of resolution, using a coarse scale ("coarse filter" in TNC jargon) of dominant vegetation and plant community-type, delineated using vegetation classifications and maps; and a fine scale ("fine filter") of on-site searches to verify presence of elements of concern (Jenkins 1991, Noss 1987). TNC methodology provides a framework for general habitat assessment by narrowing geographic scale from coarse to fine in order to capture information on many elements of diversity, including rare species.

TNC methods are a good example of the relationship of empirical field work to coarse-scale indicator or modeling approaches. Coarse scales are used to narrow a search area, and a field study verifies indications that were based on coarse habitat information. Essentially, the coarse filter is a model, and the fine filter is a verification step.

After inventory of the species of concern has taken place, monitoring and management of the species should become a part of the recovery process. In each case, agencies are required by ESA law to manage populations of listed species for recovery and they may initiate relevant research. Even where a formal recovery plan is not in effect, agency research efforts should further the goal of recovery, or at least not hinder it (Lucy Jordan, Botanist, USFWS Region 6, personal communication).

4.4 Recommendations to Agencies Addressing the Survey, Inventory, and Monitoring of Plant TES

Applications of habitat-based modeling to plant TES management is really unknown because of the few existing examples of management-oriented, habitat-based modeling for individual plant species, and the lack of tests of accuracy of basic research models of vegetation. Development and testing of such models and tests of applications may represent a fruitful area of potential research. This section presents some potential applications to plant TES in the contexts of survey, inventory, and monitoring.

We distinguish five areas in which agencies can address plant TES. These are: survey, inventory, monitoring, research, and management. This report will address the first four of these. These levels are interrelated, in that appropriate management should be guided by the findings of the other four levels and vice versa. For example, some level of descriptive habitat characterization would be necessary to delineate research areas in field studies. Inventory and monitoring also entail research, however, we treat them separately from our research section because they are legally required for TES management. Listed below is a brief summary of the four areas (Bonham and Bousquin 1996):

A preliminary survey determines which TES species may be present on an installation. This includes consultation with local experts, searches of literature and information bases, and reviews of relevant past work. A field survey will locate and verify that TES occur (or do not occur), and the geographic location of these populations.

Inventory is a "one-point-in-time" assessment which may or may not be considered as separate from data collection. This may include listings of TES present on the installation, listings of geographic information for populations, counts of individuals within populations, and more detailed data collection. The purpose of the inventory is to describe the important characteristics of species, populations, and their habitats.

Monitoring is the quantitative assessment of population or habitat conditions over time to detect change, direction of the change, and its magnitude. To be useful for monitoring, models must have the ability to detect change in populations or habitat.

Research refers to detailed study of, for example, ecological relationships, genetics, or metapopulation dynamics, which may enhance knowledge and management of a TES.

Other authors do not make these same distinctions (e.g., Ayensu 1981, Palmer 1987, Cropper 1993, Given 1994); however; because there is little consensus, our list is intended to express the important levels of TES information.

A preliminary survey determines which TES may be present on an installation. The first step in this process is the collection of information about potential TES occurrence. In some cases, previous research may provide part or all of the information necessary to locate potential TES.

The next step in the preliminary survey will involve gathering information for potentially occurring TES from the literature and experts. This existing information can then be used to identify special requirements, biotic-abiotic associations, sensitivities of TES, and to refine installation habitat information. Collection of existing habitat and land-use information for the installation will then aid in the process of delineating possible habitat areas for plant TES. Various predictive methods (e.g. statistical models, HSI models, PATREC models) can be used to facilitate the location of TES populations: predictive or extrapolative models can be used to help identify possible locations of extant populations of plant TES, either for verification by field searches, or, where field searches are not practicable, for tentative identification and protection of potential habitat. Even where populations do not exist, potential habitat may be regarded as valuable as a possible site for reintroduction of populations. For example, a predictive model may utilize a particular vegetation type to characterize the expected habitat of a plant TES.

Predictive efforts should be followed by field surveys to verify predictions. The field survey will serve to verify and determine the geographic locations of any TES present. The field survey is guided by habitat and species information.

The TES inventory may include lists of TES present on the installation, listings of and geographic information on TES populations, counts of individuals within populations, and more extensive quantitative data collection on individuals, populations, and habitats as a basis for habitat characterization, modeling, and monitoring. This information also can be used to further refine models selected for use to evaluate survey results. Data is collected on relevant habitat (i.e. environmental and vegetation associations) for all or a representative sample of documented populations of each TES.

Monitoring is usually considered to be a site-specific activity that quantitatively assesses the condition of a set of variables over time to detect change, to determine the direction of that change, and to measure its magnitude (Bonham 1989, Given 1994). Habitat-relationships models can have two functions for habitat monitoring: 1) using model predictions to indicate the amount and quality of habitat available as discussed above, and 2) using model predictions to "monitor" or predict change in extant populations.

Habitat-based models may have some applications for the monitoring of non-TES populations (Morrison et al. 1992) by providing an "early warning" of population changes that may result from habitat change. Habitat-based models assume that, because habitat quality affects population demographics, monitoring of habitat quality can be used to indicate present or future population change. This assumption leads to several complications within the model, however. First, this approach presumes that habitat itself is monitored in sufficient detail to detect habitat change. Next, habitat quality is not the only factor that affects population density. Therefore monitoring of populations via predictive habitat-based models proceeds through an additional level of extrapolation from model assumptions. Lastly, the predictions of habitat-based models are likely to become less accurate as time proceeds. Noon (1986) stated that "for successful, long-term management of most wildlife species, information on biological processes (i.e., birth and death rates) will be essential." As quoted in our introduction, Morrison et al. (1992), among others, maintain that model-derived inferences about the population trends of sensitive wildlife species should be avoided, and that population trends should be monitored directly.

Predictive uses of habitat-based models do have appeal because they avoid data-intensive demographic monitoring (described below) (Van Horne 1983, Morrison et al. 1992). In general, such uses are most prominent in broad-scale assessment models, such as models developed under Habitat Evaluation Procedures (USFWS 1980b, Boyce 1980, Benson and Laudenslayer 1983). Such broad-scale models are probably not sensitive enough for use in monitoring plant TES, however.
In rare plant conservation, the term monitoring refers to some level of site-specific population monitoring (e.g., Synge 1981, Palmer 1987, Cropper 1993, Given 1994). While population monitoring is often viewed as expensive and data-intensive (Morrison et al. 1992), useful and direct information on the status of populations can be obtained from various intensities of data collection with a corresponding range of effort. Population-level monitoring is well-documented and accepted in the literature of rare plant conservation (e.g., Menges 1986, Palmer 1987, Schemske et al. 1994). Van Horne and Wiens (1991) consider population monitoring to be a useful means of testing habitat-relationships models. See our discussion of adaptive management for examples of applications of monitoring to TES management.

Palmer (1987) distinguished three levels of intensity for population monitoring of plant TES species. These are:

1). Simple census of individuals in the population over time, which can provide a rough estimate of population stability from changes in population size.
2). Census of vegetative and reproductive individuals using a repeatable sampling design (e.g., transects or quadrats, see Bonham 1989). This method can project population trend but cannot identify life-stages crucial to the growth rate of the population (Schemske et al. 1994).
1). Demographic monitoring of the vital rates (birth and death rates) of populations. This level allows use of matrix simulation models to project population trajectory and to identify life stages that most affect population growth rate (Leslie 1945, Lefkovitch 1965).

The design of a monitoring program should follow methodology compatible with both the inventory and survey stages of TES habitat characterization. This may include considerations of QA/QC criteria, which should be considered in monitoring efforts to track population dynamics and related habitat changes. Monitoring results should be analyzed periodically to evaluate military and other effects on TES populations. Further, monitoring should be continued at some level as a feedback mechanism for timely adaptive management response to changes in population status and other trends.

Monitoring or specific questions about a plant TES may lead to research on habitat requirements or associations that can be addressed through various forms of habitat characterization or habitat-based research. Long-term monitoring and additional research may be planned on a revised schedule based on emerging information. For example, research may be designed to determine the causes of declines in extant TES populations or to evaluate the possibility of their relocation to another area. Habitat-oriented research to which habitat-based models may apply is discussed in our sections describing particular model types.

By definition, TES are in imminent danger of extinction, often as a result of habitat loss or modification. Further, TES often have specific habitat requirements, and for these reasons may be particularly sensitive to habitat perturbations. The provisions of ESA and regulations developed by USFWS prohibits alteration or destruction of TES habitat (with few exceptions), and requires federal agencies to actively conserve TES. If these agencies are to use habitat characterization models in their attempts to conserve TES, then the sources of error, constraints, and limitations of habitat characterization models should be well-understood and accommodated for. For these reasons, validation and field checking of all model predictions is especially important for TES applications.

Continue to Sections V & VI of Part II
References Cited in Part II

Part I: Protocol for Inventory and Monitoring
of Threatened and Endangered Plant Species
Sections I-III
Sections IV & V

Glossary of Acronyms

Document placed on the Web Monday, April 2, 2001
This page has been accessed times.