An Adaptive Approach to Habitat Management for Migratory Birds in the Southeastern United States

Michael J. Conroy 1

ABSTRACT— Modern tools for habitat management of migratory birds include models describing habitat-population relationships, coupled with remote sensing and geographic information systems (GIS). These approaches implicitly assume some degree of underlying understanding about the functioning of bird populations and communities in response to habitat modifications. First I discuss some general principles in modeling, with emphasis on the use of models as tools for generating testable predictions from our provisional understanding. I then describe some approaches for modeling bird-forest habitat relationships, with emphasis on recent mechanistic models based on individual bird behavior. I discuss a specific application of modeling in the management of habitats for Wood Thrush (Hylocichla mustelina) populations in Georgia, and how a conceptual model of habitat-population dynamics led to a management experiment designed to test underlying hypotheses of the model. I then discuss some difficulties in parameterization of spatially explicitly models, and some recent work on statistical models for providing habitat-specific estimates of survival and movement rates. Finally, I argue that scientific management only happens when management is treated as "experimentation" and models and other predictions about management impacts as testable hypotheses. Research and monitoring must be integrated with management to assure that the data gathered are relevant to decision making and used to inform decision makers. An adaptive management (Walters 1986) paradigm provides a unifying framework for accomplishing these goals.


Biologists and managers responsible for conservation of migratory birds are faced with a complex array of often confusing information about the potential impacts of forest management on these species. Here "management" includes both manipulation of habitats and populations, such as forest cutting, and the "preservation" of habitats through reserves or other means. A plethora of GIS and other analytical tools, including models, that provide forecasts of bird abundance and species composition under various alternative scenarios of management is now available. Some forest planning tools, traditionally limited to predictions about timber and other commodity production, have or will be modified to include the conservation of biodiversity, often birds, as a component of the objective function (e.g., Smith et al. 1981). Other tools such as Gap Analysis (Scott et al. 1993) are intended to provide managers with a comprehensive landscape-level approach, with biodiversity conservation as the principal goal. Like other planning tools, these rely on the use of models to make projections about the status of populations, communities, and ecosystems under alternative future conditions, including management.

These uses of models have several common features and potential difficulties. First, unlike many applications of models in the physical sciences, the true nature of the processes that ecologists seek to model is typically both highly variable and incompletely understood. Therefore, models cannot reasonably be expected either to exactly describe a particular system (e.g., a community of forest birds) at a given point in time, or to predict very well its state or condition into the future. Further, because of inherent stochasticity (e.g., because of randomly changing weather conditions), difficulties in quantifying the system (e.g., estimating abundance, demographic parameters, movement rates), and uncertainty about the nature of functional relationships (e.g., is the population resource limited? Is it limited by dispersal or the availability of habitats?) it is even less likely that forecasts about the future will be accurate. In some instances, the state of knowledge of the organisms and their habitats may be so poor that any model constitutes at best a crude guess. For the above reasons, and many more, conservation biologists frequently view models and modeling with distrust and disdain, and question the value of modeling as a part of management.

Much of this skepticism is warranted. In my view, models in ecology, perhaps especially in wildlife management, frequently have been used in an inappropriate and uncritical manner. However, the problem lies not with models or modeling per se, but with the failure to use models as a component in a scientific process leading both to more rationale (i.e., better informed) decisions, and to improved understanding. The moment that a modeler/biologist behaves as if his or her model is the truth, and then uses the model to dictate what a management policy should be, he or she has exited the realm of scientific natural resource management and embarked on a pilgrimage of faith.

On the other hand, judicious use of models as part of a scientific process may prove fruitful, as long as models are recognized simply as the mathematical expressions of our provisional understanding of how a system might work. Models should be treated as hypotheses, and their output as predictions of how the world might look under some management scenario, instead of prescriptions determining how it will look (Conroy 1993a). And in fact, most managers have their own conceptual "models" of how forests, populations, and ecosystems function, and therefore how to manage them, although few will have formally described or written down a model (let alone attached equations to one), and most would not see themselves as "modelers."

In this paper I show how such a process might work, first by describing modeling in a bit more detail, then by describing some types of models which have potential use for managing migratory birds. I then will describe a conceptual model that colleagues and I have used to orient field research directed at the impacts of forest management practices on a breeding population of a Neotropical migratory bird (NTMB), and show how our experiments should provide a refined model that may improve decision making. Next, I will discuss the critical issue of the relationship of habitat to demographic processes, and describe some work on methods for empirically estimating these relationships. Finally, I will describe some very recent work, in which model and other sources of uncertainty are explicitly accounted for in conservation decision-making. Dynamic animal and forest growth models are imbedded in an adaptive optimization procedure to provide for optimal decision making to achieve long-term goals, and to orient research and monitoring programs to meet these objectives.


Critical components to model development

In considering the development of a model relating forest habitat to bird populations, clearly stating the objectives of the model is important. I assume that conservation biologists will principally be interested in models that relate in some manner to management, either directed toward conservation of bird populations, or perhaps more commonly, toward some other primary goal, with conservation of birds either a secondary objective or a constraint.

The value of modeling to the manager will be determined by whether it helps the manager better understand or predict the consequences of management. However, because models are abstractions of reality, these predictions will be imperfect; in some instances the model will provide little more than an educated guess about what might happen to populations under various management scenarios. Thus, it is critical that modeling not be viewed as a recipe for what will happen (and thus, what should be done), but rather as a mathematical means of expressing what might happen, if the model is true (Conroy 1993a). Because a single model is unlikely to be completely true, it is crucial for managers to consider alternative models, including some that may be diametrically opposed to the model the biologist believes most likely. The consideration of alternative models is both prudent in terms of hedging management options, and consistent with the scientific method. Closely related, for a model to have scientific credibility, both it and its competitors must produce testable predictions, i.e., must be capable of generating statements that can be supported or refuted with observations. When models become ensconced as articles of faith (or matters of governmental policy) and are not subject to testing, they no longer are a part of science.

Also crucial is the issue of observability. For the most part, useful models must produce predictions about observable phenomena, and must be constituted of parameters that have biological and physical meaning. If predictions cannot be compared to observations, then there is little prospect for verification or falsification of the model, or of discrimination between competing models. For example, habitat suitability models sometimes are claimed to predict the "potential occurrence" of animals, a phenomenon that does not appear to be observable (Conroy 1993a).

The realism of a model is the degree to which it contains the essential elements of the system of interest. For birds in forest systems, realism suggests that the model include key components of environmental and biotic factors (e.g., habitat) and relate in a meaningful way to bird abundance and distribution through time, ideally in terms of demographic parameters (survival and reproduction rates) and behavior (e.g., movements and dispersal). If model parameters have no relation to biology, then the model has little realism (Levins 1966) and is unlikely to provide predictions beyond a narrow range of current conditions, even if empirically validated (Conroy 1993a, Conroy et al. 1995). Realism may but does not necessarily require individually based and/or spatially explicit models (e.g., Dunning et al. 1992, 1995), recognizing that these approaches carry a heavy burden in terms of model parameterization, validation, and updating (Conroy et al. 1995). The point to remember is that the "realism" of a model depends on the purpose for which it is being constructed: i.e., the research questions that are being asked and the management decisions that will be made using the model. A model that is sufficiently detailed ("realistic") for one purpose may be totally inadequate for a second, and unnecessarily complicated for a third (see also Levins 1966).

As alluded to earlier, an additional, desirable feature of a model is its generality, i.e., its ability to describe or predict over a broad range of conditions. Usually (but not always), models based on biological mechanisms, which have been validated under broad conditions, will have greater generality than models that are strictly empirical or validated under narrow conditions. However, as with realism, the generality needed depends on the application; if the application is narrow, the model need not be general, so long as future applications remain narrow, and conditions remain similar to those under which the model was constructed and validated.


I will illustrate some of the above principles by means of two types of models used in wildlife management, including management of NTMBs. In the first of these, animal abundance or (more typically) presence or absence, is predicted on the basis of vegetation type, structure, configuration, or other attributes (see Van Horne and Wiens 1991 for a critical review of these models for birds). This type of model includes Habitat Suitability Index (HSI) and Wildlife Habitat Relational (WHR) models developed by biologists in, respectively, the U. S. Fish and Wildlife Service (USFWS) and U. S. Forest Service (USFS). These models are typically based on literature surveys or "expert opinion;" they often have limited field validation; and they usually do not have parameters with biological interpretation. A further difficulty, alluded to earlier, is that these models usually predict habitat 'potential' rather than occupancy or other observable phenomenon, making validation problematic (Conroy 1993a). Despite these limitations, HSI and WHR models may provide some qualitative statements about habitat suitability, and may be all that are available for many species.

A second type of model is based on mechanisms operating at the individual or local population level, and includes single population, patch dynamic (Levins 1969) including source-sink (Pulliam 1988), and spatially explicit (SEPMs; Dunning et al. 1992, 1995; McKelvey et al. 1993) models. For example, the BACHMAP model (Dunning et al. 1992) used a spatially explicit approach in which the movement and fate (survival, reproductive status) of individual Bachman's sparrow (Aimophila aestivalis) in a dynamic (i.e., undergoing succession and management) landscape is simulated. By summarizing birds' fates across the population, it is possible to assess the effects of various landscape configurations and compositions on population viability. Similarly, McKelvey et al. (1993) used a spatially explicit, individual-based model (OWL) to simulate impacts of forest management on northern spotted owl (Strix occidentalis) populations.

Models such as BACHMAP and OWL tend to have a strong theoretical base and parameters that have biological interpretation. Further, they generally predict observable phenomena such as animal abundance and spatial distribution, and thus are apparently more subject to empirical validation (or refutation). However, this complexity can make these models difficult to parameterize, for example, values for survival, reproduction, movement, and other parameters. Also, the resolution of observations is frequently insufficient to truly validate such models; for example predictions from spatially explicit models that are aggregated to correspond to broader-scale field observations may not be useful in validating the underlying model (Conroy et al. 1995). Nonetheless, in my opinion, the mechanistic nature of these models makes them more likely to perform over a range of conditions, and more amenable to validation and testing against alternatives, than purely empirical or ad hoc models.


In my research group, we currently are taking several simultaneous approaches to the problem of model development and application in the management of NTMBs. I will describe these briefly, and try to show how they fit together into a comprehensive and adaptive framework.

Effects of forest management on Wood Thrush populations

Populations of Wood Thrush have declined in recent decades, as have a number of other forest nesting NTMBs (Sauer and Droege 1992). Forest fragmentation on the breeding grounds may be a causal factor, but in significant portions of the range, the population is declining during concurrent periods of increase in the total acreage and average stand size of forests (Turner and Ruscher 1988, Odum and Turner 1990), suggesting that if an impact occurs through forest management on the breeding grounds, it is more complex than simple area or compositional effects. One possibility is that increased patchiness (internal fragmentation) of managed stands may have resulted in decreased reproductive success and survival and changes in dispersal rates, potentially resulting in landscape-level population impacts (Powell et al. 1995). However, we view these potential impacts as predictions emanating from our underlying view of what makes Wood Thrush populations work, and freely acknowledge that we may be wrong. We have used our hypotheses about functional relationships between habitat features and Wood Thrush populations, together with general notions of population dynamics, to construct a conceptual model that is spatially explicit and contains alternative hypotheses as specific sub-models within an over-arching model structure .

The basic components of our model (Fig. 1) involve 1) compartment level representation of breeding season survival (adult and juvenile) and reproduction rates; 2) movement rates between-compartment (intra- and inter- year), 3) proportional (temporary) effects of treatments on demographic and movement parameters; and 4) between-year survival and fidelity rates (possible impacts on the latter by treatments).

Figure 1. Conceptual model for Wood Thrush breeding population, Piedmont National Wildlife Refuge, Georgia. A population of size Nt arrives (comprised of individuals breeding or produced on this landscape last year plus "immigrants" from other local breeding populations). Of Nt, E individuals select other landscapes, leaving a local breeding population of Nt - E. Shaded ellipses represent local habitat conditions (stand size, structure, composition) influencing mortality losses (M) and reproductive gains (B). Landscape-level events (e.g., movement among stands to search for food or new nest sites, represented by double-headed arrows) also influences M. Summation symbol represents integration of landscape and local effects on net number of individuals entering fall migratory population. See Table 1 for hypothesized local and landscape-level effects of management.

conroyPif001.GIF (1266 bytes)

We recently have completed a field experiment that will help us to parameterize our conceptual model, and give us at least a preliminary understanding of the potential effects of landscape changes through management on Wood Thrush population dynamics (Powell et al. 1995, 1997; Conroy and Krementz 1997). Our study site is located on the Piedmont National Wildlife Refuge and adjoining Oconee National Forest in Georgia (Fig. 2). Management on these federal lands is directed toward a variety of purposes, including production of timber and sustainable harvest of wildlife populations. An important additional component of management is the maintenance of viable populations of Red-cockaded Woodpeckers (RCW, Picoides borealis). RCW management includes the thinning of loblolly pine (Pinus taeda) stands to encourage the growth of large, older trees as potential nest sites, and the removal of understory and midstory vegetation mechanically and through prescribed fire. Because such management removes potential nesting and foraging structure for Wood Thrushes, we were interested in evaluating its impact on Wood Thrush population parameters, and using our conceptual model of Wood Thrush populations, we developed specific predictions regarding these possible impacts (Table 1). These hypotheses and predictions operate at two levels of spatio-temporal resolution. Local effects are those that are thought to directly affect survival and reproduction at the stand or finer scale through alteration of the habitats. For example, the number and success rates of nests would presumably be lowered by the removal of trees structurally suited to contain nests; survival might be lowered by the removal of fruiting trees and shrubs. Landscape effects are those that affect the population at a broader population scale, for example, nest site fidelity of returning birds, immigration of birds from other local breeding population, dispersal of juveniles, or intra-season movement among desirable habitat patches (stands). Again, alteration of the landscape through management is hypothesized to play a role here, but detecting these effects requires a broader scale analysis and the explicit consideration of the spatial configuration of treatments. The bottom line for the population, though, is the integration of the local and landscape effects in the production of a fall migratory population, of which a fraction will survive and return the following breeding season, thus completing the annual cycle.

Figure 2. Location of Wood Thrush study area in Piedmont of Georgia.    

conroyPif002.gif (994 bytes)

Table 1. Hypothesized effect of treatments for RCW on Wood Thrush populations.

Local effects
Nest Success Higher Lower
Survival Higher Lower
Density Higher Lower
Landscape effects
Survival Higher Lower
Fidelity Higher Lower
Density ??
Juvenile Dispersal Lower Higher

Our experimental design (Powell et al. 1995; Conroy and Krementz 1997; Fig. 3, Fig. 4) allowed us to test the compartment-level effects directly, by the comparison of density, survival rates, reproduction rates, and other parameters on treatment (subjected to RCW management) and control (not recently treated) compartments, two years prior to and following the treatment application. In addition, monitoring conducted within compartments will allow evaluation (albeit non experimentally) of gradients in fragmentation, for example, due to uneven application of the treatments. The results of the experimental study are being used in conjunction with the SEPM and GIS to make provisional projections and recommendations regarding the impact of RCW management on Wood Thrush (Powell et al. 1997).

Figure 3. Schematic of experimental design for study of impacts of forest management on Wood Thrush breeding population, Piedmont National Wildlife Refuge, Georgia. Treatments are comprised of thinning and prescribed burning for RCW management (see text).

Schematic of experimental design for study of impacts of forest management on Wood Thrush breeding population, Piedmont National Wildlife Refuge, Georgia.

Figure 4. Location of treatment and control compartments for study of impacts of forest management on Wood Thrush breeding population, Piedmont National Wildlife Refuge and Oconee National Forest, Georgia. Letters represent treatment-control pairs.

conroyPif004.gif (1568 bytes)

Estimation of habitat specific survival and movement rates from radio telemetry

As noted earlier, mechanistic models such as patch dynamic and SEPMs require the estimation of (or assumptions about) demographic parameters such as survival rates and reproduction rates, and additional parameters such as movement rates or probability of movement among habitat patches. Empirical estimation and hypothesis testing is needed for these models, for example to evaluate whether a complex model such as a SEPM really is needed (Conroy et al. 1995). A number of studies have quantified reproduction rates of NTMBs, through nest searching and monitoring and other methods. Survival and movement rates tend to be more problematic, usually requiring the use of banding, color marking, or radio telemetry. Design and analysis for survival estimation is straightforward when individual birds occupy single habitat patches over study periods of interest. For example, mark-recapture experiments (Pollock et al. 1990) may be used to provide estimates of survival for samples of birds stratified by habitat type. However, if movements of birds among habitat patches and/or types occurs during the study, extensions of mark-recapture models may be needed to allow separate estimation of patch-specific survival and movement probabilities (Hestbeck et al. 1991, Brownie et al. 1993). If movements are very frequent, methods based on radio telemetry may prove more useful. Conroy (1993b) and Conroy et al. (1996) evaluated methods for analysis of radio marking data, in which birds freely move (e.g., on a daily basis) among habitats, for testing hypotheses about habitat-specific survival and movement rates. Unfortunately, it appears that large marked samples (100 or more) are needed to detect all but the largest habitat specific differences in survival. Because most study designs would fall short of such sample sizes, the issue of parameter estimation and model comparison for complex models, especially SEPMs, remains problematic (Conroy et al. 1995).

Development of an adaptive management framework for research and monitoring

As indicated above, models are frequently used to predict the effects of human impacts on habitats—such as loss through agriculture or urban development or alterations through management practices—on bird populations. Models relating bird and other animal populations to habitats are now included as part of planning tools such as Gap Analysis (Scott et al. 1993) and forest growth models (e.g., Smith et al. 1981), and have been used in conjunction with optimization procedures (e.g., linear programming) to derive strategies for maximization of an objective function incorporating constraints or tradeoffs, for instance, balancing timber revenue with species diversity. Approaches such as these typically have failed to consider one or more difficulties inherent in wildlife models. Commonly the system (e.g., bird population) is treated as a deterministic response to habitat or other effects, and usually only a single model describing habitat-population relationships is used. Models are frequently invoked that have not been properly validated, and for which there are no built in "reality checks."

Given that bird populations, forest landscapes, and other natural resource systems are inherently unpredictable—or predictable only with great uncertainty—and that our understanding of these systems is far from complete, many biologists are tempted to treat modeling efforts with disdain. Another reaction is simply to pretend that these sources of uncertainty do not exist, and to proceed with making management decisions based on models or other information as "the best we can do." An alternative to either approach is to explicitly deal with uncertainty: Not just as a modeling exercise, but as part of the process of making decisions (Lindley 1985). Further, it should be possible to reduce some types of uncertainty through monitoring and research, providing better understanding of how the system under management functions, and how to manage it (Walters 1986).

Johnson et al. (1993) applied stochastic dynamic optimization procedures (Williams 1989, Lubow 1993, Puterman 1994) to the management for the optimal long-term harvest of waterfowl in the face of uncertainty. Crocker and Conroy (1995a, b) extended this approach to the problem of managing for multiple species in a dynamic landscape, with species conservation as a component of the objective function. This approach is explicitly stochastic and allows for model uncertainty and competing models as hypotheses. Previous approaches have included some applications of optimization methods (e.g., Bedward et al. 1992, Pressey and Nicholls 1989) and decision theory (Lindley 1985; e.g., Maguire 1986, Haight 1995, Conroy and Noon in press) but so far none has integrated the achievement of a long-term optimum in an uncertain system, with the iterative incorporation of knowledge, in a truly adaptive sense (Walters 1986).

Consider a simple representation of the species persistence problem, where the "system state" is the number of species persisting (S=0,1,2; Fig. 5). Given the present state (e.g. S=2) we would like to make a decision so as to maximize the average (over uncertain future events) of some objective or "utility", which expresses in this case the tradeoff between species preservation (S=2 having highest utility) and the use of resources for other purposes. In this example the possible decisions are to conserve either 50% or 10% of the total landscape. The utility of the decision-outcome combination represents a value (1 highest, 0 lowest) received by that outcome. For instance, the best outcome would be to conserve both species by reserving 10%, and has utility 0.9; two decision-outcome combinations have utility 0, (no species persist regardless of the decision to conserve). For each decision, we need to make a prediction about the future state of the system. For example, species-habitat relationship models might predict that each of the outcomes S=0,1,2 have different probabilities of occurring depending on the amount of habitat conserved. Finally, the expected utility of each decision is obtained by averaging over all the uncertain outcomes, and the optimal decision is the one that maximizes this average or expectation. In this example, the optimal decision clearly is a=0.5.

Figure 5. Diagram of a hypothetical decision tree for a species conservation problem. Boxes represent decision nodes, circles represent uncertain events. In illustrated scenario, the decision maker observes that there are 2 species present (S = 2) of a hypothetical 2 total, and is faced with making 1 of 2 decisions: conserve 50% (a = 0.5) or 10% (a = 0.1) of the landscape. For each decision, there are 3 possible outcomes:
S = 0, 1, or 2, with probabilities of occurrence denoted by P, dependent on the decision. Each of the 6 decision-outcome combinations has a value or utility (u) to the decision maker, in this case taken as
S/2(1 - a), reflecting 1) the desirability of conserving as many of the 2 species as possible, but 2) imposing a penalty for the proportion of the landscape "removed" from other uses.

Diagram of a hypothetical decision tree for a species conservation problem.

The schematic illustrates a simple, one-iteration process: a decision is made one time. In reality, at each decision-outcome (6 combinations here) the decision maker would again be faced with making a decision, and the optimal policy is a vector of decisions through time. For instance, the optimal decision may be to conserve 75% habitat at t=1, and the rest at t=2, or to conserve none now, but 90% of the remaining habitat at equally spaced intervals, noting that these are not necessarily one year apart. The optimal decision will depend on, among other things, system dynamics (e.g., loss of non conserved habitats) and economic considerations (e.g., commodity and amenity values, discount rates).

We have applied such an approach to a simple system containing four species and two habitat types, with an inherent loss of habitats not conserved, to evaluate the effects of model uncertainty (three simple model structures) and observability (coefficients of variation on the observed system states of 0-100%). The results of preliminary analyses suggest that both model uncertainty and observability have great potential for influencing 1) what the optimal policy is (e.g., how much habitat to conserve, of what type, and when), and in turn have ramifications for the optimal design of survey and research programs for meeting resource management objectives (Crocker and Conroy 1995a, b). Our next steps are to expand this approach to allow more realism, including spatial explicitness and more complex landscapes and species assemblages. We also will work toward a formally adaptive procedure, in which the relative belief in alternative models is updated through monitoring and research (Johnson et al. 1993, Crocker and Conroy 1995a, b).


Models of bird population dynamics and movements can be a useful component of management, research, and monitoring programs. However, model users (and modelers) must recognize that models are vast simplifications of the real world. Further, at best, models represent mathematically our present knowledge about how populations function: No model ever has produced a single additional scientific fact about a species or a system. Nonetheless, models can be useful in summarizing extant knowledge, providing testable predictions based on underlying assumptions, and directing future research efforts, for example, toward understanding key parameters or functional relationships.

On the other hand, models need not be extremely complex or completely realistic to prove useful to management. By simplifying a system to a few key components and predictions, models sometimes can crystallize a problem in a way not possible by other methods. Further, models offer the possibility of examining multiple combinations of possible management scenarios by the mathematical combination of inputs and parameter values. However, users must recognize that these so called "experiments" are simply manipulations of underlying assumptions, are not better than the model and the assumptions on which they are based, and are no substitute for the real thing.

By integrating research and monitoring with management, we can assure that the information gathered is relevant to decision making, and can discriminate between information that is essential for reaching our resource management goal, versus that which would simply be nice to have. Likewise, conservation decisions should continually be reevaluated in the light of new information, and modified as old assumptions change in the light of new data (Fig. 6). Further, the actions of managers will themselves frequently provide new information, particularly if management is done in an experimental manner, e.g., with suitable controls and replication. Viewed in this manner, models are simply one component in a continual process of scientific management.

Figure 6. Diagram of adaptive learning process showing the roles of modeling, research and monitoring, and management.

conroyPif006.gif (1428 bytes)


Bedward, M., R. L. Pressey, and D. A. Keith. 1992. A new approach for selecting fully representative reserve networks: Addressing efficiency, reserve design, and land suitability with an iterative analysis. Biological Conservation 62:115-125.

Brownie, C., J. E. Hines, J. D. Nichols, K. H. Pollock, and J. B. Hestbeck. 1993. Capture-recapture studies for multiple strata including non-Markovian transitions. Biometrics 49:1173-1187.

Conroy, M. J. 1993a. The use of models in natural resource management: Prediction, not prescription. Transactions of the North American Wildlife and Natural Resources Conference 58:509-519.

Conroy, M. J. 1993b. Testing hypotheses about the relationship of habitat to animal survivorship. Pages 331-342 in J. D. Lebreton and P. M. North, Eds. The use of marked individuals in the study of bird population dynamics. Birkhauser-Verlag.

Conroy, M. J., J. E. Anderson, S. L. Rathbun, and D. G. Krementz. 1996. Statistical inference on patch-specific survival and movement rates from marked animals. Environmental and Ecological Statistics 3: 99-118.

Conroy, M. J., Y. Cohen, F. C. James, Y. G. Matsinos, and B. A. Mauer. 1995. Parameter estimation, reliability, and model improvement for spatially explicit models of animal populations. Ecological Applications 5:17-19.

Conroy, M. J. and D. G. Krementz. 1997. Effects of forest management on population parameters and habitat use of Wood Thrushes. Final Report, Work Order No. 22, Coop Agreement #14-16-0009-1551, Georgia Cooperative Fish and Wildlife Research Unit, Athens, Ga. 99 pp.

Conroy, M. J. and B. R. Noon. 1996. Mapping of species richness for conservation of biological diversity: Conceptual and methodological issues. Ecological Applications 6:763-773.

Crocker, J. E. and M. J. Conroy. 1995a. A framework for optimal decision making in the management of landscapes. 80th Annual. Meet., Ecological Society of America, Snowbird, Utah. 30 July-3 August 1995. (Abstract.)

Crocker, J. E. and M. J. Conroy. 1995b. A framework for optimal decision making in the management of landscapes. 2nd Annual. Meet., The Wildlife Society, Portland, Oreg. 12-17 Sept. 1995. (Abstract.)

Dunning, J. B., B. J. Danielson, and H. R. Pulliam. 1992. Ecological processes that affect populations in complex landscapes. Oikos 65:169-175.

Dunning, J. B., D. J. Stewart, B. J. Danielson, B. R. Noon, T. L. Root, R. H. Lamberson, and E. E. Stevens. 1995. Spatially-explicit population models: Current forms and future uses. Ecological Applications 5:3-11.

Haight, R. G. 1995. Comparing extinction risk and economic cost in wildlife conservation planning. Ecological Applications 5:767-775.

Hestbeck, J. B. J. D. Nichols, and R. A. Malecki. 1991. Estimates of movement and site fidelity using mark-resight data of wintering Canada geese. Ecology 72:523-533.

Johnson, F. A., B. K. Williams, J. D. Nichols, J. E. Hines, W. L. Kendall, G. W. Smith, and D. F. Caithamer. 1993. Developing an adaptive management strategy for harvesting waterfowl in North America. Transactions of the North American Wildlife and Natural Resources Conference 58:565-583.

Levins, R. 1966. The strategy of model building in population biology. American Scientist 54:421-431.

Levins, R. 1969. Some demographic and genetic consequences of environmental heterogeneity for biological control. Bulletin of the Entomological Society of America 15:237-240.

Lindley, D. V. 1985. Making decisions. Wiley, New York.

Lubow, B. 1993. Stochastic dynamic programming (SDP). Coop. Fish and Wildl. Res. Unit, Col. State Univ., Ft. Collins

Maguire, L. A. 1986. Using decision analysis to manage endangered species populations. Journal of Environmental Management 22: 345-360.

McKelvey, K., B. R. Noon, and R. H. Lamberson. 1993. Conservation planning for species occupying fragmented landscapes: The case of the northern spotted owl. Pages 424-450 in P. M. Karieva, J. G. Kingsolver, and R. B. Huey (Eds.). Biotic Interactions and Global Change. Sinauer Assoc., Inc., Sunderland, MA.

Pollock, K. H., J. D. Nichols, C. Brownie, and J. E. Hines. 1990. Statistical inference for capture-recapture experiments. Wildlife Monographs 107:97 pp.

Powell, L. A., J. D. Lang, M. J. Conroy, and D. G. Krementz. 1995. Effects of forest management on survival and movements of Wood Thrushes. 2nd Annual. Meet., The Wildlife Society, Portland, Oreg. 12-17 Sept. 1995. (Abstract.)

Powell, L. A., M. J. Conroy, D. G. Krementz, and J. D. Lang. 1997. Hierarchical modeling of forest management effects on Neotropical migrant songbirds. Annual Meeting of the Ecological Society of America, Albuquerque, New Mexico, 10-15 August 1997. (Abstract.)

Pressey, R. L. and A. O. Nicholls. 1989. Application of a numerical algorithm to the selection of reserves in semi-arid New South Wales. Biological Conservation 50:263-278.

Pulliam, H. R. 1988. Sources, sinks, and population regulation. American Naturalist 132:652-661.

Puterman, M. L. 1994. Markov decision processes: Discrete stochastic dynamic programming. Wiley-Interscience.

Odum, E. P. and M. G. Turner. 1990. The Georgia landscape: A changing resource. Pages 137-164 in I. S. Zonneveld and R. T. T. Forman, Eds. Changing landscapes: An ecological perspective. Springer-Verlag, New York.

Sauer, J. R. and S. Droege. 1992. Geographic patterns in population trends of Neotropical migrants in North America. Pages 26-42 in J. M. Hagan III and D. W. Johnston, Eds. Ecology and conservation of Neotropical migrant landbirds . Manomet Symposium (1989).

Scott, J. M., F. Davis, B. Csuti, R. Noss, B. Butterfield, C. Groves, H. Anderson, S. Caicco, F. D'Erchia, T. C. Edwards, Jr., J. Ulliman, and R. G. Wright. 1993. Gap analysis: A geographic approach to protection of biological diversity. Wildlife Monographs 123 123:41 pp.

Smith, T. M., H. H. Shugart, and D. C. West. 1981. The use of forest simulation models to integrate timber harvest and nongame bird habitat management. Transactions of the North American Wildlife and Natural Resources Conference 46: 501-510.

Turner, M. G. and C. L. Ruscher. 1988. Changes in landscape patterns in Georgia, USA. Landscape Ecology 1:241-251.

Van Horne, B., and J. A. Wiens. 1991. Forest bird suitability models and the development of general habitat models. U. S. Department of Interior Fish and

Wildlife Service Research Report 8:1-31.

Walters, C. J. 1986. Adaptive management of natural resources. MacMillan, New York.

Williams, B. K. 1989. Review of dynamic optimization methods in renewable natural resources management. Natural Resource Modeling 3:137-216.

1 U. S. Geological Survey, Biological Resources Division
   Georgia Cooperative Fish and Wildlife Research Unit
   D. B. Warnell School of Forest Resources
   University of Georgia  
   Athens GA 30602