Volume No. 11, 2002
Any successful program develops a momentum of its own, a consensus among its community of participants about what should be done and what the next steps should be. The problem with success―and one that is evident within the GAP community―is that this agreement often concerns tactics, the short-term actions needed to implement long-term goals originally enunciated and tacitly assumed to have remained unchanged. Glance through the programs for recent GAP meetings or look through recent issues of the GAP Bulletin and what you find is emphasis on details of assessing the accuracy of GAP models, incremental improvements on classification procedures, discussion of expert systems for inference of species presence, and, of course, reports of landmarks of progress in the GAP projects in individual states or regions. Yet the more successful the GAP community has been, the more pressing is the need to ask whether all these very worthy activities are still directed at the most useful strategy? We can grant the merits of the original goal of GAP; we can grant the merits of the current efforts to improve GAP incrementally; but we can, and should, nonetheless ask whether the accumulating GAP results indicate any need to redefine the larger GAP vision. Some lessons from the Industrial Revolution may be apt here. Early steam engines pressed for more power output had a habit of blowing up. They could always be made more powerful and safer by overengineering in the light of the available practical experience. But the most rapid advances came when engine performance was analyzed in the light of thermodynamic principles. No longer were engineers restricted to “cut and try” approaches: instead engines could be designed successfully for use in novel environments within which they had never before been deployed. So where, and how, might the GAP community be most innovative in deploying its collective skills and expertise?
GAP’s basis is in mapping the distribution of potential habitat. If GAP scientists unequivocally demonstrate that Kirtland’s warbler (Dendroica kirtlandii) is a denizen of young jack pine (Pinus banksiana) stands in the eastern United States, what exactly does a map of jack pine distribution across Michigan, for example, imply in conservation terms? The origin of GAP was in the notion that it meant a lot. If no stands of jack pine were in some form of protected status, then GAP asserts that one can validly infer that the conservation of Kirtland’s warbler will be enhanced by acquiring protection for some blocks of this habitat. Whether the threshold for effective protection should be 10% or 50% or 90% is thereafter considered to be largely a research and conservation management question, to be solved by incremental research. Here I maintain that one can evaluate such threshold questions in unconventional ways that may be better than the incremental advance possible with conventional thinking. In particular I want to suggest that the GAP concept of species distributions is one of container habitats rather than one of habitat correlates. A container can hold the species but need not always do so, and the relevant strategic questions are therefore, first, how specific the specification of the container is to the species in mind, and second, under what conditions the species will actually be present in the container. In contrast, the habitat correlate concept of species distribution envisages an equilibrium world in which a species is always present in its habitat, and the problem is merely one of obtaining a yet better statistical model with which to describe that habitat. This can, in turn, be best done for a species in equilibrium and yields poor results otherwise. More important yet, though, is that this latter notion holds poorly in the growing appreciation of the role of limits and carrying capacities as constraints on species distribution (Huston 2002, O’Connor 2002).
A GAP assumption is that one can determine a species niche accurately, that one can correctly identify the habitat or environment characteristic of a particular species. GAP sees jack pine forest as a container within which Kirtland’s warblers may occur and assumes that a tight one-to-one correspondence exists between the two: jack pine means Kirtland’s warblers (assumption I) and Kirtland’s warblers mean jack pine (assumption II). Reality quickly cuts in for the first assumption, and it is readily acknowledged that not all jack pine will necessarily hold Kirtland’s warbler. There are two possible reasons for this. The first is that it may not really be jack pine that is the habitat but rather (say) jack pine in which tree density is above some critical density, and if we but knew that fact we could redefine the habitat to be appropriately high-density jack pine stands. This merely moves the logic on one step. But even if we had perfect knowledge of all such issues, the perfectly defined habitat may yet remain only locally occupied. Some sites may be unoccupied because a severe winter (or dispersal or migration stresses) killed the birds that would have occupied them, in which case waiting for the population to build up again and fill these stochastic gaps will resolve a temporary violation of the GAP assumption. But other sites may be unoccupied simply because the population is limited in the long term by factors other than habitat―perhaps the local use of pesticides, lack of winter habitat, and so on. In that case, there are not enough birds around to occupy all the available jack pine habitat (or whatever variant of it needs to be specified to describe optimal habitat), and the GAP assessment is in long-term error. For stock doves (Columba oenas) in Britain in the 1950s, for example, this was the case: thousands of square miles of arable farmland lay open to use, but organochlorine use there limited the population to less than replacement demographic rates (O’Connor and Mead 1984).
Now consider assumption II above, that Kirtland’s warblers mean jack pine. For many species individuals make use of secondary habitat when densities are locally at high levels, and they contract back into the core habitat when the population shrinks (Fretwell and Lucas 1970, Lidicker 1962). For such species the niche is, so to speak, somewhat elastic. At low densities individuals are exclusively in a core, optimum habitat (habitat A; ignoring any influence of site fidelity from previous episodes of high density; O’Connor 1985). At high densities, on the other hand, some individuals are forced into a secondary habitat (habitat B) where breeding is also possible but with less success than in habitat A. (This sequence may extend to a third, fourth, etc., habitat in a hierarchy of breeding suitability.) When researchers determine what to treat as the habitat of this species, their conclusion will depend on the prevailing population level. It will be “habitat A or habitat B” if the information comes from a time of generally high densities, for both habitat types are in use in such conditions. But it will be “habitat A alone” if the information comes from a time of generally low densities. The latter is then merely an analogue of the stenotope example discussed above, but the former will lead us to consider, and possibly protect, areas of type B. But although such protection is designed to be most valuable should the species decline, habitat of type B is the very type of habitat that is not used when population levels are low! Thinking in terms of the principles of habitat occupancy in this way immediately transforms (or, at least, should transform) thinking among the GAP community about the nature of predictive modeling accuracy and about what the concepts of omission and commission error mean.
In the warbler-jack pine example, omission error (failing to predict the occurrence of a species that actually occurs on the site), could result (1) if the jack pine habitat is too narrow a specification of the habitat tolerance of the warbler (e.g., if it routinely uses other stand types than just jack pine) or (2) if it is subject to Fretwell-Lucas dynamics, and the test of accuracy was done at a time of high population while the model of its habitat use was developed at times of low population. I acknowledge that this could be seen as a special case of (1), but the conservation implications are so different they should be kept separate! (The error patterns arising from the other combinations of differential population levels between model development and time of testing are elaborated by Krohn [1996]). Commission error (predicting the presence of a species that does not occur on the site), on the other hand, could result (3) if the habitat container needed to be specified in greater detail than hitherto appreciated (e.g., to allow for a critical density of trees, as above), or (4) if a nonhabitat factor, e.g., the pesticide discussed for stock doves in Britain, was limiting. But commission error will also appear to occur (5) if the census was inadequate to detect birds actually present.
I will assert unequivocally that these principles mean that it is bad science to treat either commission or omission error within GAP methodology as though error were a unitary phenomenon. It is meaningless to report error as 20% or 50% or 80% without considering the different types of error possible. Moreover, it is at least poor science, and maybe should even rate as bad science, to make the unitary measurement for either type of error and then simply to discuss these different origins as possible factors influencing the result. If we know that these ecological processes are at work, surely they should be taken into account in the design of the error measurement. This need not mean gathering huge amounts of new data: simply thinking through the processes involved allowed Schaefer (2002), for example, to set upper and lower bounds on the possible error in GAP assessment for individual species, and these bounds turned out to be far closer to each other than one would naively have expected. In other cases it may indeed be true that to distinguish between some of the possibilities above requires sustained research, and that the documentation of a significant error rate has to be the first step towards establishing the influences of one or more of these factors. Yet I can safely assert that most treatments of GAP error present their results as though they are definitive determinations of error rather than as a delineation of the scope of underlying uncertainty in GAP conclusions. The error assessment process needs, I suggest, to be more conscious of the temporal dynamics of the species and perhaps needs to eschew the use of “found data” (extant data whose collection protocol was not expressly designed for this new use) in quantifying error. For a species that is essentially equilibrial in distribution, one would be justified in setting aside those of my concerns based in biological dynamics. However, between half and two-thirds of all bird species examined both in Britain (Greenwood and Baillie 1991) and in the United States (Boone 1991) have proved to have the density-dependence typically associated with Fretwell-Lucas dynamics, and the methodology used in both the studies cited are such that failure to detect density dependence in any species is more a “not proven” verdict than proof of density independence. Lidicker (1962) and Bowers (1994) provide analogous evidence for small mammals. Most species, therefore, need some thought about their dynamics to figure into the interpretation of error assessment. Even in the absence of empirical data for many taxa, one can conduct the thought experiment of calculating how sensitive one’s conclusions would be to a (say) 25% decline in species abundance or to a 10% density-compensation; if one’s results are not stable against such perturbation of population, they are perhaps too shaky to be offered as firm findings. Moreover, such findings may indicate that better science will result from repeating a previous survey to determine the extent of population stability in a habitat than from doing something new.
However, such issues also affect GAP’s logic with respect to conservation action, for the nature of GAP protection depends on the scale of the spatial extent and dynamics of the species involved. The famous multimillion dollar land swap in Hawaii engineered by Michael Scott and his colleagues (Scott et al. 1987) reflects, I suspect, the effective deployment of GAP thinking in a Hawaiian context limited in space and population excursions. Within the conterminous United States, though, scale is more important. As GAP moves to recommend a 10% protection level, the internal logic on which GAP has historically been based should lead it to accept any 10% of the identified habitat, subject only to concerns for territory size vis-a-vis reserve size. (Though this discussion is in respect to a single species, similar thinking applies to multiple species protection, naturally constrained more by issues of complementarity.) Some species have a mosaic distribution, distributed as small population islands within the block of habitat. Others congregate into one large population within the block. Protecting any 10% of the mosaic distribution may well be as effective as protecting any other 10%, but protecting 10% of the “central occupancy” species is effective or ineffective to the extent that the protected area lies within the core of the species distribution. Yet to know where a proposed reserve lies relative to the distribution of the populations involved requires information that is not intrinsic to GAP projects! Evidence as to the scale on which species dynamics play out is only beginning to accrue, and then primarily for birds (Koenig 1998), but shows a range of dimensions as small as 40 km (rather few species) to several hundred kilometers (many species; O’Connor 1996). Because disjunct populations look like island ones when viewed over large extent, knowledge of scale will surely be helpful in advancing the utility of GAP analysis. But even if we have only crude information about the location of core range, taking the core effect into consideration will yield better conservation decisions than will ignoring the effect (Wilcove and Terborgh 1984).
Many of the same issues critically limit any idea that GAP data can be used as the basis for monitoring populations or communities over time. (If I appear to be flogging a straw man here, as far as most of the GAP community is concerned, it is because I have been present at a workshop in which a leading GAP scientist passionately and persistently argued for GAP to be considered as the basis for population monitoring.) GAP has little to say about monitoring a species that undergoes Fretwell-Lucas dynamics within its fundamental niche. The loss of habitat B above has no immediate conservation significance for a once abundant, but now declining, species that has retreated to habitat A, yet attempting to monitor from GAP data would infer there was such significance (namely, whenever habitat B happened to increase or decrease in ubiquity). It is true that recording losses of habitat A has long-term significance for the species―in the extreme, the habitat might be less abundant than needed to support the remnant population. But a species could decline dramatically in response to new stressors within a GAP-critical habitat that is otherwise unchanged in extent. The loss of stock dove populations from arable habitats in Britain as organochlorine use proliferated is one such example; the declines of osprey (Pandion haiaetus) and brown pelican (Pelecanus occidentalis) in the United States are others. Even where a tight one-to-one linkage between organism and habitat element has been established for GAP purposes, monitoring from GAP data is not possible: one has no certainty that the association of species occurrence and habitat has remained unchanged. In Britain the stock doves switched to breeding in coastal and island habitats, which they previously avoided for competition from rock doves (Columba livia), thereby destroying the dove-arable correlation that would have underlain a GAP assessment of their status. Basically GAP as a monitoring tool fails to pass what I call “the organochlorines test”: if I fly a helicopter across the area monitored, spraying 1960s-era organochlorines over the area, will the program subsequently detect the effects of this treatment on the bird populations present? A monitoring program that registers only the presence of the habitat in the area and assumes, in the absence of evidence, that the species-habitat link continues indefinitely, will fail to detect such habitat-independent losses. This is also Young and Hutto’s (2002) explanation for the failure of many management initiatives based on premanagement correlation studies. Even viewed as a series of snapshots of the state of a species, GAP is ineffective as a monitoring tool unless the underlying species-habitat associations are determined afresh for each snapshot.
What might these issues imply for GAP in its pursuit of utility in conservation? The first message is that GAP’s mapping of potential is closer to the fundamental notion of carrying capacity than are correlation-based analyses of habitat requirements (O’Connor 2002). The source of the issues above lie paradoxically in this strength. GAP is not a poor person’s version of sophisticated habitat models, to be improved by the introduction of better correlation techniques. Instead improvement must come from efforts (1) to understand that the habitat association models behind GAP give different results at different population densities: a founder population may be in optimal habitat, but populations at saturated equilibrium may use far wider niches. For species with slow dynamics any given habitat association result is likely to have long-term validity. For species with fast dynamics the value of any association is likely to be ephemeral. And (2) GAP also needs awareness of the likely implications of a dynamic population fluctuating under a carrying capacity ceiling approximated in GAP by an effectively “habitat container” notion, quite independently of the temporal robustness with which that container was determined. The implications of these two points in relation to the current thrust towards determining the commission and omission rates associated with GAP models were discussed above.
A second message for GAP is the need to think more about the spatial patterning of the habitat containers. The issue of where to position possible protected areas is quite different where blocks of suitable habitat are distributed in mosaic fashion than where the habitat occurs as a single large, contiguous block. This difference essentially originates in the uncertainty as to the level of occupancy of the habitat. GAP in principle proceeds as if occupancy was certain, but in practice a variety of GAP-related studies have felt it necessary to try to correct for the error in this assumption, for example, by proposing the use of likelihood-of-occurrence ranks (Boone and Krohn 1999) or by developing species-specific error estimates (Schaefer 2002). What such methods are trying to do is to improve the specificity of the presumed association of the habitat with the species of interest. A more promising approach might be to incorporate the Dufrêne-Legendre (1997) index, devised to reflect the ecological indicator value of a habitat in respect to a particular species, into the characterization of that association. The Dufrêne-Legendre index for a given habitat and species is
I = 100(n/H)(n/S)
where n is the number of (the given) habitat units that contained the species, H is the number of (the given) habitat units examined, and S is the total number of units (of all habitats) that contained the species (see McCune et al. 2002 for additional explanation and implementation). Such data require that one samples representatively within the species range and within the habitat distribution, perhaps a promising avenue for future GAP investigation. The two fractions in parentheses correspond to assumptions I and II above, and with perfect specificity of association the index is 100. For partially filled habitat containers or for Fretwell-Lucas dynamics, one or the other of the two fractions is reduced, reflecting the lower specificity of the association. Such indices would normally fill a species-habitat matrix, but for GAP this collapses to one value per species and, when summed across the species entering any GAP map, provides a pathway to indexing the reliability of any conservation conclusion drawn.
Finally, let the above not be seen as the routine cry for ever greater elaboration of the research effort. I am well aware of the heroic efforts that have been needed, in the face of shoestring budgets, to make GAP such an effective component of applied science within the contemporary conservation map. Applied science is about crafting or engineering a solution to a science problem under the twin limitations of finite (very finite) resources and limited technical information. But the best engineering, as with the steam engines mentioned above, has always been achieved when fully informed by the principles of physics and chemistry. Moreover, contemporary conservation biology emphasizes principles as much as practice. It will not hurt, therefore, nor will it be an exercise in futility, to ask whether each current GAP activity is more craft than science. Nor to ask if each current task is still strategically important for conservation through GAP. Nor to ask if new and unanticipated questions may surface on thinking about how basic population dynamics permeate GAP. If nothing else, the effort will make for stimulating conversations about the goals, methods, and priorities of GAP.
I thank Michael Jennings and William Krohn for reviewing an early draft of this manuscript.
Boone, R.B. 1991. Construction of a database used in the study of bird populations and agriculture, with a study of density dependence. University of Maine, Orono, Maine. M.S. thesis.
Boone, R.B., and W.B. Krohn. 1999. Modeling the occurrence of bird species: Are the errors predictable? Ecological Applications 9:835-848.
Bowers, M.A. 1994. Use of space and habitats by individuals and populations: Dynamics and risk assessment. Pages 109-122 in R.J. Kendall and T.E. Lacher, editors. Wildlife toxicology and population modeling: Integrated studies of agroecosystems. CRC Press, Boca Raton, Florida.
Dufrêne, M., and P. Legendre. 1997. Species assemblages and indicator species: The need for a flexible asymmetrical approach. Ecological Monographs 67:345-366.
Fretwell, S.D., and H.L. Lucas, Jr. 1970. On territorial behavior and other factors influencing habitat distribution in birds. I. Theoretical development. Acta Biotheoretica 19:16-36.
Greenwood, J.D., and S.R. Baillie. 1991. Effects of density-dependence and weather on population changes of English passerines using a non-experimental paradigm. Ibis 133: Suppl. 1:121-133.
Huston, M.A. 2002. Ecological context for predicting occurrences. In J.M. Scott, P.J. Heglund, J.B. Haufler, M.L. Morrison, M.G. Raphael, W.B. Wall, and F. Samson, editors. Predicting Species Occurrences: Issues of Accuracy and Scale. Island Press, Washington, DC.
Koenig, W.D. 1998. Spatial autocorrelation in California land birds. Conservation Biology 12:612-620.
Krohn, W.B. 1996. Predicted vertebrate distributions from GAP analysis: Considerations in the designs of statewide accuracy assessments. Pages 147-162 in J.M. Scott, T.H. Tear, and F.W. Davis, editors. Gap Analysis: A landscape approach to biodiversity planning. American Society of Photogrammetry and Remote Sensing, Bethesda, Maryland. 320 pp.
Lidicker, W.Z., Jr. 1962. Emigration as a possible mechanism permitting the regulation of population density below carrying capacity. American Naturalist 96:29-33.
McCune, B., J.B. Grace, and D.L. Urban. 2002. Analysis of ecological communities. MjM Software Design, Gleneden Beach, Oregon. 300 pp.
O'Connor, R., and C.J. Mead. 1984. The stock dove in Britain, 1930-80. British Birds 77:181-201.
O'Connor, R.J. 1985. Behavioural regulation of bird populations: A review of habitat use in relation to migration and residency. Pages 105-142 in R.M. Sibly and R.H. Smith, editors. Behavioural ecology: Ecological consequences of adaptive behaviour. Blackwell Scientific Publications, Oxford (BES Symposium Nr. 25).
O'Connor, R.J. 1996. Towards the incorporation of spatio-temporal dynamics into ecotoxicology. Pages 281-317 in O.E. Rhodes, Jr., R.K. Chesser, and M.H. Smith, editors. Population dynamics in ecological space and time. University of Chicago Press, Chicago, Illinois.
O’Connor, R.J. 2002. The conceptual basis of species distribution modeling: Time for a paradigm shift. Pages 25-33 in Scott, J.M., P.J. Heglund, and M.L. Morrison, editors. Predicting Species Occurrences: Issues of Accuracy and Scale. Island Press, Washington, DC.
Schaefer, S.M. 2002. An assessment of methods for testing the reliability of wildlife occurrence models used in GAP Analysis. University of Maine, Orono, Maine. M.S. thesis.
Scott, J.M., C.B. Kepler, P. Stine, H. Little, and K. Taketa. 1987. Protecting endangered forest birds in Hawaii: The development of a conservation strategy. Pages 348-363 in Transactions of the 52nd North American Wildlife and Natural Resource Conference.
Wilcove, D.S., and J.W. Terborgh. 1984. Patterns of population decline in birds. American Birds 38:10-13.
Return to Table of Contents