Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
8 Dealing With Uncertainty This chapter concerns the management of ecological systems in the face of uncertainty. Some uncertainty is unavoidable in ecologists' predictions about ecological systems, but decisions that might lead to unexpected environmental changes have to be made. To deal with uncertainty in ecological prediction, we must first identify its sources and conse- quences this information is often the most useful that an ecologist can give to an environmental manager. SOURCES OF UNCERTAINTY Many ecological systems among them those most affected by human activities are poorly understood. We often lack the detailed information necessary for making accurate predictions, particularly under the unusual conditions that we impose on ecological systems. Several broad categories of uncertainty make the precise prediction of ecological changes difficult. Complexity Because the relationships among species are complex, changes in one can lead to unexpected changes in another. Indirect effects those prop- agated through a number of species links are particularly difficult to predict (Brown, in press). For example, when DDT was introduced into terrestrial environments, no one imagined that it would eventually be found in marine fish (Chapter 24~. Equally hard to predict are the effects of 88
DEALING WITH UNCERTAINTY 89 multiple perturbations, such as the introduction of different kinds of pol- lutants from multiple sources into a river (Chapter 9~. The response of a population or community can become nonlinear at some degree of perturbation, perhaps with a threshold beyond which the response changes qualitatively. Fish populations can undergo rapid col- lapse when overharvested (Murphy, 1977; Ricker, 1963), and lake com- munities can undergo distinct changes in composition when nutrient concentrations change (Chapter 204. Continued reduction and fragmen- tation of habitat can also lead to the extinction of local populations and whole species (Chapter 171. In some cases, nonlinearity of population dynamics can lead to chaotic behavior even in environments that are homogeneous in space and time (May, 1981; Chapter 21. Natural Variability Populations and communities vary in space and time because of both intrinsic processes and changes in their physical and biological environ- ments (see also Chapter 51. In most species, reproduction is seasonal, and mortality is neither constant throughout the year nor equally distributed with respect to sex and age. An estimate of the makeup of a population or community at any particular time is like a snapshot often one with poor resolution of one state among many possible states. The responses of populations and communities are all too often influenced by their site. The presence or absence of a particular species, for example, can change a community's response to environmental change. Ecological systems commonly respond in a site-specific or situation-specific fashion, so our predictive capability is likely to be poor when appropriate analog studies for nearby or similar sites are unavailable. Temporal variability in ecological systems can have several implica- tions. Some systems are highly variable because of periodic natural dis- turbances or intrinsic cyclic behavior. For example, the marine intertidal communities on the western coast of North America are subject to severe storms that can temporarily destroy them. Many organisms in such systems are adapted to those disturbances and have characteristics that facilitate recovery (Sousa, 19841. Changes in the systems must be interpreted through an understanding of the natural history of the organisms present. Similarly, changes in plankton communities that undergo seasonal turnover must be interpreted in the light of population changes in natural conditions. Changes in climate can also complicate assessment of human-caused environmental changes. For example, it is often unclear whether changes in fish populations are due to fishing pressure or to changes in ocean conditions (e.g., Chapter 12; Parrish and MacCall, 1978; Steele, 1984;
90 KINDS OF ECOLOGIC KNOWLEDGE ED THEIR PLACATIONS Wooster, 19831. Environmental changes can interact with biological phe- nomena to produce large and complex changes in population numbers and perhaps alternative stable states (Steele and Henderson, 1984; Wooster, 19831. Changes in climate can be severe but brief, such as El Ninos (e.g., Barber and Chavez, 1983), or they can take place over hundreds of years with large effects on the animals and plants distributed over wide areas (Bryson and Murray, 19771. Natural variability makes it difficult to establish baseline conditions. The possibility that changes in populations or communities are due to natural changes in the abiotic environment should always be kept in mind in making predictions about ecological systems. Random Variation Because many forces acting on populations are more or less random, populations behave in probabilistic rather than deterministic ways, and their responses to perturbations can never be predicted precisely. Popu- lation numbers can fluctuate randomly and on a periodic or cyclic basis seasonally or over longer or shorter periods. If the scale of observation does not match the scale of natural fluctuations, even cyclic (e.g., sea- sonal) fluctuations can be perceived as random noise that makes deter- mination of the state of a system difficult. Because variation is sometimes random and sometimes systematic, we need to know about the variability of important population or community measures mean values are not enough if we are to manage them adequately and predict the effects of changes. Errors of Estimatior' In addition to errors stemming from random variation, measurement error is unavoidable, and it can be large. For example, independent ob- servers counting migrating animals can disagree substantially if the animals pass them rapidly and in great numbers. Estimates of population size must often be made indirectly, as when fish catch statistics are used to estimate stock size (e.g., Chapter 121. Whether estimation is direct or indirect, inevitable errors of estimation add to the uncertainty of management pro- jections. Errors of estimation commonly combine with natural variability to make the detection of ecological effects challenging. Sampling programs nec- essary to distinguish an effect from background noise become more elab- orate, more time-consuming, and more expensive as the accuracy of estimation required and the extent of natural variability increase. In view
DEALING WITH UNCERTAINTY 91 of the practical difficulty of distinguishing among effects, it is surprising that we traditionally worry about concluding that an effect has occurred when it has not (type 1 error), but rarely worry about concluding that an effect has not occurred when it has (type 2 error) (Zar, 1976~. If multiple type 2 errors are made for a given class of environmental change, serious cumulative ecological effects can occur (Chapter 91. Lack of Knowledge We often lack experience with a particular type of environmental per- turbation. When neither appropriate studies nor reliable theories or models are available, accurate prediction is not possible, and it might be necessary to conduct extensive ecological studies as an aid to management decisions. However, if even the best possible ecological studies are inadequate to support prediction of acceptable accuracy, a planned project or action must be viewed as an experiment in itself (Chapters 6 and 21~. MANAGING IN THE FACE OF UNCERTAINTY The point of discussing the many obstacles to making accurate predic- tions is not to argue the futility of trying, but to show that the process of prediction must be viewed as complex and probabilistic. An appropriate approach to managing ecological systems recognizes the random com- ponent of population dynamics by dealing with measures of system var- iability, the possibility of nonlinearity, and the consequences of errors (e. g., Beddington 1984a,b; Walters, 1 9841. For example, attempting to achieve maximal sustainable yields in many fisheries can result in more variable yields, greater population fluctuations, and a greater likelihood of population collapse (May, 19801. Regulating a fishery at higher stock size can stabilize yield and decrease the likelihood of catastrophic failure of the fishery. When populations are very variable and exceeding a critical threshold would have severe consequences, the best strategy could be to hedge one's bets to reduce risk by trying not to exceed the threshold (Beddington, 1984a; Jewell and Holt, 1981~. When we have little understanding of either the major potential effects of a perturbation or the dynamics of the receiving system, our predictive capability is likely to be extremely poor. Developing a project or initiating a management plan can then be viewed as conducting a large-scale ex- periment. Information from well-designed monitoring studies can allow management to adapt to the occurrence of unexpected effects (Holling, 1978) and can be valuable in the planning of similar actions (Chapter 12~.
92 KINDS OF ECOLOGIC KNOWLEDGE ED THEIR PLACATIONS In some cases, particularly those involving the management of renew- able biological resources, information can be gathered by managing ex- perimentally purposefully manipulating the system to learn about the effects (Bar-Shalom, 1976; Beddington, 1984a; Ludwig and Hilborn, 1983; Walters, 19841. Models can aid in conceptualizing a problem and in developing hypotheses that can be used to design research and monitoring studies (Holling, 19781. The use of multiple working hypotheses permits one to compare the potential effects of several possible perturbations. A number of methods are available for dealing with randomness in ecological systems. Error analysis is a formal method of dealing with random effects when models are formulated explicitly (Meyer, 19751. Random error is inserted into a model to simulate variation, and the model is explored by Monte Carlo simulation or simple first-order analytical techniques (Walters, in press). This method of modeling has been used to determine the minimal viable population for endangered species (Shaf- fer, 19811. Survival is a probabilistic affair, and one who would determine the minimal tolerable population size must decide for how many gener- ations a species is to be protected and the acceptable probability of ex- tinction. Such modeling efforts are usually limited to analyzing the effects of random error, but it is also possible to model systematic errors, in an attempt to identify potential causes of extinction. Sensitivity analysis is a method for assigning different values to the parameters of a model to explore the consequences of errors in choosing and measuring them (Meyer, 19751. The analysis helps to identify the variables to which the system is most sensitive. It permits exploration of alternative management plans and can be used to decide which parameters need to be estimated most accurately. The obstacles to making accurate quantitative predictions of the behavior of populations and communities are formidable. The natural variability and complexity of ecological systems will always limit our ability to make precise predictions. In the long run, a better understanding of natural variability and of the factors that make some systems more responsive to change than others will improve our predictive ability. But as long as we continue to alter our environment in new ways, uncertainty will always be associated with the effects of our manipulations. Environmental ma- nipulations will always be experimental to some extent, and our most promising course is to structure each one so that we can learn as much as possible from it.