Landscape Management Lecture Notes for the Beijer Institute Advanced Course on Ecological Modeling Santa Fe Institute, December 1998. Louis J. Gross The Institute for Environmental Modeling University of Tennessee, Knoxville gross@tiem.utk.edu Copyright 1998. Louis J. Gross Outline: Ecological theory and applications Computational ecology Regional management and assessment Prediction for natural systems Landscape-level management Multimodeling Spatially-explicit conrol Sources of uncertainty A relative assessment protocol The ATLSS example Ecological Theory and Applications The inability of models derived from the general theory in ecology to effectively address environmental problems has been one of the strongest criticisms of ecological theory (Shrader-Frechette and McCoy. 1994. Method in Ecology: Strategies for Conservation; Sarkar, S. 1996. Bioscience 46:119-207) and is a theme that arises periodically. This indicates: (1) Theorists are doing a poor job explaining the various purposes for constructing theory and models (2) Much of classical ecological theory may have little to say about practical issues - it was not designed to do so and should not be expected to. There are now a wide variety of new approaches which build upon classical theory, sometimes requiring extensive computation and completely different formulations from the standard dynamical systems approaches traditionally used. Computational Ecology Computational Ecology is "an interdisciplinary field devoted to the quantitative description and analysis of ecological systems using empirical data, mathematical models (including statistical models), and computational technology". Computational methods are essential to deal with many ecological issues, particularly those involving several organismal, temporal, or spatial scales. A Workshop on this subject dealt with 3 general areas of: Data Management, Modeling, and Visualization - report is: THE STATE OF COMPUTATIONAL ECOLOGY (J. Helly, T. Case, F. Davis, S. Levin, and W. Michener, editors) http://www.sdsc.edu/compeco_workshop/report/helly_publication.html Thus computational ecology links together observational data from the field and remote sensing, with mathematical models and computer simulations, and offers the potential to address issues at regional scales for which standard models in mathematical ecology are inappropriate. The tradition in mathematical ecology is to focus on biotic interactions and ignore the dynamics of underlying environmental factors which may be more critical to understanding of natural system response that just the biotic interactions. This is done for very pragmatic reasons - dealing with non-autonomous systems is difficult analytically without strong assumptions on the nature of the driving factors (e.g. assuming periodicity). Data Management: Ecological data are relatively sparse, irregular in character, contains a mixture of data types, and scales of measurement vary widely over time and space. The metadat, used to describe the data, are as diverse as the data itself. Problems then arise in how to maintain such data so as to be usable for diverse researchers with varying hardware and software. Thus standardization is a major concern. Mathematical modeling: To date this has focused on ascertaining general properties of natural systems from basic assumptions. Taking into account stochastic factors, the range of organismal scales from individual through ecosystem, and external forcing functions such as weather and human-controlled impacts represent a very small fraction of the modeling work done to date. Although sufficient computational power now exists to handle such models taking these into account, it is not part of the culture of the field, which appreciates generality over precision and realism. Visualization: A wide variety of statistical techniques have been developed and/or applied to ecological data sets historically to aid in elucidating patterns in these data. Visualization methods have developed to the point where we can emphasize information with particular features in complex data sets. Not only are such methods important for observational data, but they are critical to analysis of model output and comparison of such output to observations. Validation: Very little agreement has been reached on how we decide when a particular applied model is acceptable for predictive purposes - a very contentious area. Regional ecological management and assessment One of the most important functions of the environmental sciences today is to analyze the impacts of human actions on ecosystems and to provide management recommendations to ameliorate these impacts. In all parts of the world ecosystems are affected by the shrinkage and dissection of natural areas, disruptions of natural cycles, and the input of pollutants. The spatial extent of the effects of these anthropogenic impacts range from very local to regional and therefore require assessments that can span these scales as well. Environmental scientists are increasingly using mathematical or computer modeling approaches for impact assessment. Some of these modeling approaches are tailored to deal with small spatial extent concerns such as effects of toxicants on local biological populations. Other approaches, such as analyses of potential land use changes, aim at the county-level spatial level, whereas a few address questions on much larger, regional levels; for example, the problems of northwestern forest management as it impacts the spotted owl. Because most cases of anthropogenic impact include specific problems on a number of different levels, it is appropriate to develop general methods for across-level coupling of models to provide input to the assessment of these impacts on natural systems. Ecological Assessment Ecological assessment refers to the determination of the impacts of various anthropogenic influences on a natural system. Common components of such an assessment would be: Changes in population densities of "important" species, either culturally or economically Biodiversity effects Non-native species introductions Changes in community structure (which may not necessarily be associated with biodiversity changes) Effects of pollutant inputs Direct effects of human actions on the system (e.g. hunting, deforestation, sewage/waste disposal) Indirect effects of human actions (e.g. habitat fragmentation, soil erosion, salinity changes) Coupled with the above for regional assessment would be taking account of the human actions impacts on human systems as well, including: Human population density changes Economic impacts Land use changes and effects on urban/rural/commercial/residential percentages and the long term impact of these on future human needs Agricultural productivity Social/cultural changes Cultural attitudes towards conservation Regional Environmental Issues Over regional spatial extents (e.g. on order of 100-1000's of square km), environmental modeling requires taking account of smaller spatial heterogeneity in underlying habitats, trophic structures, and human impacts. Typical aggregated models, in which a few compartments represent major components of the system (e.g. primary production, nutrients, biomass density) and the model tracks changes in these components through time, require either large spatial data sets to parameterize at regional levels, or else make many assumptions about how basic physical and biotic processes scale from smaller, more accurately understood systems. Large spatial data-sets are few, except for those available from remote-sensing information, making both the construction of defensible aggregated models (as well as the validation of any regional extent model) truly challenging. The most important recent technological advance associated with regional modeling and assessment is the use and availability of Geographic Information Systems (GIS), allowing for the rapid visualization and analysis of two-dimensional images, such as those obtained from satellite or airplane remote sensors. GIS data are readily available for a variety of habitat characteristics, including basic vegetation maps, land-use maps, soil maps, road maps, population density, etc. In utilizing these data however, one must be aware of inaccuracies (e.g. ground-truthing is expensive and difficult to do correctly without long-term support mechanisms), and be aware that to date there are relatively few dynamic data sets available for characteristics which would be needed for ecological assessment (e.g. dynamics of vegetative succession). GIS data, in addition to generally being static and thus providing only a "snapshot" of the system, do not readily allow one to track the animal components of a system, without using some proxy models, such as habitat suitability indices. Such indices have their own inaccuracies, as they assume that localized population estimates may be based totally upon habitat measures, ignoring biotic interactions. Although the technology is available to radio tag and track individual animals, except for a few large mammals and commercial species, this has been too expensive to apply in general (and probably will be for the forseeable future). The above limitations of GIS has led to a call for linking spatially explicit ecological models to GIS data, allowing one to produce dynamic models at local extents within a GIS framework, and allowing at least for the potential to produce models that can analyze the effects of management systems on a variety of components of the natural system, not just those which can be observed remotely. The easiest method to produce a spatially-explicit ecological model is to take a standard ecosystem-type model (e.g. for biomass in different trophic compartments), link it's parameters to local habitat variables available in a GIS framework, run the model independently in each pixel (or some combination of pixels, depending upon the scale for which the model is appropriate), and then link the spatial components by having some movement of state variables between pixels. This is the approach of some commercial packages (e.g. RAMAS/GIS). There are numerous computational and modeling issues associated with this approach, and there are alternatives. When Can an Ecologist Ignore Physiology? Physiology (morass of within-organism characteristics which specify how an individuals life processes respond to environment) seems to skip levels in ecology - very much ignores at population/community level, but important in ecosystems approaches. Why? Focus on nutrient and energy flows leads to reductionist focus on processes. How does physiology enter into other hierarchicval levels: (a) Defines a structuring of populations based upon physiological characteristics - VARIATION (b) Provides a means to couple organisms responses to environment - PROCESS So if focus avoided differences between individuals and was not dependent upon environmental conditions, can ignore physiology - which is exactly case in classical population biology models. If focus is on how individual variability affects population and community dynamics, can still ignore physiology by making assumptions about nature of variability between individuals - p.d.e. and stochastic d.e. approach. I claim that to consider effects of environmental factors requires either a phenomenological approach (based upon observed responses of physiology), or mechanistic approach (process models which track a "standard individual") or individual-based approaches. Prediction for natural systems Computational methods allow us to investigate far more realistic ecological models than we might do otherwise. It is driven by the need to improve our predictive capabilities - to more accurately assess the future impact of human actions on natural systems. The phenomena that ecologists need to include to carry this out frequently operate on spatial and temporal extents larger and longer than any individual can study effectively. This naturally implies the importance of teams of researchers collaborating over long periods, rather than the single-investigator with students approach typical of much of ecology in the past. The difficulty of manipulating and replicating experiments at landscape level raises issues about experimental design, the regularity and longevity of sampling, and the integration and storage of data. A central issue in computational ecology is the need to link dynamic processes that operate across differing spatial regions and at different rates. How do we link natural and anthropogenic forces that influence the demand for biological resources with the dynamics of those resources? How much averaging and smoothing of very high resolution biological data must be done to match the lower resolution of geophysical data while preserving the prediuctive capabilities of the approach for the underlying natural systems? All this is clearly tied in with both what it is we wish to predict, as well as the accuracy desired for such a prediction to be useful. Landscape-level Management: Much of applied resource management occurs at the landscape level, and has the potential to make use of spatially-explicit information (often included in some form of GIS data base) to analyze current and past trends and effects (e.g. on animal population sizes, vegetation community structure, etc.) and make predictions about effects of possible management scenarios. This can involve linking models for landscape change at various extents to the economic impacts of such changes. As of yet, we have experience with very few such approaches (for one example involving a Markov-transition approach to landscape change see the LUCAS Home Page at http://www.cs.utk.edu/~lucas/index.html), and yet regional assessment programs aimed towards comparing various management plans require the type of fairly detailed analysis provided by extensions of such approaches. Multimodeling: Historically, much of modeling in both theoretical and applied ecology has dealt with models that aggregate across a variety of levels (temporal, spatial, and organismal). Thus classical models have been dynamical systems with state variables being the densities of species, and these have served as the basis for much of ecosystem modeling. Taking account of spatially-heterogeneous systems, with different trophic levels having different inherent spatial and temporal extents, requires a mixture of modeling approaches rather than a single one-model-fits-all view. Thus, we have been developing (in the ATLSS project) the methodology for a multimodel (for non-biological examples see the site http://www.cis.ufl.edu/~fishwick/) which combines process-oriented compartment models for the lower trophic levels, structured population models for intermediate trophic levels, and individual-based models for higher-level consumers. Procedures for developing and analyzing such ecosytem-level multimodels in combination with economic and social impact models remains an area of great future importance. ATLSS Home Page: http://www.tiem.utk.edu/~gross/atlss_www/atlss_frame.html Spatially-explicit Control: Management that occurs at landscape level (e.g. forest harvesting, water flow management, conservation preserve design, etc.) is not an all-or-nothing affair that occurs uniformly in space. Rather, realistic management scenarios must take account of spatial heterogeneity in underlying resources, as well as how such heterogeneity interacts with management through time (local ecological succession for example). Given that there are a variety of potential criteria which affect the system management, so that the underlying non-spatial issue may be viewed as a multiple criteria optimization problem, how should the "control" of the system be applied spatially in order to carry out the optimization? This is a little-developed area of applied mathematics, particularly in systems in which there are stochastic factors which interact with the management scheme. Yet it lies at the heart of much of applied ecology today. The Institute for Environmental Modeling Staff Members for ATLSS - Everglades Restoration Faculty: Louis Gross - EEB/Math Postdocs: John Curnutt Phil Nott Full-Time Staff: Charley Comiskey - ATLSS Vegetation analysis Jane Comiskey - ATLSS deer/panther, BPI models Mark Palmer - ATLSS Production run manager Michael Peek - ATLSS database, systems manager Scott Sylvester - ATLSS hydrology, pseudotopography Graduate Students: Holly Gaff Rene' Salinas http://www.tiem.utk.edu/~gross/atlss_www/atlss_frame.html ATLSS Project Manager: Donald DeAngelis, U.S.G.S. Other Personnel: O. Bass, R. Bennetts, J. Chick, P. Fishwick, W. Loftus, K. Rice, J. Trexler, W. Mooij, W. Wolff See: DeAngelis, D. L. et al. (1998) Landscape Modeling for Everglades Ecosystem Restoration. Ecosystems 1:64-75. ATLSS (Across Trophic Level System Simulation) Objective: Analyze the ecological impacts of hydrologic planning across the Everglades region of South Florida. ATLSS as a multimodel: a mixture of modeling approaches based upon the inherent temporal resolution and spatial extent of various trophic components, linked together by spatially-explicit information on underlying environmental (e.g. water, soil structure, etc.), biotic (e.g. vegetation), and anthropogenic factors (e.g. land-use). Current approaches: static spatially-explicit indices, compartment analysis, differential equations for structured populations and communities, and individual-based models. Linking models at very different spatial and temporal resolutions has been a major challenge, requiring a variety of spatial interpolation methods, and careful design of model interfaces and linkages with remote sensing data and GIS. Multimodeling is an essential tool to link together disparate data sets, aid monitoring programs, and produce relative assessments of the ecological impacts at landscape-levels of spatially-explicit management plans. Sources of Uncertainty in Regional Assessments: 1. Uncertainty resulting from lack of knowledge about future climate and weather. 2. Uncertainty resulting from imperfect understanding andrepresentation of major processes in physical and biological models. 3. Uncertainty resulting from imprecise measurements of important physical and biological parameters used in equations that describe processes or initial conditions. 4) Uncertainty resulting from randomness in models with stochastic components How can we reduce the problems induced by such uncertainties for regional planning? a. Adaptive management b. Extensive sensitivity analysis of models and their robustness to model modifications c. For stochastic models, use sufficient replication c. Use RELATIVE ASSESSMENTS: Here, we do not claim that the exact quantitative results from any model are accurate predictors of future changes in the system, but rather that a relative comparison of two or more alternative scenarios provides an accurate assessment of the relative impacts of the different scenarios. We can then provide the public with a ranking of alternative scenarios, which may be the main need rather than exact quantitative predictions of the future from any particular scenario. Advantages of Relative Assessment: If the uncertainties mentioned above do not interact differentially with changes in scenarios (which is reasonable for scenarios produced by methods external to the models being used to assess them, in this case ATLSS), then errors should propagate similarly in model runs on different scenarios. This is TESTABLE by varying uncertain components exactly the same way in two different scenarios and determining whether the relative differences in scenarios change in any significant manner. We have used relative assessment extensively in ATLSS for comparing ecological impacts over a 30-year planning horizon by developing difference graphs and maps. The results have been extensively used by a variety of government agencies and have had a major impact on the planning process.