Quantitative Predictions of the Responses of Aquatic Ecosystems to Disturbance II
Presiding: C P Hawkins, Utah State University; L Yuan, U.S. Environmental Protection Agency; A I Pollard, U.S. Environmental Protection Agency
NB41E-01 INVITED 08:30h
Theory Can Help Structure Regression Models for Projecting Stream Conditions Under Alternative Land Use Scenarios
We built multiple regression models for Emphemeroptera/ Plecoptera/ Tricoptera (EPT) taxon richness and other indicators of biological condition in streams of the Willamette River Basin, Oregon, USA. The models were used to project the changes in condition that would be expected in all 2-4th order streams of the 30000 sq km basin under alternative scenarios of future land use. In formulating the models, we invoked the theory of limiting factors to express the interactive effects of stream power and watershed land use on EPT richness. The resulting models were parsimonious, and they fit the data in our wedge-shaped scatterplots slightly better than did a naive additive-effects model. Just as theory helped formulate our regression models, the models in turn helped us identify a new research need for the Basin's streams. Our future scenarios project that conversions of agricultural to urban uses may dominate landscape dynamics in the basin over the next 50 years. But our models could not detect any difference between the effects of agricultural and urban development in watersheds on stream biota. This result points to an increased need for understanding how agricultural and urban land uses in the Basin differentially influence stream ecosystems.
http://oregonstate.edu/Dept/pnw-erc/
NB41E-02 INVITED 08:45h
Simulating Biological Impairment to Evaluate the Accuracy of Ecological Indicators
Environmental management depends on accurate monitoring and assessment of ecological conditions. Bioassessments are generally based on the measurement of selected indicators. However, the accuracy of these indicators is difficult to evaluate because the true biological impairment of a system is almost always unknown. We describe a simulation procedure that allows objective comparisons of estimated indicator values against known impairment. We modeled how densities changed with increasing stress as a function of tolerance values. We applied this procedure to data from five reference sites and compared non-metric multidimensional scaling (NMDS) and observed taxa richness for detecting known impairments. NMDS showed that the trajectories of impairment caused by stress were all evident at 100, 300, or 600 counts. However, the discrimination of stress levels within ordination space greatly improved with increasing count. Estimates of richness based on 100- or 300-count samples often severely underestimated true taxa loss and frequently indicated taxa gain. Estimates of taxa loss based on 600 counts also underestimated true taxa loss, but these estimates were strongly correlated with true taxa loss. This simulation procedure should be applicable to the evaluation of how well a variety of biotic indicators measure biological impairment.
NB41E-03 INVITED 09:00h
Predicting Ecosystem Responses with AQUATOX, a Mechanistic Fate and Effects Model
AQUATOX, a mechanistic fate and effects model, simulates the significant physical, chemical, and biological processes affecting aquatic biota. The user can represent the food web with as little or as much complexity as desired. Generality is balanced with site specificity. The parameters governing the biological processes are designed to be as general as possible, such that a parameter set for a group of organisms should transfer from site to site with little or no recalibration. For example, maximum photosynthetic rates for specific plants should be "global," but the predicted time-varying site photosynthetic rates change with temperature, nutrients, light, and toxic chemicals. AQUATOX also contains informative analytical tools. Control and Perturbed simulations isolate the effects of the differences due to a particular stressor, much like a controlled laboratory experiment. The predicted in situ rates due to ecological processes (such as consumption, mortality, and reproduction) and limitations on photosynthesis can be saved and graphed, enabling the analyst to identify the important processes and environmental controls operating at any given time. Built-in uncertainty analysis allows the user to test which driving variables and parameters are most important to the particular endpoints of interest. Examples of simulations of stream ecosystems will be given.
NB41E-04 INVITED 09:15h
A Bayesian Approach for Combining Data Sets to Improve Estimates of Taxon Optima
Predictions of the responses of stream ecosystems to different types of disturbance would be facilitated by accurate knowledge of the environmental preferences of different taxa. These preferences are often expressed in terms of taxon optima, or the location along an environmental gradient where a particular taxon is most likely to be observed. Several empirical methods exist for estimating taxon optima from a single data set. However, these optima estimates can be biased by the range of environmental conditions sampled within the data set and by environmental gradients that covary with the gradient of interest. Optima estimates can therefore differ between different data sets. By combining different datasets, we can potentially improve optima estimates by increasing sample sizes and better controlling for covarying gradients. Here, we present a Bayesian approach for improving optima estimates using disparate data sets. The Bayesian model provides a means of accounting for factors that are unique to each data set, while estimating a parameter (the optima) that is assumed to be fixed. We present estimates of temperature optima for several aquatic insects that were developed using different data sets. For certain taxa, this modeling approach may lead to more reliable estimates of their optima.
NB41E-05 INVITED 09:30h
Using knowledge elicitation to inform a Bayesian belief network model of a stream ecosystem
The identification of the causal pathways leading to stream impairment is a central challenge to our understanding of ecological relationships. Bayesian belief networks (BBN's) are a promising tool for modeling presumed causal relationships, providing a modeling structure within which different factors describing the ecosystem can be causally linked and uncertainties expressed for each linkage. Relationships can be specified empirically or by knowledge elicitation from a group of experts. We conducted a pilot study to examine the effectiveness of knowledge elicitation for a simple scenario (impairment of a Midwestern, low-gradient stream by excess fine sediments). Five stream ecologists guided by BBN facilitators then defined relevant chemical, physical, and biological aspects of the ecosystem and how the components interacted, and predicted quantitatively how different attributes of the macroinvertebrate assemblage would change in response to increased levels of fine sediment. The exercise provided insights into how best to adapt knowledge elicitation methods to ecological questions, and informed the assembled stream ecologists on the elicitation process and on the potential benefits of this modeling approach. The explicit quantification of uncertainty in the model not only enhances the utility of the model predictions but can also help guide future research
NB41E-06 INVITED 09:45h
A Bayesian Network of Eutrophication Models for Synthesis, Prediction, and Uncertainty Analysis
Accurate prediction in complex adaptive systems is difficult. Even in a simple programmed system where all the rules of interaction are known exactly, unexpected behavior can arise. Thus, forecasting in real ecological systems, where system behavior is often poorly understood, requires characterization of the inherent prediction uncertainty. We developed a Bayesian probability network to predict the effect of nitrogen load reductions on eutrophication symptoms in the Neuse River Estuary, NC. The model consists of a set of sub-models, each independently estimated, and capable of probabilistic prediction. Probabilistic forecasts are propagated through the network in accordance with fundamental probability rules. This approach is extremely flexible as it can accommodate information from detailed process-based models, empirically-based models, probabilistic knowledge elicited from experts, or any combination of these sources. Additionally, the Bayesian framework facilitates model updating, in an Adaptive Management context. As management actions to reduce nitrogen inputs become effective the response of system can be monitored and the new data rigorously assimilated into the model via Bayes Theorem. In the Neuse Estuary this latter feature should be particularly informative because the influences of river flow and nitrogen input are highly confounded making their independent effects difficult to estimate.