Predictive Classification Modeling of Freshwater Ecosystems
In many water-rich regions of the world, researchers and managers struggle to develop an understanding of what causes observed variation among ecosystems (and how they respond to change). One of the major constraints to these efforts is often the sparse amount of within-ecosystem (i.e., lake nutrients or stream flow rates) data that are available. Without the data and understanding needed to customize management activities towards individual ecosystems, responsible parties often treat aquatic ecosystems as if they were all the same, despite evidence that one-size-fits-all policies can cause declines in ecological and social resilience (Carpenter and Brock 2004). Therefore, we need ways to transform the vast diversity of freshwater ecosystems into a tractable number for effective management and conservation. One tactic is to classify ecosystems into a more manageable number of relevant classes. Predictive classifications are in contrast to other traditional approaches to classification in which the variable of management interest (which is typically an in-lake variable) is the one that classifies the ecosystems. Our research team is actively engaged in developing predictive classifications in which landscape features (typically widely available through GIS systems, etc) are used to classify lakes. An overview of our general approach to predictive classification modeling is provided in Soranno et al. 2010.
In addition to creating the classifications as end products themselves, we are actively engaged in refining approaches for the development and application of predictive classifications. For example, grounded in our landscape limnology conceptual framework, we recognize that classification modeling should reflect the hierarchical spatial landscape structure in which aquatic systems are embedded. Toward that end, we are comparing the relative contribution/explanatory power of a variety of regionalization schemes, such as ecoregions, hydrologic units, each of which has been developed for different purposes and based on various features. Surprisingly, there has been inadequate exploration of: (a) the relative ability of these regionalizations for aquatic classification purposes has been explored inadequately, and (b) the inferences that can be drawn regarding processes likely driving aquatic ecosystems at regional spatial scales. We have done so for lake chemistry in the state of Michigan (Cheruvelil et al. 2008). Currently we are evaluating the influence of spatial extent and resolution for water chemistry and fish species richness across a five state region (Cheruvelil et al. in prep).
From management and conservation perspectives, a critical step in predictive classification modeling is identification of the ultimate application of the classification before it is developed. Past and ongoing research efforts in our group involve predictive classification development for fish growth rates, fish species richness, and water chemistry.