We at the Ecosystems Research Division have developed an integrated modeling system, linking diverse science modules together in a structered software framework, that can simulate the impact of drivers such as global climate change, landuse change, atmospheric deposition of mercury, and terrestrial loadings of fertilizers, on specific ecosystem services within a region. We developed this modeling package using the Albermarle-Pamlico drainage for our test case, and focused on water quality, water quantity, and fisheries production in headwater basins as ecosystem services of interest. As one of many steps in getting the system up and running, we had to estimate fish communities in streams which had not been previously sampled. Utilizing a large database of sample events from a variety of sources, a cluster analysis was done using relative species abundance. This produced a number of "hallmark" fish communities seen at multiple sites and times throughout the region. This analysis was followed by a predictive discriminant analysis (PDA) using a variety of characteristics of the stream and its watershed to estimate which of the fish communities would be most likely found at that stream. This tool could then be used to predict probable fish communities at unsampled sites whose stream and basin characteristics were known.
Results/Conclusions
A total of 57 communities were found by cluster analysis. Seventeen of these did not occur often enough to be considered indicative of the region. Many of the remaining 40 clusters shared close similarities with other clusters, so I performed another clustering of these samples in hopes of producing more distinctive results. This 2nd-stage clustering resulted in 11 communities that were indeed very distinctive, but obviously more variable than the original 40. Geographic specificity was demonstrated in that these communities were primarily found only in certain ecoregions, either the Blue Ridge/Highlands, the Piedmont, or the Coastal Plains of North Carolina. Discriminant analysis found the following site characteristics most useful in predicting fish communities: stream temperature, water velocity, ecoregion, stream slope, longitude, and elevation. Using cross-validation, the PDA was 48% successful in classifying streams with known communities.
