By Mark Scott, Guest Blogger

The noted flowing-water ecologist H.B.N. Hynes wrote about the stream and its valley, pointing out that water quality, channel form, biological communities, and other stream conditions are linked to characteristics of the drainage basin. Countless studies have since reinforced that observation, and although Clean Water Act provisions to address point source pollution have allowed U.S. waters to rehabilitate significantly since the days of rivers afire (see: Cuyahoga River), we are still grappling with effective ways to control nonpoint sources. It’s relatively simple to measure the concentration of a contaminant coming out of a pipe and relate it back to toxicology studies to set protective limits on discharges. But the diffuse runoff from landscapes containing a cocktail of potential contaminants interacting with fine sediment, not to mention the highly fluctuating flows delivered from impervious surfaces, complicates things to say the least. It’s not as easy to figure out how much landscape disturbance is too much, or pinpoint how much green space will need to be restored to rehabilitate a stream degraded by nonpoint sources.

The South Carolina Department of Natural Resources, working with researchers at Clemson University, collaborated to make a tool available to help answer these sorts of questions, driven by the information gleaned from the South Carolina Stream Assessment (SCSA). The mapper and decision tool uses data collected during the probability-based assessment of wadeable streams to predict conditions in streams all across the state. The predictions are based on modeled relationships between instream data and watershed characteristics obtained from the National Fish Habitat Assessment (NFHA) developed by Wang and colleagues (2011). While SCSA provided data on water quality, stream habitat features, and biological communities, NFHA provided information on topography, land cover, point sources of pollution, and dams, among other things, attributed to catchments in the National Hydrography Dataset Plus.

Under the classic concept that the stream is influenced by its valley, we used the NFHA data to develop predictive models of select fish assemblage metrics at our ~400 SCSA sample sites. We employed a form of machine learning called random forest, which has demonstrated favorable results when used with robust sampling designs, to create the models in R Statistical Software. Random forests require no assumptions about distributions, error structures, or functional forms in data, handle outliers well, and capture non-linearities such as threshold responses. Identification of thresholds is a particularly relevant bit of information in making management recommendations and actions. Other useful features include a ranking of predictor importance and plots of functional relationships depicting the behavior of the response while varying a particular predictor across the range of values, holding other predictors constant. Each model is saved and loaded into the South Carolina Stream Conservation Planning Tool.

In the Map Viewer section of the tool, the user selects a response of interest – for example, richness of conservation priority fish species- and static maps are presented with predicted conditions for all 1:100K scale catchments draining wadeable streams, powered by Esri’s ArcServer. It includes various base maps and layers selectable to aide orientation with the typical ability to zoom in to local areas of interest. The most useful capability for conservation planning is found in the ‘Create Alternative’ box, where one can select a pointer tool to pick a catchment of interest. The most important anthropogenic predictors from the model pop up with their values, along with slider bars for creating alternative scenarios.

Upon making a ‘what-if’ change to a predictor (or predictors) in one or more catchments, the model is called to make dynamic prediction using the new values, and the result is returned and mapped. Not only is the selected catchment mapped, but all downstream catchments are re-calculated and mapped as well, providing a visual reminder to the user of the connectedness of aquatic systems, how cumulative effects propagate, and the consequences of land management to aquatic ecosystem integrity. Because the predictions apply only to wadeable streams, the downstream analysis only applies to catchments <150km².

By selecting multiple catchments, users can cover greater spatial scales to investigate a host of possible scenarios. For instance, one could map predicted aquatic responses to proposed municipal growth plans, search for locations where installing a given amount of greenspace would be expected to produce the greatest conservation result, or map expected consequences of proposed development projects in a watershed.

For more information on SCSA, our modeling approach and error estimation, and to explore the Stream Planning Tool, go to this link.

Citations

1. Wang , D. Infante , P. Esselman , A. Cooper , D. Wu , W. Taylor , D. Beard , G. Whelan and A. Ostroff. 2011. A Hierarchical Spatial Framework and Database for the National River Fish Habitat Condition Assessment. Fisheries 36(9): 436-449