Citation:

HILL, R., M. H. WEBER, E. FOX, S. G. LEIBOWITZ, AND D. THORNBRUGH. Using StreamCat and the NHDPlus framework to model and map the biological condition of USA streams and rivers. Presented at AWRA Summer Specialty Conference, GIS and Water Resources IX, Sacramento, CA, July 11 - 13, 2016.

Impact/Purpose:

The StreamCat Dataset is a database of upstream features for 2.6 million streams within the conterminous US (CONUS). It was built with the National Hydrography Dataset Plus (NHDPlus) and provides information on natural (e.g., soils) and anthropogenic (e.g., urbanization) watershed characteristics for stream-based studies. In this presentation, we provide an example of using StreamCat in conjunction with the NHDPlus framework to model and predict the biological condition of nearly 1.1 million streams within the CONUS. We found that the visualization of predicted biological condition allowed us to assess the model in ways that were not possible with standard model evaluation techniques. Visual assessments of model predictions helped guide development of the final model and map of predicted biological condition. This study supports the development of a robust national map of stream condition for perennial streams, which is of interest to the Monitoring Branch within the Office of Water. The map produced by this work will also be used to investigate national patterns in health and economic benefits. It also contributes to an FY16 deliverable under SSWR 3.01B.

Description:

The US EPA’s National River and Stream Assessment (NRSA) uses spatially balanced sampling to estimate the proportion of streams within the conterminous US (CONUS) that deviate from least-disturbed biological condition (BC). These assessments do not infer BC at un-sampled streams, nor the anthropogenic stressors that degrade BC. A national map of BC could provide an important tool for prioritizing monitoring and restoration of streams. Our objective was to model and map predicted probabilities of CONUS streams being in good BC. We used the US EPA’s StreamCat Dataset to provide upstream landscape indicators for both model development and application. StreamCat was developed within the NHDPlus (version 2.1) and easily links to this framework to provide both natural (e.g., climate, soils, geology) and anthropogenic (e.g., dams, agriculture, urbanization) watershed characteristics for stream-based studies. We used random forest modeling and data from 1380 streams that NRSA determined to be in “good” or “poor” BC to model and predict the probable BC of all perennial streams within the NHDPlus. The link between StreamCat and NHDPlus allowed rapid application, mapping, and visualization of predicted probabilities. This visualization of predicted probabilities provided important insight and guidance for model development. For example, we compared four candidate models that random forest reported as having similar and excellent model performances. However, despite similar performances, each model produced markedly different maps. In addition, maps of local probabilities helped identify predictions that were sometimes unrealistic. Subsequent modeling and maps showed that these predictions were improved with regional models rather than a single, national model. These insights were not possible through standard model evaluation techniques and helped to improve both the accuracy and reliability of a final national map of predicted BC.

Record Details: