Development of empirical models to estimate channel dimensions in the contiguous US
On this page:
Stream channel dimensions influence many critical features of stream ecosystems, such as temperature, habitat quantity, and water quality, and serve as key inputs for many ecohydrological and habitat models used for management. Although field measurements of channel dimensions are ideal to characterize habitat and parameterize management models, they are not always available, and obtaining parameter measurements can be costly and limit the spatial extent of modelling applications. Therefore, estimates of channel dimensions at unmeasured locations could facilitate the characterization of streams when measurements are unavailable for initial model parameterization, and help improve the management of these ecosystems. We used empirical modeling to interpolate stream widths and depths to 1.1 million stream segments across the contiguous US. Specifically, we created four models: (1) wetted width, (2) thawleg depth, (3) bankfull width, and (4) bankfull depth. To do so, we used channel dimension measurements at 3,233 sites from the USEPA’s 2008/09 and 2013/14 National Rivers and Streams Assessment. Stream dimensions were modeled with random forests and predictor variables from the USEPA StreamCat dataset, which contains several hundred watershed metrics of land use, climate, geology and others. The wetted width, thawleg depth, and bankfull width models performed well with high r-squared values (≥79%) and low root mean square errors relative to observed widths across low to high stream orders (RMSE ranges from 1-10 order streams - wetted width: 4.16-351.60 m, thawleg depth: 0.30-4.70 m, and bankfull width: 5.60-447.64 m). The bankfull depth model performed less well, explaining just 40% of the variation in measured values. We tested regional models of bankfull depth but found no improvement in model performance. A possible reason for poorer performance in the bankfull depth model could be the difficulty of identifying and measuring bankfull stage in the field and the sensitivity of depth measurements relative to bankfull width. Similar models in the literature were unavailable to compare with our models of wetted width, thalweg depth, and bankfull depth. However, we could compare the bankfull width model to another model from the literature. Through this comparison we observed that our model produced a similar, and sometimes better, spread of regional predictions, r-squared values, and root mean square errors. These results indicate that this modeling effort improved on previous models and point to the potential utility of our interpolations when measured channel dimensions are unavailable. Upon publication of these models, we plan on making the interpolated values for 1.1 million stream segments publicly available through the StreamCat dataset, which will facilitate their use by aquatic resource managers and researchers.