Predicting natural channel patterns based on landscape and geomorphic controls in the Columbia River basin, USA

Based on known relationships of slope, discharge, valley confinement, sediment supply, and sediment caliber in controlling channel patterns, we developed multivariate models to predict natural channel patterns across the 674,500 km(2) Columbia River basin, USA. We used readily available geospatial data sets to calculate reach slopes, 2 year flood discharge, and valley confinement, as well as to develop hypothesized landscape-level surrogates for sediment load and caliber (relative slope, percent of drainage area in alpine terrain, and percent of drainage area in erosive fine-grained lithologies). Using a support vector machine (SVM) classifier, we found that the four channel patterns were best distinguished by a model including all variables except valley confinement (82% overall accuracy). We then used that model to predict channel pattern for the entire basin and found that the spatial distribution of straight, meandering, anabranching, and braided patterns were consistent with regional topography and geology. A simple slope-discharge model distinguished meandering channels from all other channel patterns, but did not clearly distinguish braided from straight channels (68% overall accuracy). Addition of one or more of the hypothesized sediment supply surrogates improved prediction accuracy by 4-14% over slope and discharge alone. Braided and straight channels were most clearly distinguished on an axis of relative slope, whereas braided and anabranching channels were most clearly distinguished by adding percent alpine area to the model.