GIS Approaches for Channel Typing in the Columbia River Basin: Carrying Fine Resolution Data to a Large Geographic Extent

We are developing a series of GIS procedures to predict channel types across the Columbia River basin (a total reach length of 444,121 km) using fine and medium resolution GIS data (10-m digital elevation data and approximately 1:50,000 resolution hydrography). However, we have encountered several significant accuracy issues in predicting reach-level attributes from fine and medium resolution data. Our goal is to predict reaches into four mountain channel types (cascade, step pool, plane bed, and pool riffle) and four floodplain channel types (straight, meandering, island-braided, and braided) based on geomorphic principles and GIS techniques. Our channel type prediction heavily relies on the quality of the digital elevation model and a stream layer, as channel slope is the most critical attribute for our channel type prediction. However, we found that channel type prediction is particularly difficult in floodplains because of low vertical resolution in 10 m DEM. To overcome the limitation of the 10 m DEM, we developed and tested several GIS methodologies. We adopted integer as a data type in DEM to shorten the processing time, however, the minimum vertical resolution of the data resulted in poor slope estimates in wide floodplains. Surprisingly, reverting to use of floating point data made little improvement in estimation of channel slope. To overcome data resolution and time constraints, we developed a method to aggregate multiple reaches based on attributes for the slope estimation. Our method to estimate floodplain also solely depends on DEM. After we tested multiple approaches to estimate bankfull width including piecewise linear regression, an empirical bankfull width model was constructed with drainage area (D) and mean annual precipitation of the drainage (P) (BFW = 0.177-D0.397 -P0.453, r2 = 0.845, p < 0.001, df = 267). Then, floodplain width was calculated based on the bankfull width and estimations were compared to field measurements (Peason's correlation coefficient = 0.698, p < 0.001, df = 55). Similar to the slope estimation, we found that accuracy of floodplain estimation became lower in wide floodplains. We also evaluated accuracy of stream line data by comparing our data with a 1:24,000 stream layer.