Unsupervised hydrologic classification of rivers: Watershed controls on natural and anthropogenic flow regimes, Alabama, USA

摘要

Watershed hydrology has often focused on modelling studies of individual watersheds, which consider each river system as unique. Classification is an alternative approach that instead focuses on the similarities among different watersheds. Although both supervised and unsupervised hydrologic classifications have been developed, few previous studies have used classification to assess the degree of anthropogenic modification of hydrologic regime. Here, we conducted an unsupervised hydrologic classification of 189 U.S. Geological Survey gages, including 41 minimally impacted gages from the Hydro-Climatic Data Network (HCDN), in the five major interstate river basins in the U.S. state of Alabama. For the natural classification, the most significant predictor variables for cluster membership were related to compressive strength of bedrock, bedrock depth, hydraulic conductivity, elevation, temperature, and soil texture, and several land-cover variables were also significant in the anthropogenic classification. We then developed two random-forest models: one based on all 189 gages using both natural and anthropogenic variables from the Stream-Catchment (StreamCat) dataset and one based on the 41 HCDN gages using natural StreamCat variables only. We used the random-forest models to predict natural and anthropogenic normative hydrologic class for over 158,000 National Hydrography Dataset Plus catchments in the study area. Catchments that changed their class between the natural and anthropogenic classifications can be identified as those that have a large amount of anthropogenic influences on their hydrologic regime, including many catchments on the coast, in the north-western Coastal Plain, in the Interior Low Plateaus, and in the Piedmont. Using unsupervised hydrologic classifications is a promising approach for uncovering the physical processes that affect hydrologic regime. There are also potential applications in river management, including predicting the hydrologic behaviour of ungaged watersheds, identifying relatively unimpaired rivers to serve as conservation and restoration targets, and regionalization of environmental instream flow standards and climate-change impacts.