Poster Details
PREDICTING DAILY DISCHARGE OF NON-PERENNIAL STREAMS
Long-term flow records are needed to produce hydrologic classifications in support of ecological assessments. However, in our study area (the arid southwestern US), temporal and spatial data gaps limit analyses of flow regimes. We therefore conducted a pilot project to determine if empirical models could be used to both fill-in missing values from existing gauge records and predict long-term flow dynamics for ungauged streams. We used long-term flow data from 43 basins located throughout the southwestern US to test if random forest models could accurately predict daily discharge (Qd) for individual streams. 50% of observations were used for validation. We modeled each basin's Qd as a function of daily flows observed at the other 42 basins. Model performance (as R2) ranged between 4 and 99%. 29 basins had R2 > 90%, and 12 other basins had R2 > 56%. Two southern Arizona basins had R2 < 13%. Both basins had low mean Qd and extreme desert climates. We are exploring if model performance in these low mean Qd regions can be improved by adding basin-specific climate variables as predictors.
Angela Merritt (Primary Presenter/Author,Co-Presenter/Co-Author), Department of Watershed Sciences, National Aquatic Monitoring Center, and Ecology Center, angela.merritt@aggiemail.usu.edu;
Charles Hawkins (Co-Presenter/Co-Author), Utah State University, chuck.hawkins@usu.edu;