CSUFlow25: Colorado Streamflow Ungaged Prediction Models

Kathryn Rosalie Willi , Stephanie Kampf, William Keenan, Juan De La Torre, Matthew RV Ross

CSUFlow25 is an updated set of statistical models for predicting streamflow in ungaged basins in Colorado, USA. It builds on the earlier CSUFlow18 models (Eurich et al., 2021) by extending the streamflow record from water years 2001–2018 to 2001–2024, expanding the pool of candidate predictor variables, and increasing the number of streamflow metrics predicted. Models were developed from a rigorously screened dataset of 131 watersheds in Colorado and surrounding states, with sites filtered for transbasin diversions, large dams at watershed outlets, urban land use, and nested watersheds to better represent natural streamflow. Multiple linear regression models were fit for mean annual and mean monthly flows, high and low flows (Qmax, Q95, Qmin, Q5), the date of those flow quantiles, the fraction of annual flow during the monsoon season, and bankfull (1.5-year return period) flow, using predictors describing topography, climate, geology, land cover, soil, and hydrologic region. Model generalizability was assessed with 5-fold cross-validation and evaluated using the Nash–Sutcliffe efficiency (NSE) and percent bias (PBIAS). Performance is strongest for mean annual flow, snowmelt-season monthly flows, high flows, and bankfull flow, and weaker for low flows, winter months, and the monsoon fraction. Across nearly all months CSUFlow25 outperforms both CSUFlow18 and the development version of USGS StreamStats. The models are implemented in a publicly available R Shiny application that delineates a watershed from a user-selected point, compiles predictor variables, and returns predicted flow values with 95% confidence intervals.