This repository contains the code used to develop CSUFlow25, an updated ungauged streamflow prediction model for Colorado. The models predict 25 different streamflow variables including mean annual and monthly flows, high and low flow metrics, flow timing quantiles, monsoon flow fractions, and bankfull flow estimates. Built upon multiple linear regression using an extensive dataset of 118 stream gauging sites with drainage areas ≤1,500 km², the models incorporate predictor variables spanning topography, climate, geology, land cover, soil properties, and hydrologic regions. The dataset generated in this project covers water years 2001-2024 and implements strict data quality criteria requiring >85% data availability over at least 20 years, with comprehensive screening for transbasin diversions, flow modifications, and urban influences done outside the coding environment.

The codebase includes scripts for preliminary data preprocessing and watershed delineation, predictor variable extraction from multiple national datasets (Daymet, StreamCat, NLCD, MTBS), model development using exhaustive subset selection, cross-validation procedures, and performance evaluation using Nash-Sutcliffe efficiency and percent bias metrics.

For access to the raw data, please contact the ROSSyndicate.