New Long-Lead Forecasts for the Colorado River using Machine Learning
![Lake Powell land no longer under water](/cadswes/sites/default/files/styles/large_image_style/public/article-image/lake-powell-cover-nps_crop_0.jpg?itok=k8pNahzw)
Operating the Colorado River effectively requires forecasting the highly variable flows beyond the seasonal timescale. The traditional NWS River Forecast Center method based on ESP is not skillful past 5 months when flows are much above or below average. A new modeling approach developed at the University of Colorado is based on the random forest machine learning technique. It has demonstrated improved forecasts at 6- to 18-month lead times using model predictors such as antecedent basin conditions and large-scale climate teleconnections as well as the mean ESP CBRFC forecasts. The research, led by PhD candidate David Woodson, was funded by the Bureau of Reclamation and the National Oceanic and Atmospheric Administration.
The study is published in "Long-Lead Forecasting of Runoff Season Flows in the Colorado River Basin Using a Random Forest Approach" by David Woodson, Balaji Rajagopalan and Edith Zagona in the ASCE Journal of Water Resources Planning and Management.