Return to the Research Projects at NSIDC page

Improving Operational Streamflow Forecasting in the Upper Colorado River Basin

This project is funded by NOAA OHD


To assess and account for uncertainty with hydrologic modeling with a view to improving operational capabilities


Andrew Slater

Project Summary

The Colorado River Basin is a snow-dominated watershed and provides a useful testbed for this study. Our research addresses the main areas of uncertainty within hydrologic forecasting: inputs, parameters, model structure, and initial conditions.

To address input uncertainty, we developed a method for generating high-resolution ensemble inputs for the hydrologic model. The method, which is based on locally-weighted regression, was used for precipitation generation in regions with complex terrain. We preserved both the temporal and spatial correlation structures of the ensemble variables by using correlated random field in the stochastic sampling strategy. Probabilistic verification techniques showed that our resulting ensembles were statistically reliable and provided good discrimination in terms of having probabilities that differ significantly between cases when specific events occur and when they do not. Our method is flexible and can be extended to other model input fields.

We can reduce the uncertainty of initial conditions through data assimilation. Building on our ensemble generation work, we applied an Ensemble Kalman Filter (EnKF) data assimilation system to the problem updating snow water equivalent (SWE) in the National Weather Service River Forecast System (NWSRFS) module, SNOW-17. The uncertainties of the model and assimilated data were both derived directly from observed data through cross validation. SWE improvements were most evident during the early accumulation and later melt periods of the snow season. Accounting for the temporal correlation in SWE values further improved results. Within the limits of available information, our assimilation results were consistently superior to either the model or interpolated observations.

Recognizing that estimates of snow covered area (SCA) are more readily available than SWE, we proposed a method for assimilating this quantity into models. The effectiveness of SCA assimilation is limited in regions where significant amounts of snow melt occur before bare ground is exposed. For regions with ephemeral snow it could prove a useful strategy, but overall the result points to dual SWE-SCA assimilation as perhaps being most effective.

Structural uncertainty describes uncertainty inherent in a model due to its underlying philosophy and construction, but we were able to disentangle the complex interplay of parameter and structural uncertainty. Within this framework, parameters maintained the same meaning across all models, thus allowing for identification of parameters important to each structure. Different parameters were more clearly identifiable in different models, and parameter sets with the lowest error could vary markedly between different model structures. However, most model parameters examined in the study were poorly identifiable. Poorly identifiable parameters mean that equally accurate streamflow simulations can be obtained in a number of different ways.

Related Resources


Disclaimer: This internet site was prepared by the National Snow and Ice Data Center, part of the Cooperative Institute for Research in Environmental Sciences (CIRES), with support in part from the National Oceanic and Atmospheric Administration, U.S. Department of Commerce, under cooperative agreement NA17RJ1229 and other grants. The statements, findings, conclusions, and recommendations are those of the author(s) and do not necessarily reflect the views of the National Oceanic and Atmospheric Administration or the Department of Commerce.

Contact NSIDC User Services for more information.