Research Proposal: Multi-site Seasonal Streamflow Forecast on the
Upper Colorado River Basin
Advisor: Balaji Rajagopalan
A process for predicting streamflows at multiple locations upstream of the Lees Ferry
gage on the Colorado River is proposed. Lees Ferry is a crucial gage because it is
considered the boundary between the upper and lower Colorado River basins. The
method seeks to increase lead-time on spring streamflow forecasts to allow basin
managers to make more informed operating decisions. Large-scale climate predictors
such as sea surface temperature, and geopotential height will be identified by their
correlation with historical natural streamflow data. An optimal subset of predictors will
be determined and an ensemble of forecasts will be made for the Lees Ferry gage. Upon
evaluating the skill of the predictions at Lees Ferry, a disaggregation model will be used
to extend predictions to upstream gages. The forecasted streamflows can be used with an
existing decision support model.
Introduction and Motivation
Due to rapidly growing populations, water resource managers in the western United
States face increased demand on the region’s renewable water supply. In the western
United States, 44% o the renewable water supply is consumed annually while only 4% is
consumed in the rest of the country (Grantz et. al. 2005). The Colorado River provides a
crucial water source to many states including Colorado, Utah, Wyoming, Arizona, New
Mexico and California. Management of the Colorado River is a complex issue that relies
in part on the on the skilful prediction of seasonal streamflows. Skilful predictions allow
basin managers to understand the impact of operational policy on aspects of the Colorado
River system (Prairie 2006). Recent droughts have increased the importance of not only
skilful flow predictions but also long lead times.
Currently, the primary predictor for streamflow in the upper basin of the Colorado River
is winter snowpack or Snow Water Equivalent (SWE) (Grantz et. al. 2005). Predictions
based on SWE are available in January at the earliest though skilful forecasts of spring
flows are also important for planning and decision-making. A growing body of literature
supports the idea that large-scale climate features (e.g. ENSO, PDO) influence the
regional hydrology of the western United States in a significant way (Regonda et. al.
2006). The identification of large-scale climate predictors (e.g. SST, geopotential height)
offers a forecast of spring streamflows during December through February.
Lees Ferry is the boundary between the upper and lower Colorado River basin located
near the border between Utah and Arizona. The gauging station located there is of
interest in this study. A recently developed disaggregation model provides the means to
expand flow predictions at the Lees Ferry gage to make predictions at upstream gages
(Prairie et. al. 2007). A dissaggregation method captures the spatial interdependency of
the upstream gages rather than simulating them all separately. The method is
parsemoneous because predictors need only be developed for one gage.
The objectives of this study are to:
1. Perform Climate diagnostics in order to identify large-scale climate predictors,
which have an impact on the flow at the Lees Ferry gage (Grantz et. al. 2005).
2. Using an optimal subset of predictors, provide a seasonal forecast of the
streamflow at various lead times and evaluate the skill.
3. Employ a disaggregation model to predict flows at gages upstream of Lees Ferry
and evaluate the skill.
4. The final objective of this study is to provide input to the decision support model
for the upper Colorado River Basin though this is out of the scope of this study.
Figure 1 gives a graphical representation of the objectives of this study.
Figure 1-Flow Chart of Objectives
This section will parallel the objectives section with descriptions of the methodology at
1. A forecast season will first be developed based on historical data. The task of
identifying predictors involves identifying regions of high correlation to
seasonally averaged flows for various lead times. This will be done using the
Climate Diagnostics Center’s (CDC) website (www.cdc.noaa.gov). The average
predictor value over the region of highest correlation is then used in a local
polynomial fit model, which captures local nonlinearity. In addition to
identifying predictors, extreme flow years will be singled out for composite
analysis in hopes of identifying large-scale mechanisms that drive basin-scale
2. A statistical method called Generalized Cross Validation (GCV) will be used to
determine an optimal subset of predictors (Grantz et. al. 2005). The subset of
predictors will then be used to generate an ensemble of forecasts in the form of
probability density functions. The skill will be evaluated using ranked probability
skill scores and likelihood functions (Grantz et. al. 2005).
3. The ensemble of forecasts will be run through the disaggregation model which
may be modified slightly in order to fit the desired number of upstream gages.
The data analysis, generation of correlation and composite maps, identification of
predictors, and determining an optimal set of predictors is expected to take 1-2 weeks.
The end of June will be spent using the LOCFIT package to fit local polynomials and
generate an ensemble of forecasts. July will be spent evaluating the skill of the
predictions, using the disaggregation model to predict upstream flows, and interpreting
data. The end of July and the beginning of August will be spent compiling and analyzing
data and preparing the final report.
1. Prairie, J., B. Rajagopalan, U. Lall and T. Fulp, A Stochastic Nonparametric
Technique for Space-Time Disaggregation of Streamflows, Water Resources
2. Regonda, S., B. Rajagopalan, M. Clark and E. Zagona, Multi-model Ensemble
Forecast of Spring Seasonal Flows in the Gunnison River Basin, Water Resources
Research, 42, W09494, 2006
3. Grantz, K., B. Rajagopalan, M. Clark and E. Zagona, A Technique for
incorporating large-scale climate information in basin-scale ensemble streamflow
forecasts. Water Resources Research, 41, W10410, 1-13, 2005
4. Prairie, James. Stochastic Nonparametric Framework for Basin Wide Streamflow
and Salinity Modeling: Application for the Colorado River Basin. PhD thesis,
University of Colorado, 2006.