Today, there are many river basins facing the challenges and complexities of ensuring a sustainable water supply as future water demands increase and threats such as climate change alter the hydrologic regime. The Colorado River, in particular, faces these challenges and many more as the river struggles to meet its over-allocated demand and continued drought. Monthly to seasonal water supply forecasts are essential in planning and managing water resources, particularly for providing water for irrigation, operating reservoirs and water distribution networks, and allocating water among competing users.
Riverside is working with the Colorado Basin River Forecast Center, Colorado State University, and Utah State University to couple advanced data assimilation techniques with distributed hydrologic modeling to provide improved water supply forecasts for the Colorado River basin. In addition, we will work with Denver Water and the Dolores Water Conservancy District to demonstrate how the probabilistic ensemble forecast information can be used to improve water management decision making.
Riverside has developed an innovative approach to hydrologic model calibration that uses multiple objective functions to capture hydrograph fit and expert knowledge to maintain parameter realism. The approach has been extended to distributed models that are an essential component of this research. A framework has been designed for automated parameter optimization in a parallel computing environment, but it also allows interactive evaluation of non-dominated parameter sets. Currently, we are running the parallel NSGA-II calibration for the Animas basin utilizing 28 cores on the NAS Broadwell Node on Pleiades. Some initial results are shown below for one non-dominated parameter set.
Multivariate spatial regression is being used to interpolate Z-Scores of basin snow-water-equivalent (SWE) observations. Long-term modeled SWE will be used to estimate the mean and standard deviation of gridded SWE to allow transformation of the Z-Scores to real-time estimates of gridded SWE. The figure below shows interpolated SWE Z-Scores and associated standard deviations for the Colorado River basin on April 1, 2007. The standard deviations are computed from the posterior predictive distribution at each grid point to estimate the uncertainty of the estimates. These data will be used along with MODSCAG areal snow cover observations in an ensemble Kalman filter data assimilation process.
Colorado River Basin
Application Readiness Level
ARL = 2 (Invention)
Most of the necessary components have been developed in some form, but they need to be integrated in a framework that can be made operational. These components include:
- Algorithms for computing precipitation from GPM. CSU will develop regionally calibrated real-time GPM-based precipitation estimates for the Colorado River basin.
- Community Hydrologic Prediction System (CHPS). This system is operational at the Colorado Basin River Forecast Center (CBRFC) and it will serve as the forecasting system framework for this work. The CHPS framework will allow multiple models to be run in parallel to support model validation and comparison of different forcing datasets.
- Distributed Hydrologic Model. The NWS Research Distributed Hydrologic Model (RDHM) will serve as the modeling framework within the CHPS forecast system.
- Energy Balance Snow Model. Utah State University has developed and tested the Utah Energy Balance Snow Model. This model will be incorporated into CHPS and implemented in key Colorado River basin watersheds so that it can be used for operational forecasting.
- Satellite Snow Products. CBRFC has been working with NASA/JPL to evaluate the potential of utilizing the MODSCAG and MODDRFS snow products for model state updating. This research will develop and evaluate data assimilation techniques for utilizing these products in the distributed hydrologic models developed as part of this work.
- Ensemble Forecasting. CHPS includes functionality for ensemble forecasting, however, it has never been applied to distributed modeling. This research will explore the computational challenges of ensemble processing for distributed models.
Gerald (Jay) Day - Riverside
Christian Kummerow, Colorado State University, Cooperative Institute for Research in the Atmosphere (CIRA)
David Tarboton, Utah State University
Michelle Stokes, NWS Colorado Basin River Forecast Center
Elizabeth McNie, Western Water Assessment
Nicholas Flores, University of Colorado
Shaun Carney, Paul Micheletty, Jonathan Quebbeman, Gi-Hyeon (John) Park, Abigail Watson, Riverside
Collaborators and Stakeholders
Dolores Water Conservancy District
The National Weather Service (NWS) Colorado Basin River Forecast Center (CBRFC) produces the official seasonal water supply forecast used by agencies throughout the basin. CBRFC uses both statistical and deterministic models to forecast future water supply. The deterministic models are conceptual hydrologic models that have been applied on a lumped basis. In the case of the Colorado River Basin, watersheds are subdivided into elevation zones to improve the models’ ability to simulate snow accumulation and melt. The two major hydrologic models used by CBRFC are the NWS Snow-17 model and the Sacramento Soil Moisture Accounting (SacSMA) model. The models require minimal input data, mainly estimates of precipitation and temperature. These models have been applied in a wide variety of hydrometeorological regimes, and they have proven to be robust for operational river forecasting applications, however, these conceptual lumped model formulations cannot easily accommodate data from remote sensor measurements.
In this project, we propose to work directly with the CBRFC Community Hydrologic Prediction System (CHPS) operational forecast framework to deploy a distributed modeling environment that will accept gridded precipitation estimates (e.g., GPM) and facilitate the assimilation of a variety of sensors including SNOTEL point data, as well as MODIS-based snow products. We propose to implement the NWS distributed Hydrologic Model (HL-RDHM), which utilizes the Snow-17 and SacSMA models as a first step for transitioning to operational distributed modeling at the CBRFC. Distributed snow modeling will provide better representation of the spatial distribution of the snowmelt process and should lead to improved forecasts even before advanced data assimilation is introduced.
In parallel, the Utah Energy Balance (UEB) model will be implemented in pilot watersheds to address the difficulties and assess the potential value of incorporating an energy balance snow model in an operational environment. CHPS will provide a flexible environment that will support the management of gridded forcing and measurement datasets, allow the use of multiple models in parts of the basin, and facilitate the incorporation of future advancements. In addition, CHPS will provide a framework that will enable the research team to assess the performance of different components of the forecasting process, i.e., forcing datasets, models, and data assimilation procedures. CBRFC recognizes this project provides a low-risk approach to transition to an environment that will support improved process modeling and the incorporation of advanced observations.
The enhanced forecast system will be used to produce ensemble reforecasts that will allow objective comparison of different operating rules and their impact on water management performance objectives.
Related Research Areas