National Weather Service California Nevada River Forecast Center Department of Commerce National Oceanic and Atmospheric Administration National Weather Service

General Ensemble Theory for Streamflow Forecasts

NWS River Forecast Centers across the U.S. generate short-term flood forecast guidance at nearly 4000 locations. The modeling system is a collection of data processing, hydrologic simulation models, and associated interfaces. The modeling system has three subsystems: (1) calibration, (2) operations, and (3) ensemble. These three subsystems share architectural and information resources.

Hydrology Modeling components - calibration, operations, and ensemble

NWS hydrologists use the Calibration System to select appropriate models, develop and adjust parameters and fit the resulting simulated flow to that historically observed. This also includes integrating those parameters and supporting realtime data into the Operations System.

The Operations System is used daily by NWS river forecasters to produce short term river discharge forecasts and guidance. Real time data are used to update model states (soil moisture, snow water, etc.) Forecasts of precipitation and air temperature are included in the simulation of the “future” to develop the streamflow forecast. Simulations are deterministic (single value) and extend 10 days into the future. The CNRFC uses a standard 5-day horizon for operational forecasts. You can view these forecasts at:

The Hydrologic Ensemble Forecast Service (HEFS) utilizes the same models and states as the Operational System, but in a slightly different fashion. In this case, the Operations System is run, starting with today's current conditions, once for each ensemble member. Most watersheds in the CNRFC area are calibrated with between 30 and 40 years of data. Thus, the HEFS is designed to produce a similar number of ensemble members . These possibilities, called "traces", can extend for up to one year into the future. Each streamflow trace is influenced by short-term meteorological forecast information, and as forecast skill diminishes with longer lead times, the climatological meteorological data is the stronger influence. Beyond 28 days of lead time, the streamflow traces are 100% driven by climatological meteorology (see graphic below).

Diagram showing timeline of HEFS forecasts.  Observed weather occurs, followed by weather model input for days 0 through 28, and then climatology for days 29 through 365.  Streamflow forecasts are produced from this and current watershed conditions.

Each member of meteorological forcings (precipitation, temperature, and freezing level) are processed through the same hydrologic models as the deterministic forecasts with the same starting watershed conditions, one member at a time. Probabilities can be calculated from the collection of ensemble streamflow forecasts.

Below is a figure of streamflow traces for the Cosumnes River at Michigan Bar starting on January 7th, 2017, and going out 30 days. You can see that the traces are much more aligned in the first few days, and become less uniform for the longer lead times. The HEFS system relies more on climatology as weather model skill degrades with longer lead times. As stated above, the ensembles rely 100% on climatology after day 28. As of October 2022, the climatology used in HEFS covers years 1980-2021. At the beginning of a new water year, another year will be added to the climatology, increasing the total number of years by one.

Example of HEFS traces for Consumnnes River at Michigan Bar, showing traces uncertainty increasing with time.

There are a number of short-range interactive products displaying statistical information that is focused on the first 10 days. They can be found by clicking on an ensemble location on the CNRFC map: The Build Your Own tool ( is another interactive tool that users can use to generate probabilistic information, and is designed for longer-range analysis going out months/seasons/water year. The ensemble data utilized by the Build Your Own tool is all full-natural flow (no regulation effects are included in the forecasts) since the forecasts go out a full year and little is known about regulation effects that far out in the future.