Seasonal streamflow forecasts: Difference between revisions

Revision as of 18:46, 5 April 2021

Overview

Seasonal streamflow forecasts, also called seasonal water supply forecasts, are widely consulted by water managers and water users throughout the Colorado River Basin to inform general expectations for the coming runoff season and specific operational decisions such as reservoir release schedules and water allocations. Forecasts are issued from mid-winter (Dec/Jan) through July for a several month forecast period: April-July (most common), March-June, or May-August, depending on the usual runoff peak.

There are two primary federal providers of operational seasonal streamflow forecasts for the basin:

Colorado Basin River Forecast Center (CBRFC), one of the 11 NOAA National Weather Service River Forecast Centers
Natural Resource Conservation Service (NRCS) National Water and Climate Center

CBRFC and NRCS use different modeling approaches and their respective forecasted streamflow volumes for the same forecast point may differ. Before 2012, they coordinated their forecasts so that their forecasted streamflow volumes were the same.

CBRFC seasonal streamflow forecasts are critical inputs into Reclamation's operational models, 24-Month Study and MTOM.

Methods

In the snowmelt-dominated headwaters of the Colorado River Basin, predictability of streamflow on seasonal timescales in the basin arises from initial moisture conditions: the observed snowpack and soil moisture conditions. The skill of the forecasts increases through the winter and spring months as the accumulating snowpack provides more information about the upcoming runoff. Soil moisture deficits (or surpluses) in the fall, before snow begins to accumulate, indicate whether an an unusually low (or high) proportion of the snowpack will become runoff in the stream.

Lower-elevation watersheds which have little or no snowmelt contributing to the annual hydrology cannot be forecast as skillfully as higher-elevation watersheds that are snowmelt-dominated. In all cases, the primary source of forecast uncertainty and error is due to lack of knowledge of how the weather and climate will evolve between the date the forecast is issued (e.g., April 1st) and the end of the forecast period (e.g., July 31st). Other uncertainty and error comes from inaccurate depictions of the snowpack and/or soil moisture as input into the forecast model, and biases in the model itself.

The CBRFC uses a conceptual (or simple dynamical) modeling framework for its seasonal forecasts, coupling a snow model (SNOW-17) with a rainfall-runoff model (Sacramento Soil-Moisture Accounting; SAC-SMA). Conceptual models draw from our understanding of the physics of the real-world watershed, but have very simplified representations of watershed attributes and processes, with 5-20 components and tunable parameters to relate them to each other. CBRFC runs their models throughout the forecast season (December-July), using historical daily precipitation and temperatures from each year of 1981-2015 to stand in for the unknown future weather. This creates an ensemble of 35 forecast traces, each starting with the same observed initial moisture conditions but individually evolving the modeled snowpack and runoff through the end of the forecast period given the winter/spring weather of 1981, 1982,...2014, 2015. These ESP (Ensemble Streamflow Prediction) forecasts are issued daily, and then become the primary input to the official forecasts issued monthly or semi-monthly.

NRCS uses a statistical modeling framework, employing separate regression models for each gage that are calibrated on the historical relationships between streamflow and several watershed predictors, typically snowpack (SWE) and accumulated water-year precipitation at 2-5 SNOTEL sites. Statistical regression modeling has been the standard-bearer for seasonal streamflow forecasting in the western U.S. for the past 50 years, and only recently has its skill been exceeded by dynamical forecasts such as those from CBRFC--which previously relied more heavily on regression modeling itself.

[[File:<FILENAME>|frame|Figure 1. (Image source: [ ])]]

Relevance and Application

Data and Tools

Additional Resources

State of the Science Report

Chapter XX of the State of the Science report (XXXXX

New and Notable Research (2020-present)

[<URL FOR PAPER>]

Summary

@@ Line 13: / Line 13: @@
 ==Methods==
-In the snowmelt-dominated headwaters of the Colorado River Basin, predictability of streamflow on seasonal timescales in the basin arises from ''initial moisture conditions'': the observed snowpack and soil moisture conditions. The skill of the forecasts increases through the winter and spring months as the accumulating snowpack provides more information about the upcoming runoff. Soil moisture deficits or surpluses in the fall, before snow begins to accumulate, indicate whether an an unusually low or high proportion of the snowpack will become runoff in the stream.
+In the snowmelt-dominated headwaters of the Colorado River Basin, predictability of streamflow on seasonal timescales in the basin arises from ''initial moisture conditions'': the observed snowpack and soil moisture conditions. The skill of the forecasts increases through the winter and spring months as the accumulating snowpack provides more information about the upcoming runoff. Soil moisture deficits (or surpluses) in the fall, before snow begins to accumulate, indicate whether an an unusually low (or high) proportion of the snowpack will become runoff in the stream.
-Lower-elevation watersheds which have little or no snowmelt contributing to the annual hydrology cannot be forecast as skillfully as higher-elevation watersheds that are snowmelt-dominated. In all cases, the primary source of forecast uncertainty and error is due to lack of knowledge of how the weather and climate will evolve between the date the forecast is issued (e.g., April 1st) and the end of the forecast period (e.g., July 31st). Other uncertainty and error comes from inaccurate representation of the snowpack and/or soil moisture observations used as inputs into the forecast model, and biases in the model itself.
+Lower-elevation watersheds which have little or no snowmelt contributing to the annual hydrology cannot be forecast as skillfully as higher-elevation watersheds that are snowmelt-dominated. In all cases, the primary source of forecast uncertainty and error is due to lack of knowledge of how the weather and climate will evolve between the date the forecast is issued (e.g., April 1st) and the end of the forecast period (e.g., July 31st). Other uncertainty and error comes from inaccurate depictions of the snowpack and/or soil moisture as input into the forecast model, and biases in the model itself.
-The CBRFC uses a relatively simple dynamical or process-oriented model framework for its seasonal forecasts, coupling a snow model (SNOW-17) with a rainfall-runoff model (Sacramento Soil-Moisture Accounting; SAC-SMA).
+The CBRFC uses a ''conceptual'' (or simple ''dynamical'') modeling framework for its seasonal forecasts, coupling a snow model (SNOW-17) with a rainfall-runoff model (Sacramento Soil-Moisture Accounting; SAC-SMA). Conceptual models draw from our understanding of the physics of the real-world watershed, but have very simplified representations of watershed attributes and processes, with 5-20 components and tunable parameters to relate them to each other. CBRFC runs their models throughout the forecast season (December-July), using historical daily precipitation and temperatures from each year of 1981-2015 to stand in for the unknown future weather. This creates an ensemble of 35 forecast traces, each starting with the same observed initial moisture conditions but individually evolving the modeled snowpack and runoff through the end of the forecast period given the winter/spring weather of 1981, 1982,...2014, 2015. These ESP (Ensemble Streamflow Prediction) forecasts are issued daily, and then become the primary input to the official forecasts issued monthly or semi-monthly.
+NRCS uses a ''statistical'' modeling framework, employing separate regression models for each gage that are calibrated on the historical relationships between streamflow and several watershed predictors, typically snowpack (SWE) and accumulated water-year precipitation at 2-5 SNOTEL sites. Statistical regression modeling has been the standard-bearer for seasonal streamflow forecasting in the western U.S. for the past 50 years, and only recently has its skill been exceeded by dynamical forecasts such as those from CBRFC--which previously relied more heavily on regression modeling itself.