Understanding Water Supply Forecast Error on the Colorado River
The Colorado River is perhaps the most critical resource to the southwestern United States, providing water to over 40 million people. In an average year, over half of this water comes from western Colorado, primarily in the form of snowmelt when high-elevation seasonal snowpack dissipates in the spring.
The Colorado River has been managed with a large series of reservoirs. Seasonal water supply forecasts made by agencies like the Colorado Basin River Forecast Center (CBRFC) and the National Resources Conservation Service (NRCS) using data from high-elevation Snowpack Telemetry (SNOTEL) stations offer water managers insight into how much water to expect each year. These forecasts allow for less uncertainty and better management of these important reservoirs.
However, 2020 and 2021 were low water supply years, much lower than one would expect based on snowpack values alone. Researchers from the Colorado Climate Center questioned whether very low, if not record low, soil moisture levels at high elevation were causing a smaller fraction of snowmelt to runoff than in a normal year, and further, whether these conditions are likely to occur more frequently in a warmer climate. “On the Sources of Water Supply Forecast Error in Western Colorado” is the result of a research project funded by NOAA’s National Integrated Drought Information System (NIDIS) to explore this question.
The research team created hindcasts of April–August streamflows using SNOTEL snowpack and precipitation data from 1981–2021, inputting modeled soil moisture and groundwater data to predict streamflow. In this case, “hindcast” refers to a prediction of streamflow in a previous subset of years using a statistical model that was trained based on a separate subset of years. In this way, the researchers mimicked an actual water supply forecasting environment without including the known answer into the model (See the AMS article for a more detailed explanation of the methods).
The researchers paid special attention to hindcasts using the data available on April 1. April 1 is near peak snowpack season, and these numbers have historically been used as a benchmark for how much water to expect in the coming spring. In 2021, for instance, adding soil moisture data from the Western Land Data Assimilation System’s km resolution land surface model reduced April 1 streamflow hindcast in all four major basins, and lowered hindcast error across the board.
Results indicate that a much larger fraction of the error in the hindcast is attributable to the weather that occurs after April 1 and 2021 was no exception. While antecedent soil moisture conditions were record-low in western Colorado, 2021 also had a much drier-than-normal spring for western Colorado with a record dry April. This was not well predicted or incorporated in operational streamflow forecasts. April 1 numbers do not tell the whole story, because what happens before the snowpack season, and what happens from April 1 through early June, also have a marked impact on the year’s water supply.
The findings from this study are important because they establish a ceiling for how skillful we can expect water supply forecasts to be without significant, successful investments in sub-seasonal to seasonal forecasting over the Intermountain West. Soil moisture data can be implemented more effectively in the future and incrementally lower water supply forecast error for this crucial region. However, with time, larger-scale errors in sub-seasonal to seasonal prediction will need to be addressed to create a consistently accurate forecast of water supplies.
