Why do runoff forecasts diverge from snowpack conditions?

[Note: Since we’re getting closer to the April 1st, historically a key date for for water supply forecasting and decision-making, I thought I would post an explainer I wrote up in spring 2020 for the Colorado River District. Looking at how 2021 is shaping up, it’s clear that the second factor described below—low antecedent soil moisture—will be the largest one causing a difference between the SWE and forecasted runoff. In fact, 2021 is looking a lot like 2013—another year with very low antecedent soil moisture.]

While the runoff forecast for the spring-summer season is primarily based on the level of the snowpack, the forecasted runoff as a percentage of normal often differs from the snowpack as a percentage of normal. The chart below compares the April 1 snow-water equivalent (SWE) for the Colorado River basin above Cameo, CO (purple) with the NOAA CBRFC (dark blue) and NRCS (light blue) runoff forecasts for that gage for each of the past 10 years. In most years, there are differences of 5-15% between the April 1 SWE and the forecasted runoff, and up over 20% in some years. There are several factors at work here—all of which may be operating in a given year, sometimes offsetting each other, sometimes accentuating each other. Other gages in the Colorado River Basin, and in adjacent basins (South Platte, Arkansas, Rio Grande) will also see these same factors.

What’s normal? Median vs. Average
The snowpack conditions and the runoff forecasts are usually presented using different ‘normals.’  The NRCS presents both site and basin-average SWE as the percentage of the long-term (1981-2010) median, because medians are less prone to distortion by a few very high values. The CBRFC and NRCS runoff forecasts, however, are usually presented as % of the average over the same period. If we recalculated the runoff forecasts as % of median to match the SWE, this would make all of the forecasted runoff values higher by 4-10 points (proportional with the % of average), since the median seasonal runoff at Cameo is lower than the average seasonal runoff. In about half the cases, the % median forecasted runoff is closer (than the % average forecasted runoff) to the % median SWE.

Low soil moisture in the previous fall
If there was a dry summer and fall in the previous year, the soil moisture deficit going into the snow season will decrease the runoff efficiency come spring, as it effectively creates a “bucket” that needs to be filled first. Previous fall soil moisture is explicitly accounted for in the CBRFC forecast model, so low soil moisture will tip the CBRFC forecasted runoff downward, sometimes strongly, below what the snowpack alone would predict. While the NRCS statistical forecast equations don’t include soil moisture directly, they do use water-year-to-date precipitation, so a dry October that exacerbates low soil moisture will still factor into the forecasted runoff. We can see this dry-soils effect in both of the runoff forecasts for the 2020 season shown above: They are much lower than what would be expected from the snowpack alone, reflecting the very dry summer and fall of 2019. We can also see this dry-soils/dry-fall effect in the runoff forecasts for 2013, after the extreme drought in 2012.

SNOTEL averages vs. snow data as incorporated in the runoff forecasts
The basin SWE % of average reported by NRCS is based on the arithmetic average of the SWE values at all of the SNOTEL sites within and adjacent to the basin. The CBRFC runoff forecast procedure uses precipitation data from SNOTEL sites, and other climate data, to build up a more sophisticated spatial representation of the basin snowpack than a simple average of SNOTEL SWE data. The NRCS forecast equations use SWE data from selected SNOTEL sites—not all of the sites—within and near the basin above each forecast points, and apply different weightings to the SWE values at each site, rather than simply averaging them together. So due to the specific handling/estimating of snow data embedded in their respective forecast approaches, the forecasted runoff from both CBRFC and NRCS can diverge from the NRCS % of normal SWE.

Runoff efficiency as a function of snowpack size
The fraction of the snowpack that actually gets translated into runoff varies from year to year; in general, this fraction is higher in large-snowpack years than in small-snowpack years. Large snowpacks last later into the spring than do average or small snowpacks, and then melt out faster under the higher-angled sun of May and June, which (paradoxically) means that a smaller fraction of the snow’s water content gets lost to the soils and the atmosphere. So the “rich” years tend to get “richer” as large snowpacks are more efficiently converted to runoff. Both the CBRFC forecast model and the NRCS forecast equations should capture this effect to some extent. It appears that the 2011 forecast shows this effect, and it may have also been seen in the 2019 forecast but was canceled out by the dry-soils effect operating that year as well.

Gages where the forecasts are for flows that are not adjusted for upstream use
This is a very important factor for a few gages/forecast points in the Colorado River Basin, most critically Lake Powell inflows. For nearly all CBRFC forecast points, the gage record has been adjusted to account for the effects of most upstream management and water diversions, and these “naturalized” flows are what is being forecasted. But for Lake Powell inflows, CBRFC modifies their forecast of naturalized flow to fit with Reclamation’s needs for modeling inputs, which call for “unregulated flows”—which means that the effects of reservoir operations are accounted for, but not upstream diversion and consumptive use; i.e., all of the Upper Basin consumptive use. Because the amount of Upper Basin consumptive use doesn’t vary all that much from year to year, once that amount is taken away, the remaining flows—the unregulated flows—will show greater fluctuations from normal on a percentage basis than the natural flows, and by extension, than the (natural) snowpack.