On the Sources of Water Supply Forecast Error in Western Colorado

Peter E. Goble aColorado Climate Center, Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado

Search for other papers by Peter E. Goble in
Current site
Google Scholar
PubMed
Close
and
Russ S. Schumacher aColorado Climate Center, Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado

Search for other papers by Russ S. Schumacher in
Current site
Google Scholar
PubMed
Close
Open access

Abstract

Annual spring and summer runoff from western Colorado is relied upon by 40 million people, six states, and two countries. Cool season precipitation and snowpack have historically been robust predictors of seasonal runoff in western Colorado. Forecasts made with this information allow water managers to plan for the season ahead. Antecedent hydrological conditions, such as root zone soil moisture and groundwater storage, and weather conditions following peak snowpack also impact seasonal runoff. The roles of such factors were scrutinized in 2020 and 2021: seasonal runoff was much lower than expectations based on snowpack values alone. We investigate the relative importance of meteorological and hydrological conditions occurring before and after the snowpack season in predicting seasonal runoff in western Colorado. This question is critical because the most effective investment strategy for improving forecasts depends on if errors arise before or after the snowpack season. This study is conducted using observations from the Snow Telemetry Network, root zone soil moisture and groundwater data from the Western Land Data Assimilation Systems, and a random forest–based statistical forecasting framework. We find that on average, antecedent root zone soil moisture and groundwater storage values do not add significant skill to seasonal water supply forecasts in western Colorado. In contrast, using precipitation and temperature data after the time of peak snowpack improves water supply forecasts significantly. The 2020 and 2021 runoffs were hampered by dry conditions both before and after the snowpack season. Both antecedent soil moisture and spring/summer precipitation data improved water supply forecast accuracy in these years.

Significance Statement

Seasonal water supply forecasts in western Colorado are highly valuable because spring and summer runoff from this region helps support the water supply of 40 million people. Accurate forecasts improve the management of the region’s water. Heavy investments have been made in improving our ability to monitor antecedent hydrological conditions in western Colorado, such as root zone soil moisture and groundwater. However, results from this study indicate that the largest source of uncertainty in western Colorado runoff forecasts is future weather. Therefore, improved subseasonal-to-seasonal weather forecasts for western Colorado are what is most needed to improve regional water supply forecasts, and the ability to properly manage western Colorado water.

© 2023 American Meteorological Society. This published article is licensed under the terms of the default AMS reuse license. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Peter Goble, peter.goble@colostate.edu

Abstract

Annual spring and summer runoff from western Colorado is relied upon by 40 million people, six states, and two countries. Cool season precipitation and snowpack have historically been robust predictors of seasonal runoff in western Colorado. Forecasts made with this information allow water managers to plan for the season ahead. Antecedent hydrological conditions, such as root zone soil moisture and groundwater storage, and weather conditions following peak snowpack also impact seasonal runoff. The roles of such factors were scrutinized in 2020 and 2021: seasonal runoff was much lower than expectations based on snowpack values alone. We investigate the relative importance of meteorological and hydrological conditions occurring before and after the snowpack season in predicting seasonal runoff in western Colorado. This question is critical because the most effective investment strategy for improving forecasts depends on if errors arise before or after the snowpack season. This study is conducted using observations from the Snow Telemetry Network, root zone soil moisture and groundwater data from the Western Land Data Assimilation Systems, and a random forest–based statistical forecasting framework. We find that on average, antecedent root zone soil moisture and groundwater storage values do not add significant skill to seasonal water supply forecasts in western Colorado. In contrast, using precipitation and temperature data after the time of peak snowpack improves water supply forecasts significantly. The 2020 and 2021 runoffs were hampered by dry conditions both before and after the snowpack season. Both antecedent soil moisture and spring/summer precipitation data improved water supply forecast accuracy in these years.

Significance Statement

Seasonal water supply forecasts in western Colorado are highly valuable because spring and summer runoff from this region helps support the water supply of 40 million people. Accurate forecasts improve the management of the region’s water. Heavy investments have been made in improving our ability to monitor antecedent hydrological conditions in western Colorado, such as root zone soil moisture and groundwater. However, results from this study indicate that the largest source of uncertainty in western Colorado runoff forecasts is future weather. Therefore, improved subseasonal-to-seasonal weather forecasts for western Colorado are what is most needed to improve regional water supply forecasts, and the ability to properly manage western Colorado water.

© 2023 American Meteorological Society. This published article is licensed under the terms of the default AMS reuse license. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Peter Goble, peter.goble@colostate.edu

1. Motivation

Western Colorado’s hydrological cycle is characterized by a seasonal accumulation of snowpack at high elevations through the winter and early spring, followed by a snowmelt cycle in the late spring and early summer. This snowmelt replenishes water supplies in western Colorado and beyond. Six states and two countries rely on water that comes from Colorado’s annual snowmelt (USBR 2013).

Colorado has seen a shift in summertime weather conditions in recent years with much hotter, and often drier, conditions persisting into late summer and early autumn (NCEI 2022). The July–September periods of 2018–20 produced the three lowest recorded 3-month standardized precipitation–evapotranspiration index measurements for Colorado Climate Region II, or western Colorado (Fig. 1). The recent warming trend for Colorado, and for the entire western United States, is expected to continue with high confidence (Gonzalez et al. 2018).

Fig. 1.
Fig. 1.

Time series of the standardized precipitation evapotranspiration index (SPEI) for Colorado Climate Division II (1895–2021). SPEI is calculated using the average monthly temperature and precipitation over the climate division in the NOAA nClimGrid dataset (Adams 2017), a Thornthwaite potential evapotranspiration estimate, and a Pearson distribution trained on data from 1901 to 2020.

Citation: Journal of Hydrometeorology 24, 12; 10.1175/JHM-D-23-0004.1

Higher temperatures, particularly during summer, are likely to reduce usable water volumes in an already arid region. While neither observations nor climate models show a clear downward trend in precipitation for Colorado (Lukas et al. 2014; Lukas and Payton 2020; NCEI 2022), higher temperatures raise evapotranspiration rates and change the fraction of water available for runoff (Christensen and Lettenmaier 2007; Talsma et al. 2022). Udall and Overpeck (2017) show that while the first 15 years of the twenty-first century were not the driest recorded 15-yr stretch over the Upper Colorado River basin, they did result in historically low streamflows due to higher temperatures. Xiao et al. (2018) estimate that over half of the decreasing trend in annual Colorado River flows is attributable to an increase in temperatures rather than a decrease in precipitation. This trend is expected to continue with projected annual streamflow losses of 3%–10% °C−1 warming over preindustrial levels (Hoerling et al. 2019; Milly and Dunne 2020).

Hotter, drier, and longer summers increase the likelihood of low soil moisture and low water tables entering the winter snowpack season (e.g., Seager et al. 2013; Talsma et al. 2022). This reduces both the fraction of precipitation that becomes runoff, or runoff efficiency (Koster et al. 2010), and peak season streamflow predictability (Livneh and Badger 2020). Runoff efficiency is reduced because a higher fraction of cold season snowpack is partitioned to soil moisture recharge. Runoff in the following season becomes more uncertain because peak season snowpack is a less skillful measure of subsequent season runoff as the climate warms, the snow season shortens, and the fraction of snowpack directly translating into runoff decreases.

Annual runoff forecasts have been made historically using statistical relationships between peak spring snowpack values, generally occurring around 1 April, and subsequent peak streamflow values (Church 1935; Fleming and Goodbody 2019). Such approaches do not account for changes in antecedent soil conditions and are less skillful if the fraction of seasonal snowmelt partitioned to soil moisture and groundwater recharge increases. Current operational water supply forecasting methods are more nuanced, and some approaches do use soil moisture conditions. Notably, the Colorado Basin River Forecast Center (CBRFC) uses the Sacramento Soil Moisture Accounting Model to run dynamical streamflow forecasts for the runoff season (Vrugt et al. 2006; Moser 2021). The Natural Resources Conservation Service (NRCS) produces water supply forecasts for western Colorado as well. These forecasts indirectly account for antecedent soil moisture conditions through the inclusion of precipitation data from 1 October onward, or water-year-to-date precipitation (Fleming and Goodbody 2019).

The CBRFC and NRCS make a water supply forecast on 1 January and update these forecasts monthly through the remainder of the snowpack season and runoff season (Musselman et al. 2019). While every monthly update attracts attention from water managers, perhaps the most important is the one released on 1 April. The 1 April water supply forecast updates from CBRFC and NRCS attract additional interest for a couple of reasons: 1 April forecasts are issued nearest the time of peak snowpack conditions, and it is the time of year when streamflows begin to rise in response to melting mountain snow.

Annual runoff in western Colorado was much lower than normal in 2020 and 2021 (USDA NRCS 2022a). Both these years featured near-normal snowpack measurements at the beginning of April but well below normal runoff (USDA NRCS 2022a). The CBRFC and NRCS correctly anticipated below normal runoff years across western Colorado in both 2020 and 2021, but April–July runoff values were still lower than predicted on 1 April. Figure 2 shows the difference between 1 April snowpack, NRCS-forecasted April–July runoff, and observed April–July runoff in 2020. Antecedent soil moisture conditions are an oft-cited metric used to explain lack of runoff (Sackett 2021; Sakas 2021; Gilbert 2022), but they are not the only physically plausible explanation. Runoff conditions are also impacted by meteorological conditions occurring after the timing of peak snowpack as measured by Snow Telemetry (SNOTEL; USDA NRCS 2021) Network stations. Dry and warm spring conditions after observed peak snowpack also negatively impact runoff. Spring conditions were especially dry in 2021 where western Colorado experienced its driest April on record (NCEI 2022).

Fig. 2.
Fig. 2.

(a) The 2020 percent of 1991–2020 normal 1 Apr snowpack, (b) forecasted April–July runoff, and (c) observed April–July runoff measured at basin scale (Hydrological Unit Code 6). Black outlines are HUC 6 boundaries and blue lines are HUC 4 boundaries. Source: USDA NRCS (2022b).

Citation: Journal of Hydrometeorology 24, 12; 10.1175/JHM-D-23-0004.1

What influences seasonal runoff more: antecedent soil moisture and groundwater conditions or meteorological conditions following 1 April? Answering this question is paramount because improving forecast skill, or even maintaining it in a changing climate, requires investment in different disciplines. Much has been invested in numerical soil moisture models over the last decade (e.g., Houborg et al. 2012; Koren et al. 2014; Xia et al. 2014; Erlingis et al. 2021). If a new soil moisture modeling product can be added to a statistical forecasting framework, such as Fleming and Goodbody (2019), then the best way to improve and maintain forecast skill may be to implement, and improve, such models, or to invest in collecting more soil moisture observations. If it can be shown that most of the uncertainty in statistical water supply forecasting is attributable to what occurs after 1 April, then improving or maintaining skill will hinge upon improved subseasonal-to-seasonal (S2S) weather forecasts.

In this manuscript, we examine the role of pre- and post-snowpack accumulation season conditions in generating streamflow forecast error in western Colorado. To assess the importance of antecedent soil moisture, we combine a soil moisture and groundwater model with a random forest–based statistical streamflow forecast. To assess the importance of weather after 1 April, we evaluate how much forecasts improve if spring precipitation and temperature observations could be forecasted without error.

2. Methods

a. Data used

In this study, we produced experimental water supply hindcasts for western Colorado using observed water-year-to-date precipitation and snow water equivalent data from SNOTEL (USDA NRCS 2022a) for years 1981–2021. Antecedent subsurface moisture conditions were estimated using 1 January gridded Western Land Data Assimilation Systems (Erlingis et al. 2021) top 2-m soil volumetric water content and groundwater data over the same time frame. 1 January was the date used for antecedent hydrological conditions for three reasons: 1) This is the release date for the first water supply forecast of the season for both CBRFC and NRCS. 2) High elevation soil moisture conditions do not change much during the winter because the ground is frozen and new precipitation accumulates as snowpack. 3) Soil moisture changes quickly in response to snowmelt come springtime, after which point, it is more of a proxy for snowmelt timing. High soil moisture values in March or early April can arguably be considered a drought indicator since they are invariably indicative of early snowmelt, not a wet autumn or winter.

Hindcasts were made for April–August accumulated runoff. Typically, water supply forecasters use April–July streamflow as the predictand, but in wet years, the runoff season has not ended by 1 August in western Colorado. The predictand was, specifically, April–August naturalized streamflows for stream gages in four key watersheds: the Colorado River Headwaters, Gunnison River basin, Yampa River basin, and San Juan River basin. The specific stream gages used are shown in Fig. 3, along with the locations of all SNOTEL sites used.

Fig. 3.
Fig. 3.

Locations of United States Geological Survey stream gages (blue circles) and SNOTEL snowpack/precipitation observations (black circles) used. The yellow-shaded area represents Colorado west of the Continental Divide, blue lines are rivers, and black lines are watershed boundaries.

Citation: Journal of Hydrometeorology 24, 12; 10.1175/JHM-D-23-0004.1

On average, 13.5 million acre feet of water flow into Lake Mead annually from the Colorado River basin (USBR 2022). Much of this comes from western Colorado. The Colorado River at Cameo, Gunnison River at Grand Junction, Yampa River at Maybell, and San Juan River at Bluff in Utah combine for an average of 6.81 million acre feet of naturalized streamflow between the months of April and August every year. Naturalized streamflow refers to the amount of streamflow that would have occurred if not for human diversions. It will be referred to simply as streamflow, or flow, hereinafter. The Colorado River Headwaters is the most productive of these tributaries, accounting for an average of 2.75 million acre feet of streamflow on its own. Flows vary widely from year to year, largely as a function of precipitation and snowpack. Each basin experienced years below 50% of normal flows and above 150% of normal flows between 1981 and 2021.

Separate hindcasts are produced for each watershed. The volumetric water content and groundwater values used were derived by averaging all upstream grid points in the basin at elevations above 2400 m. Using both volumetric water content and groundwater data attempts to capture both near-surface and deeper subsurface influences on streamflow.

Hindcasts were designed to mimic the operational framework of a water supply forecaster, such as NRCS or CBRFC, in western Colorado. Hindcasts were made using only data that would have been available to a forecaster at the beginning of each month. This was done for each month starting in the middle of the snowpack accumulation season through the end of peak runoff season: 1 January, 1 February, 1 March, 1 April, 1 May, 1 June, 1 July, and 1 August. For example, a 1 February hindcast for April–August naturalized flows in the Yampa River basin would only utilize snowpack, water-year-to-date temperature, water-year-to-date precipitation through January, and antecedent volumetric water content and groundwater information, not any information from February onward.

Six sets of hindcasts were produced: The base hindcasts, referred to hereinafter as “Base,” use only observed snowpack and water-year-to-date precipitation from SNOTEL sites available in the watershed boundary above the relevant stream gage, water-year-to-date temperatures above 2400 m from the Parameter-elevation Regressions on Independent Slopes Model (PRISM) (Daly et al. 2008), and streamflow data from 1 April through the current month (e.g., April–May streamflow would be used in a 1 June hindcast). SNOTEL temperature observations were not used due to documented issues with temperature sensor drift in the network (Oyler et al. 2015). The second set of hindcasts was produced with 1 January volumetric water content data included. The third set used both 1 January volumetric water content and groundwater information. Finally, another three sets of hindcasts were generated using the same information as the first three, but with the addition of precipitation data from the current month (August) and temperature data from March–May (snowmelt season). These data will be referred to as “foresight of observations.” The hindcasts including foresight of observations data did not incorporate any observed future streamflow information. Input variables for each hindcast set are shown in Table 1. Streamflow hindcasts are verified against NRCS-adjusted (naturalized) flow estimates in western Colorado. The methodology for computing these flows can be found in USDA NRCS (2022a).

Table 1.

Input variables used (column) in each hindcast set (rows). “Y” indicates data were used; “N” indicates data were not used. Hindcast sets are named according to which variables are included. All hindcasts include water-year-to-date precipitation and snow water equivalent. VWC = 2-m soil volumetric water content. GW = groundwater storage. FO = foresight of observations.

Table 1.

b. Data quality control

Western Land Data Assimilation Systems (WLDAS) volumetric water content and groundwater datasets and the PRISM temperature dataset were spatially and temporally complete. SNOTEL snow water equivalent and precipitation datasets included missing data. To fill missing data, a standard anomaly was computed for every nonmissing station, month, and year. Missing SNOTEL snowpack and water-year-to-date precipitation values were estimated using Eq. (1). The term Pi,j,m is a missing water-year-to-date precipitation or snowpack value. The i dimension represents years, the j dimension represents stations in a basin with nonmissing data for the month and year, and the m dimension represents the month of the year:
Pi,j,m=mean(P:,j,m)×mean(Pi,:,m)/mean(P:,:,m).

c. Statistical procedures

Hindcasts were made using a random forest model. A forest of 100 regression trees was built for each basin from all input variables. In each case, a randomly selected 26 out of the 41 years of data are used as a training dataset and the 15 out-of-sample years are used as a test dataset. Hindcasts were made from the random forest model for the 15 test years. Only tracking results for test years ensures hindcasts can be interpreted as if they were actual predictions rather than just correlation analyses where the solution is implicit in the computation. Trees were not limited in number of splits or maximum depth. All 100 trees were weighted equally when hindcasting a value for test years. Additionally, a multiple linear regression value is computed from the 26 training years and used to predict values for test years. Multiple linear regression is used in place of the random forest when it is more accurate for test years. Multiple linear regression is most useful when one predictor explains nearly all the variance in the test data, and even decision trees splitting continuously along the same predictor produce a relationship with unnecessary stepwise jumps (e.g., for hindcasts made on 1 August: 1 April–31 July streamflow is a nearly perfect predictor of 1 April–31 August streamflow).

The process above is repeated 100 times. All hindcast predictors (e.g., water-year-to-date precipitation and temperature, snowpack, soil volumetric water content, and groundwater) and predictands (streamflow) were standardized before initializing the random forest model using Eq. (2). The term s(x) represents the standardized anomaly of observation x, μx is the mean of dataset x, and σx is the standard deviation of dataset x, where x could be any value used in the random forest model (e.g., basin-averaged October–March precipitation).
s(x)=(xμx)σx.
The average fraction of variance explained (r2), mean absolute error, and Heidke skill score (Heidke 1926) are then computed for each basin (four basins), hindcast issue month (8 months), and input data test set (six configurations) over 100 randomly selected combinations of test and training years, totaling 19 200 statistical hindcasts. The formulas used to compute r2, mean absolute error, and Heidke skill score are given in Eqs. (3)(5). In these equations, NF represents the naturalized streamflow from NRCS and NFfit represents the predicted streamflow values for each year derived from training data. The NF and NFfit values have been standardized using Eq. (2) before being input into Eq. (4). In the Heidke skill score formula [Eq. (5)], A = fraction of successful hits, B = fraction of false hits, C = fraction of false rejections, D = fraction of correct rejections, and E = expected value of A + D given no skill. In this study, a skillful hindcast was defined as one in which the tercile value of April–August runoff is correctly predicted (CT). For example, a 40th percentile NFfit hindcast would be considered correct if 0.33 < percentile (NF) < 0.67. In this case, E = 1/3. The data used in each set of hindcasts are shown in Table 1.
r2=[NFfitimean(NFfit)]×[NFimean(NF)][NFfitimean(NFfit)]2×[NFimean(NF)]22,
MAE=i=1n|NFiNFfiti|n,
HSS=(A+DE)(A+B+C+DE)=CTE1E.

d. Significance testing

Two significance tests were conducted: one to generally assess the skill of hindcasts and another to evaluate which variables created significantly more accurate hindcasts than Base (using only water-year-to-date precipitation, water-year-to-date temperature, and snow water equivalent data as inputs). The statistical significance of hindcast skill scores was evaluated by comparing the Heidke skill scores in the actual hindcasts with scores derived from 10 000 hindcasts, replacing all the actual input variables with data randomly drawn from standard normal distributions. This was done with multiple linear regression because the full random forest model was too computationally expensive for such a high number of iterations. Hindcasts with higher skill than 99% of the hindcasts created with random data were considered statistically significant at 99% confidence. Just as with the actual hindcasts, each of the 10 000 skill scores was computed from the average of 100 iterations where 26 years of data were randomly selected for training, and the other 15 years were selected for testing.

A similar methodology was used to determine if hindcast variable sets (e.g., foresight of observations + volumetric water content) added significant skill to Base: Nonbase hindcast variables were replaced with randomly selected data from a standard normal distribution. As with the first test, this process was repeated 10 000 times to bootstrap a distribution of skill differences. When nonbase variables added more skill than at least 99% of the model runs using random data as replacements, they were considered significant at 99% confidence.

e. 2020–21

Finally, hindcasts are created for 2020 and 2021 specifically. This analysis was conducted to see if recent years are outliers in terms of runoff efficiency. In this case, just one hindcast was made for each basin for each year using multiple linear regression, and all years 1981–2019 as training data. This process was repeated for the following sets of the input variables in Table 1: Base, volumetric water content + groundwater, and foresight of observations + volumetric water content + groundwater. These hindcasts were made using only data from October to March, mimicking the operational setting of a 1 April forecast in both years. Foresight of observations + volumetric water content + groundwater hindcasts added only April–June precipitation and temperature to the hindcast as “foresight of observations.” 1 April was chosen because these forecasts carry special importance to water managers. 1 April is near peak snowpack season, high-elevation snowmelt has already begun by this point in some years, and water managers use these forecasts to create a plan for the coming months.

3. Results

a. Hindcasts results

April–August streamflow predictions made using only Base inputs demonstrate significant skill. SNOTEL-based high-elevation snow water equivalent and water-year-to-date precipitation are crucial in current operational water supply forecast models (Vrugt et al. 2006; Fleming and Goodbody 2019). These hindcasts explain 36%–50% of the variance in April–August cumulative streamflow when made on 1 January. Skill increases by month throughout the winter, spring, and summer seasons to 98+% of variance explained on 1 August (Figs. 46). Skill scores for every basin at every initialization period were significantly higher than zero at greater than 99% confidence. 1 April forecasts, which are of particular importance to water managers (Musselman et al. 2019), explained 63%–69% of variance in April–August streamflow.

Fig. 4.
Fig. 4.

Fraction of variance in April–August streamflow explained by statistical hindcasts (1981–2021) with no future precipitation and temperature data (black), and precipitation + spring temperature data for the remainder of the water year (blue). Gray shaded area represents improvements in streamflow hindcasts to be gained from foresight of observations (FO). The blue shaded area represents error from other sources.

Citation: Journal of Hydrometeorology 24, 12; 10.1175/JHM-D-23-0004.1

Fig. 5.
Fig. 5.

Mean absolute error in April–August streamflow, represented in standard deviations, explained by statistical hindcasts (1981–2021) with no future precipitation and temperature data (black), and precipitation + spring temperature data for the remainder of the water year (blue). Gray shaded area represents improvements in streamflow hindcasts to be gained from foresight of observations (FO). The blue shaded area represents error from other sources.

Citation: Journal of Hydrometeorology 24, 12; 10.1175/JHM-D-23-0004.1

Fig. 6.
Fig. 6.

Heidke skill score for April–August streamflow explained by statistical hindcasts (1981–2021) with no future precipitation and temperature data (black), and precipitation + spring temperature data for the remainder of the water year (blue). Gray shaded area represents improvements in streamflow hindcasts to be gained from foresight of observations (FO). The blue shaded area represents error from other sources.

Citation: Journal of Hydrometeorology 24, 12; 10.1175/JHM-D-23-0004.1

Adding volumetric water content and groundwater information from WLDAS did not add statistically significant skill to seasonal streamflow forecasts for any of the four watersheds analyzed. Changes in forecast skill were near zero (Table 2). This result was surprising given the role of surface hydrology variables in current physical operational models (Vrugt et al. 2006).

Table 2.

Heidke skill scores of VWC, GW, and FO sets minus Heidke skill score of Base (Table 1) for 1 Apr hindcasts. Hindcast sets with skill significantly higher than Base at 99% confidence are in bold. VWC = volumetric water content. GW = groundwater. FO = foresight of observations.

Table 2.

Foresight of observations improved 1 January–1 May hindcast Heidke skill scores significantly for all basins at 99% confidence and improved 1 June hindcasts significantly at 99% confidence for all basins except the Gunnison. This improvement was still significant at 95% confidence for all basins. Improvements to 1 July and 1 August forecasts were not significant at 99% confidence as the majority of annual runoff has occurred by this point, and summertime rainfall does not have as large an influence on flows as seasonal snowmelt.

Figures 46 show the error in hindcasts by basin in comparison to error from other sources. Foresight of observations not only improved hindcasts significantly but also boosted all skill metrics studied here by the following amounts: fraction of variance explained increased between 0.37 and 0.43 in January and between 0.13 and 0.21 in April. Mean absolute error is reduced by 0.33 standard deviations for all basins in January and between 0.08 and 0.22 standard deviations in April. Heidke skill scores improved between 0.39 and 0.51 in January and between 0.12 and 0.24 in April.

Even with foresight of observations, water supply hindcasts are not perfect. The remaining error is represented in blue in Figs. 46 as “other errors.” These errors have several implications: first, antecedent soil moisture and groundwater conditions may have the potential to improve a statistical water supply forecast if measured ideally and integrated in the forecast ideally. It is possible that a different antecedent soil moisture dataset or in situ observations could close some of this skill gap, but this should not be the default expectation. Other errors could come from many plausible sources: for instance, all predictors (precipitation, temperature, and snowpack observations, and modeled volumetric water content and groundwater) are subject to measurement issues. The predictand (naturalized streamflow) is subject to both measurement errors and errors resulting from the conversion of measured streamflow to naturalized streamflow. Furthermore, the relevant catchment area for each stream gage is not uniformly sampled, and sometimes subject to systematic biases. For instance, SNOTEL does not evenly sample high elevations in the forecasted basins. It is estimated that 37% of seasonal runoff in western Colorado comes from above 3350 m elevation. Only 5% of SNOTEL sites are located above this elevation (CBRFC 2021). The other errors may be reduced simply by adding snowpack and precipitation observations higher in elevation, or in undersampled productive catchments.

b. 2020–21

The 2020 and 2021 runoff generated public concern about the influence of soil moisture on streamflow, especially in a warming climate (e.g., CBRFC 2021; Chow 2020; Runyon 2020; Lukas 2021). Both 2020 and 2021 produced less runoff in western Colorado than one would expect based on precipitation totals alone (Tables 3 and 4). The 1 April water-year-to-date precipitation was 97% of normal, and snowpack was 108% of normal in the Colorado Headwaters, but the Colorado River at Cameo only recorded 70% of normal April–August naturalized flows. Even 1 June water-year-to-date precipitation was 90% of normal. Streamflow hindcasts for the Colorado River at Cameo ranged from 79% of normal (foresight of observations + volumetric water content + groundwater) to 94% of normal (Base) (Table 3). The NRCS forecasted 95% of normal April–July runoff for the Colorado River at Cameo on 1 April 2020. Table 3 provides these numbers for all four basins considered in this study.

Table 3.

Hindcasted percent of 1981–2019 average streamflow from hindcast sets Base, VWC + GW, and FO + VWC + GW for 2020 compared to NRCS 1 Apr forecast and observed flow. An asterisk indicates the forecast/observation was for April–July, not April–August. Observed flows appear in bold. VWC = volumetric water content. GW = groundwater. FO = foresight of observations.

Table 3.
Table 4.

Hindcasted percent of 1981–2019 average streamflow from hindcast sets Base, VWC + GW, and FO + VWC + GW for 2021 compared to 1 Apr forecast and observed flow. An asterisk indicates forecast/observation was for April–July, not April–August. Observed flows appear in bold. VWC = volumetric water content. GW = groundwater. FO = foresight of observations.

Table 4.

Volumetric water content anomalies were negative: −1.06, −1.67, −0.55, and −2.06 for the Colorado, Gunnison, Yampa, and San Juan basins, respectively, in 2020. Adding soil moisture and groundwater information improved statistical hindcasts for 2020 for three of the four basins (volumetric water content + groundwater) and did not reduce skill for the Gunnison basin. A larger improvement came from adding April–June foresight of observations. Even when including volumetric water content and groundwater, the 1 April hindcasted streamflows were 134%, 160%, 120%, and 133% of observed flow for the Colorado River at Cameo, Gunnison at Grand Junction, Yampa at Maybell, and San Juan at Bluff, respectively. These numbers improved to 108%, 121%, 102%, and 98% of observed when adding April–June foresight of observations. Using additional spring meteorological and hydrological input variables in the hindcast, such as potential or actual evapotranspiration, may have made these hindcasts even more accurate. Studies such as Mankin et al. (2021) include more detail about the role evapotranspiration anomalies played in 2020 western United States drought conditions.

The 2021 streamflow observations were also below forecasted values, although forecasts were more accurate in 2021 than 2020 for all basins except the Yampa River basin. In this study, soil moisture and groundwater information did improve 2021 hindcasts in all four basins (Table 4). The 1 January volumetric water content anomalies in 2021 were 2.15–2.55 standard deviations below normal for the four basins, and the lowest on record for each. Furthermore, all hindcasts did not error high. Hindcasted flows using volumetric water content and groundwater ranged from 85% to 160% of observed values and included less error than those made with no volumetric water content and groundwater information (Table 4). Because volumetric water content and groundwater data did not generally improve hindcast skill in years 1981–2021 but did improve hindcast skill in 2021, it remains undetermined whether 2021 is an outlier (McEvoy and Hatchett 2023) or exemplifies a paradigm shift brought on by a warmer climate wherein significantly lower than historically normal soil moisture and groundwater conditions raise the amount of snowpack needed to generate normal streamflows. Results do indicate volumetric water content and groundwater values can be useful in streamflow forecasts in years with large negative antecedent soil moisture anomalies.

Conditions following 1 April were also exceptional in western Colorado. April 2021 was the driest April on record for Colorado west of the Continental Divide (NCEI 2022). Adding April–June foresight of observations improved water supply hindcasts for three of four basins and did not change the hindcast in the San Juan basin (Table 4).

4. Discussion

Soil moisture conditions prior to snowpack season in western Colorado impact the physics of runoff and therefore have some influence on the region’s water supply. Even so, WLDAS soil moisture and groundwater data did not consistently or significantly add skill to seasonal water supply to the statistical hindcasts for river basins in western Colorado evaluated herein. This analysis finds that a much larger and significant fraction of the uncertainty in streamflow forecasts made on or before 1 April is future precipitation and temperatures. Statistical hindcasts trained using foresight of observations were significantly more skillful than those without for all four basins analyzed, explaining over 80% of the variance in April–August streamflows in the Colorado Headwaters, Gunnison basin, Yampa basin, and San Juan basin. Thus, water supply forecasting in western Colorado is one of many problems that emphasizes the potential value to be gained from improved subseasonal-to-seasonal forecasts (e.g., White et al. 2017; Vitart and Robertson 2018; Sengupta et al. 2022).

The influence of spring weather on seasonal runoff is likely even greater than analyses here have shown. Hindcasts made with foresight of observations were significantly more skillful than those made without, but inclusion of other weather variables may lead to even more accurate results. Solar radiation, wind speed, and dust on snow anomalies all impact the runoff season (Schmidt 1982; Adam et al. 2009; Skiles et al. 2015), and none were included here.

Seasonal water supply hindcasts made with WLDAS soil moisture and groundwater data were not significantly more skillful than those made without. This null result is not without importance. A pervasive assertion exists that low antecedent soil moisture levels cause low streamflow values the following season (Sackett 2021; Sakas 2021; Gilbert 2022). This is a physically grounded hypothesis, but more evidence is needed to determine whether the effect of antecedent soil moisture and groundwater conditions on the coming season’s runoff is small or large. Given that forecasts made with only water year precipitation and temperature through August, and peak season snow water equivalent data already explain over 80% of the variance in April–August streamflow, attempts to improve water supply forecasts by integrating more observed or derived hydrological data have limited potential to improve water supply forecasts in western Colorado.

Hydrological datasets capable of significantly improving statistical water supply forecasts in western Colorado may exist. If not, such datasets may exist in the future. The NRCS is undergoing a large effort to install and maintain more SNOTEL soil moisture observations for the purpose of improving water supply forecasts. This has been done with some success in Utah (Harpold et al. 2017). SNOTEL soil moisture observations are still sparse in western Colorado and have periods of record going back to 2003 at earliest. With time, these observations may improve forecasts. The dynamics of Colorado’s water cycle will continue to change as the climate warms (Siirila-Woodburn et al. 2021). For this reason, soil moisture data may be a more important water supply forecast input in western Colorado in the future.

Perfect soil moisture observations and perfect seasonal forecasts would not eliminate water supply forecast errors in western Colorado. Numerous other sources of error exist. Potential sources of error include, but are not limited to, the following: lack of precipitation and snowpack observations in key runoff production areas, particularly at the highest elevations (CBRFC 2021), a sampling bias toward clear, unsloped terrain (Meromy et al. 2013; Aggett 2022), measurement uncertainties in SNOTEL and United States Geological Survey stream gage networks, uncertainties in NRCS naturalized flow calculations, and limitations in statistical forecasting capabilities.

Forecasting design may also serve as a source of error. Hindcast input data (e.g., precipitation, snow water equivalent, volumetric water content) are retrieved from disparate data sources, and not part of the same unified model framework. The CBRFC has historically produced similar statistical water forecasts along with their ensemble streamflow prediction model, which does consider all components of the hydrological cycle (e.g., precipitation, snow water equivalent, volumetric water content) in a unified model framework. The ensemble streamflow prediction forecasts have shown greater accuracy at most stream gage sites than NRCS statistical streamflow forecasts in some recent years (Lukas and Peyton 2020).

The narrative that western Colorado streamflows in 2020 and 2021 were reduced by antecedent soil conditions (Sackett 2021; Sakas 2021; Gilbert 2022) is incomplete. The 1 April hindcasts made with soil moisture and groundwater information were marginally better than those made without, but still higher than observed values for all four major river basins in 2020. The evidence for dry antecedent hydrological conditions weakening spring runoff is more compelling in 2021. Root zone soil moisture on 1 January 2021 was more than two standard deviations below normal for all four river basins analyzed. Volumetric water content + groundwater + foresight of observations hindcasts in 2021 did not have a positive bias and were below observations for the Yampa and San Juan River basins, but dry spring conditions after climatological peak snowpack season are also important in explaining the magnitude of negative streamflow anomalies in 2020 and 2021. April 2021 was the driest April on record for Colorado west of the Continental Divide (NCEI 2022; McEvoy and Hatchett 2023). Including these negative precipitation anomalies improved hindcasts substantially for three of four basins.

A population of over 40 million humans relies on seasonal runoff from western Colorado in some capacity (USBR 2013). This population is growing even as water resources become scarcer across the southwest United States in a warming climate (Milly and Dunne 2020). The need for accurate water supply forecasts is critical. This study demonstrates that existing soil moisture and groundwater models are unlikely to provide “low-hanging fruit” for improving forecasts. Instead, western Colorado water supply forecast skill improvements hinge notably on uncertain returns from subseasonal-to-seasonal weather forecasting research.

Acknowledgments.

This research was funded by the National Integrated Drought Information System (NIDIS) through the Cooperative Institute of Research in the Atmosphere (CIRA) Award Number NA19OAR4320073 and USDA National Institute of Food and Agriculture and Colorado Agricultural Experiment Station Project COL00703C. We would also like to acknowledge Jeff Lukas of lukasclimate.com for helpful counsel in the direction of this research.

Data availability statement.

All precipitation, snow water equivalent, and observed adjusted streamflow data may be freely accessed through the Natural Resources Conservation Services report generator (USDA NRCS 2022a) without registration (https://wcc.sc.egov.usda.gov/reportGenerator/). Soil moisture and groundwater data were downloaded from the Western Land Data Assimilation Systems (WLDAS) archive available through the National Aeronautics and Space Administration (NASA) Earthdata system. One must register an account to access these data at https://www.earthdata.nasa.gov/.

REFERENCES

  • Adam, J. C., A. F. Hamlet, and D. P. Lettenmaier, 2009: Implications of global climate change for snowmelt hydrology in the twenty-first century. Hydrol. Processes, 23, 962972, https://doi.org/10.1002/hyp.7201.

    • Search Google Scholar
    • Export Citation
  • Adams, J., 2017: climate_indices, an open source Python library providing reference implementations of commonly used climate indices. Accessed January 2023, https://github.com/monocongo/climate_indices.

  • Aggett, G., 2022: Colorado airborne snow observatory expansion plan: Project report. 73 pp., https://coloradosnow.org/.

  • CBRFC, 2021: CBRFC water year in review. Accessed October 2022, https://www.cbrfc.noaa.gov/present/yir/yir_2020.pdf.

  • Chow, D., 2020: Colorado—the entire state—is in the midst of a widespread drought. National Broadcasting Company News, accessed 26 May 2023, https://www.nbcnews.com/science/environment/colorado-entire-state-midst-widespread-drought-n1236333.

  • Christensen, N. S., and D. P. Lettenmaier, 2007: A multimodel ensemble approach to assessment of climate change impacts on the hydrology and water resources of the Colorado River Basin. Hydrol. Earth Syst. Sci., 11, 14171434, https://doi.org/10.5194/hess-11-1417-2007.

    • Search Google Scholar
    • Export Citation
  • Church, J. E., 1935: Principles of snow surveying as applied to forecasting stream flow. J. Agric. Res., 51, 97130.

  • Daly, C., M. Halbleib, J. I. Smith, W. P. Gibson, M. K. Doggett, G. H. Taylor, J. Curtis, and P. P. Pasteris, 2008: Physiographically sensitive mapping of climatological temperature and precipitation across the conterminous United States. Int. J. Climatol., 28, 20312064, https://doi.org/10.1002/joc.1688.

    • Search Google Scholar
    • Export Citation
  • Erlingis, J. M., and Coauthors, 2021: A high‐resolution land data assimilation system optimized for the western United States. J. Amer. Water Resour. Assoc., 57, 692710, https://doi.org/10.1111/1752-1688.12910.

    • Search Google Scholar
    • Export Citation
  • Fleming, S. W., and Goodbody, A. G., 2019: A machine learning metasystem for robust probabilistic nonlinear regression‐based forecasting of seasonal water availability in the US west. IEEE Access, 7, 11 994311 9964, https://doi.org/10.1109/ACCESS.2019.2936989.

    • Search Google Scholar
    • Export Citation
  • Gilbert, D., 2022: As Colorado warms, dry soil sucks up more water. That’s bad news for rivers and farmers. Colorado Sun, accessed 28 September 2022, https://coloradosun.com/2022/02/06/colorado-soil-dry-climate-change/.

  • Gonzalez, P., and Coauthors, 2018: Southwest. Impacts, Risks, and Adaptation in the United States: Fourth National Climate Assessment, Volume II, D. R. Reidmiller et al., Eds., U.S. Global Change Research Program, 1101–1184, https://doi.org/10.7930/NCA4.2018.CH25.

  • Harpold, A. A., K. Sutcliffe, J. Clayton, A. Goodbody, and S. Vazquez, 2017: Does including soil moisture observations improve operational streamflow forecasts in snow-dominated watersheds? J. Amer. Water Resour. Assoc., 53, 179196, https://doi.org/10.1111/1752-1688.12490.

    • Search Google Scholar
    • Export Citation
  • Heidke, P., 1926: Berechnung des Erfolges und der Gute der Windstarkevorhersagen im Sturmwarnungsdienst (Measures of success and goodness of wind force forecasts by the gale-warning service). Geogr. Ann., 8, 301349, https://doi.org/10.1080/20014422.1926.11881138.

    • Search Google Scholar
    • Export Citation
  • Hoerling, M., J. Barsugli, B. Livneh, J. Eischeid, X. Quan, and A. Badger, 2019: Causes for the century-long decline in Colorado River flow. J. Climate, 32, 81818203, https://doi.org/10.1175/JCLI-D-19-0207.1.

    • Search Google Scholar
    • Export Citation
  • Houborg, R., M. Rodell, B. Li, R. Reichle, and B. Zaitchik, 2012: Drought indicators based on model-assimilated Gravity Recovery and Climate Experiment (GRACE) terrestrial water storage observations. Water Resour. Res., 48, W07525, https://doi.org/10.1029/2011WR011291.

    • Search Google Scholar
    • Export Citation
  • Koren, V., M. Smith, and Z. Cui, 2014: Physically-based modifications to the Sacramento Soil Moisture Accounting model. Part A: Modeling the effects of frozen ground on the runoff generation process. J. Hydrol., 519, 34753491, https://doi.org/10.1016/j.jhydrol.2014.03.004.

    • Search Google Scholar
    • Export Citation
  • Koster, R., S. P. P. Mahanama, B. Livneh, D. P. Lettenmaier, and R. H. Reichle, 2010: Skill in streamflow forecasts derived from large-scale estimates of soil moisture and snow. Nat. Geosci., 3, 613616, https://doi.org/10.1038/ngeo944.

    • Search Google Scholar
    • Export Citation
  • Livneh, B., and A. M. Badger, 2020: Drought less predictable under declining future snowpack. Nat. Climate Change, 10, 452458, https://doi.org/10.1038/s41558-020-0754-8.

    • Search Google Scholar
    • Export Citation
  • Lukas, J., 2021: Why do runoff forecasts diverge from snowpack conditions? Accessed 25 October 2022, https://lukasclimate.com/blog/2021/2/25/why-isnt-the-seasonal-runoff-forecast-as-a-of-normal-equal-to-the-current-snowpack-of-normal.

  • Lukas, J., and E. Payton, Eds., 2020: Colorado River Basin climate and hydrology: State of the science. Western Water Assessment, University of Colorado Boulder Tech. Rep., 119 pp., https://doi.org/10.25810/3hcv-w477.

  • Lukas, J., J. Barsugli, N. Doesken, I. Rangwala, and K. Wolter, 2014: Climate change in Colorado: A synthesis to support water resources management and adaptation. A report for the Colorado Water Conservation Board, Western Water Assessment, 2nd ed., 114 pp., https://wwa.colorado.edu/sites/default/files/2021-07/Climate_Change_CO_Report_2014_FINAL.pdf.

  • Mankin, J. S., I. Simpson, A. Hoell, R. Fu, J. Lisonbee, A. Sheffield, and D. Barrie, 2021: NOAA drought task force report on the 2020–2021 southwestern U.S. drought. NOAA Drought Task Force, MAPP, and NIDIS, 20 pp., https://www.drought.gov/sites/default/files/2021-09/NOAA-Drought-Task-Force-IV-Southwest-Drought-Report-9-23-21.pdf.

  • McEvoy, D. J., and B. J. Hatchett, 2023: Spring heat waves drive record western United States snow melt in 2021. Environ. Res. Lett., 18, 014007, https://doi.org/10.1088/1748-9326/aca8bd.

    • Search Google Scholar
    • Export Citation
  • Meromy, L., N. P. Molotch, T. E. Link, S. R. Fassnacht, and R. Rice, 2013: Subgrid variability of snow water equivalent at operational snow stations in the western USA. Hydrol. Processes, 27, 23832400, https://doi.org/10.1002/hyp.9355.

    • Search Google Scholar
    • Export Citation
  • Milly, P. C. D., and K. A. Dunne, 2020: Colorado River flow dwindles as warming-driven loss of reflective snow energizes evaporation. Science, 367, 12521255, https://doi.org/10.1126/science.aax0194.

    • Search Google Scholar
    • Export Citation
  • Moser, C., 2021: Colorado Basin river forecast center. “Water year 2021 early season water supply outlook.” Accessed 5 February 2021, https://www.cbrfc.noaa.gov/present/2021/CBRFC_WY2021_EarlyOutlook.pdf.

  • Musselman, K. N., A. Nans, J. Vano, and N. P. Molotch, 2019: Reconsidering the utility of the April 1st snow water equivalent metric for water resource applications. Accessed 25 October 2022, https://westernsnowconference.org/files/PDFs/2019Musselman.pdf.

  • NCEI, 2022: Climate at a glance: Statewide time series, average temperature. Accessed 27 September 2022, https://www.ncdc.noaa.gov/cag/.

  • Oyler, J. W., S. Z. Dobrowski, A. P. Ballantyne, A. E. Klene, and S. W. Running, 2015: Artificial amplification of warming trends across the mountains of the western United States. Geophys. Res. Lett., 42, 153161, https://doi.org/10.1002/2014GL062803.

    • Search Google Scholar
    • Export Citation
  • Runyon, L., 2020: Colorado River Basin winter forecast signals dry times ahead. National Public Radio Northern Colorado, accessed 26 May 2023, https://www.watereducation.org/aquafornia-news/colorado-river-basin-winter-forecast-signals-dry-times-ahead.

  • Sackett, H. D., 2021: Drought and dry soils again will diminish Colorado’s spring runoff. Aspen Journalism, accessed 28 September 2022, https://aspenjournalism.org/drought-and-dry-soils-again-will-diminish-colorados-spring-runoff/.

  • Sakas, M. E., 2021: Dry soils and drought mean even normal snowpack can’t keep up with climate change in the west. Colorado Public Radio News, accessed 28 September 2022, https://www.cpr.org/2021/05/17/dry-soils-and-drought-mean-even-normal-snowpack-cant-keep-up-with-climate-change-in-the-west/.

  • Schmidt, R. A., 1982: Vertical profiles of wind speed, snow concentration, and humidity in blowing snow. Bound.-Layer Meteor., 23, 223246, https://doi.org/10.1007/BF00123299.

    • Search Google Scholar
    • Export Citation
  • Seager, R., M. Ting, C. Li, N. Naik, B. Cook, J. Nakamura, and H. Liu, 2013: Projections of declining surface-water availability for the southwestern United States. Nat. Climate Change, 3, 482486, https://doi.org/10.1038/nclimate1787.

    • Search Google Scholar
    • Export Citation
  • Sengupta, A., B. Singh, M. J. DeFlorio, C. Raymond, A. W. Robertson, X. Zeng, D. E. Waliser, and J. Jones, 2022: Advances in subseasonal to seasonal prediction relevant to water management in the western United States. Bull. Amer. Meteor. Soc., 103, E2168E2175, https://doi.org/10.1175/BAMS-D-22-0146.1.

    • Search Google Scholar
    • Export Citation
  • Siirila-Woodburn, E. R., and Coauthors, 2021: A low-to-no snow future and its impacts on water resources in the western United States. Nat. Rev. Earth Environ., 2, 800819, https://doi.org/10.1038/s43017-021-00219-y.

    • Search Google Scholar
    • Export Citation
  • Skiles, S. M. K., T. H. Painter, J. Belnap, L. Holland, R. L. Reynolds, H. L. Goldstein, and J. Lin, 2015: Regional variability in dust-on-snow processes and impacts in the Upper Colorado River Basin. Hydrol. Processes, 29, 53975413, https://doi.org/10.1002/hyp.10569.

    • Search Google Scholar
    • Export Citation
  • Talsma, C. J., K. E. Bennett, and V. V. Vesselinov, 2022: Characterizing drought behavior in the Colorado River Basin using unsupervised machine learning. Earth Space Sci., 9, e2021EA002086, https://doi.org/10.1029/2021EA002086.

    • Search Google Scholar
    • Export Citation
  • Udall, B., and J. Overpeck, 2017: The twenty-first century Colorado River hot drought and implications for the future. Water Resour. Res., 53, 24042418, https://doi.org/10.1002/2016WR019638.

    • Search Google Scholar
    • Export Citation
  • USBR, 2013: Managing water in the west Colorado River overview. 16 pp., https://usea.org/sites/default/files/event-/Final_Colorado_Overview.pdf.

  • USBR, 2022: Colorado River Basin water supply and demand study executive summary. 34 pp., https://www.usbr.gov/watersmart/bsp/docs/finalreport/ColoradoRiver/CRBS_Executive_Summary_FINAL.pdf.

  • USDA NRCS, 2021: Snowpack Telemetry (SNOTEL) historical data. USDA Natural Resources Conservation Services, accessed October 2021, https://wcc.sc.egov.usda.gov/nwcc/tabget.

  • USDA NRCS, 2022a: Snowpack Telemetry (SNOTEL) interactive map. USDA Natural Resources Conservation Services, accessed October 2022, https://www.nrcs.usda.gov/wps/portal/wcc/home/quicklinks/imap.

  • USDA NRCS, 2022b: Snowpack Telemetry (SNOTEL) report generator. USDA Natural Resources Conservation Services, accessed 8 December 2022, https://wcc.sc.egov.usda.gov/reportGenerator/.

  • Vitart, F., and A. W. Robertson, 2018: The sub-seasonal to seasonal prediction project (S2S) and the prediction of extreme events. npj Climate Atmos. Sci., 1, 3, https://doi.org/10.1038/s41612-018-0013-0.

    • Search Google Scholar
    • Export Citation
  • Vrugt, J. A., H. V. Gupta, S. C. Dekker, S. Sorooshian, T. Wagener, and W. Bouten, 2006: Application of stochastic parameter optimization to the Sacramento Soil Moisture Accounting model. J. Hydrol., 325, 288307, https://doi.org/10.1016/j.jhydrol.2005.10.041.

    • Search Google Scholar
    • Export Citation
  • White, C. J., and Coauthors, 2017: Potential applications of subseasonal-to-seasonal (S2S) predictions. Meteor. Appl., 24, 315325, https://doi.org/10.1002/met.1654.

    • Search Google Scholar
    • Export Citation
  • Xia, Y., J. Sheffield, M. B. Ek, J. Dong, N. Chaney, H. Wei, J. Meng, and E. F. Wood, 2014: Evaluation of multi-model simulated soil moisture in NLDAS-2. J. Hydrol., 512, 107125, https://doi.org/10.1016/j.jhydrol.2014.02.027.

    • Search Google Scholar
    • Export Citation
  • Xiao, M., B. Udall, and D. P. Lettenmaier, 2018: On the causes of declining Colorado River streamflows. Water Resour. Res., 54, 67396756, https://doi.org/10.1029/2018WR023153.

    • Search Google Scholar
    • Export Citation
Save
  • Adam, J. C., A. F. Hamlet, and D. P. Lettenmaier, 2009: Implications of global climate change for snowmelt hydrology in the twenty-first century. Hydrol. Processes, 23, 962972, https://doi.org/10.1002/hyp.7201.

    • Search Google Scholar
    • Export Citation
  • Adams, J., 2017: climate_indices, an open source Python library providing reference implementations of commonly used climate indices. Accessed January 2023, https://github.com/monocongo/climate_indices.

  • Aggett, G., 2022: Colorado airborne snow observatory expansion plan: Project report. 73 pp., https://coloradosnow.org/.

  • CBRFC, 2021: CBRFC water year in review. Accessed October 2022, https://www.cbrfc.noaa.gov/present/yir/yir_2020.pdf.

  • Chow, D., 2020: Colorado—the entire state—is in the midst of a widespread drought. National Broadcasting Company News, accessed 26 May 2023, https://www.nbcnews.com/science/environment/colorado-entire-state-midst-widespread-drought-n1236333.

  • Christensen, N. S., and D. P. Lettenmaier, 2007: A multimodel ensemble approach to assessment of climate change impacts on the hydrology and water resources of the Colorado River Basin. Hydrol. Earth Syst. Sci., 11, 14171434, https://doi.org/10.5194/hess-11-1417-2007.

    • Search Google Scholar
    • Export Citation
  • Church, J. E., 1935: Principles of snow surveying as applied to forecasting stream flow. J. Agric. Res., 51, 97130.

  • Daly, C., M. Halbleib, J. I. Smith, W. P. Gibson, M. K. Doggett, G. H. Taylor, J. Curtis, and P. P. Pasteris, 2008: Physiographically sensitive mapping of climatological temperature and precipitation across the conterminous United States. Int. J. Climatol., 28, 20312064, https://doi.org/10.1002/joc.1688.

    • Search Google Scholar
    • Export Citation
  • Erlingis, J. M., and Coauthors, 2021: A high‐resolution land data assimilation system optimized for the western United States. J. Amer. Water Resour. Assoc., 57, 692710, https://doi.org/10.1111/1752-1688.12910.

    • Search Google Scholar
    • Export Citation
  • Fleming, S. W., and Goodbody, A. G., 2019: A machine learning metasystem for robust probabilistic nonlinear regression‐based forecasting of seasonal water availability in the US west. IEEE Access, 7, 11 994311 9964, https://doi.org/10.1109/ACCESS.2019.2936989.

    • Search Google Scholar
    • Export Citation
  • Gilbert, D., 2022: As Colorado warms, dry soil sucks up more water. That’s bad news for rivers and farmers. Colorado Sun, accessed 28 September 2022, https://coloradosun.com/2022/02/06/colorado-soil-dry-climate-change/.

  • Gonzalez, P., and Coauthors, 2018: Southwest. Impacts, Risks, and Adaptation in the United States: Fourth National Climate Assessment, Volume II, D. R. Reidmiller et al., Eds., U.S. Global Change Research Program, 1101–1184, https://doi.org/10.7930/NCA4.2018.CH25.

  • Harpold, A. A., K. Sutcliffe, J. Clayton, A. Goodbody, and S. Vazquez, 2017: Does including soil moisture observations improve operational streamflow forecasts in snow-dominated watersheds? J. Amer. Water Resour. Assoc., 53, 179196, https://doi.org/10.1111/1752-1688.12490.

    • Search Google Scholar
    • Export Citation
  • Heidke, P., 1926: Berechnung des Erfolges und der Gute der Windstarkevorhersagen im Sturmwarnungsdienst (Measures of success and goodness of wind force forecasts by the gale-warning service). Geogr. Ann., 8, 301349, https://doi.org/10.1080/20014422.1926.11881138.

    • Search Google Scholar
    • Export Citation
  • Hoerling, M., J. Barsugli, B. Livneh, J. Eischeid, X. Quan, and A. Badger, 2019: Causes for the century-long decline in Colorado River flow. J. Climate, 32, 81818203, https://doi.org/10.1175/JCLI-D-19-0207.1.

    • Search Google Scholar
    • Export Citation
  • Houborg, R., M. Rodell, B. Li, R. Reichle, and B. Zaitchik, 2012: Drought indicators based on model-assimilated Gravity Recovery and Climate Experiment (GRACE) terrestrial water storage observations. Water Resour. Res., 48, W07525, https://doi.org/10.1029/2011WR011291.

    • Search Google Scholar
    • Export Citation
  • Koren, V., M. Smith, and Z. Cui, 2014: Physically-based modifications to the Sacramento Soil Moisture Accounting model. Part A: Modeling the effects of frozen ground on the runoff generation process. J. Hydrol., 519, 34753491, https://doi.org/10.1016/j.jhydrol.2014.03.004.

    • Search Google Scholar
    • Export Citation
  • Koster, R., S. P. P. Mahanama, B. Livneh, D. P. Lettenmaier, and R. H. Reichle, 2010: Skill in streamflow forecasts derived from large-scale estimates of soil moisture and snow. Nat. Geosci., 3, 613616, https://doi.org/10.1038/ngeo944.

    • Search Google Scholar
    • Export Citation
  • Livneh, B., and A. M. Badger, 2020: Drought less predictable under declining future snowpack. Nat. Climate Change, 10, 452458, https://doi.org/10.1038/s41558-020-0754-8.

    • Search Google Scholar
    • Export Citation
  • Lukas, J., 2021: Why do runoff forecasts diverge from snowpack conditions? Accessed 25 October 2022, https://lukasclimate.com/blog/2021/2/25/why-isnt-the-seasonal-runoff-forecast-as-a-of-normal-equal-to-the-current-snowpack-of-normal.

  • Lukas, J., and E. Payton, Eds., 2020: Colorado River Basin climate and hydrology: State of the science. Western Water Assessment, University of Colorado Boulder Tech. Rep., 119 pp., https://doi.org/10.25810/3hcv-w477.

  • Lukas, J., J. Barsugli, N. Doesken, I. Rangwala, and K. Wolter, 2014: Climate change in Colorado: A synthesis to support water resources management and adaptation. A report for the Colorado Water Conservation Board, Western Water Assessment, 2nd ed., 114 pp., https://wwa.colorado.edu/sites/default/files/2021-07/Climate_Change_CO_Report_2014_FINAL.pdf.

  • Mankin, J. S., I. Simpson, A. Hoell, R. Fu, J. Lisonbee, A. Sheffield, and D. Barrie, 2021: NOAA drought task force report on the 2020–2021 southwestern U.S. drought. NOAA Drought Task Force, MAPP, and NIDIS, 20 pp., https://www.drought.gov/sites/default/files/2021-09/NOAA-Drought-Task-Force-IV-Southwest-Drought-Report-9-23-21.pdf.

  • McEvoy, D. J., and B. J. Hatchett, 2023: Spring heat waves drive record western United States snow melt in 2021. Environ. Res. Lett., 18, 014007, https://doi.org/10.1088/1748-9326/aca8bd.

    • Search Google Scholar
    • Export Citation
  • Meromy, L., N. P. Molotch, T. E. Link, S. R. Fassnacht, and R. Rice, 2013: Subgrid variability of snow water equivalent at operational snow stations in the western USA. Hydrol. Processes, 27, 23832400, https://doi.org/10.1002/hyp.9355.

    • Search Google Scholar
    • Export Citation
  • Milly, P. C. D., and K. A. Dunne, 2020: Colorado River flow dwindles as warming-driven loss of reflective snow energizes evaporation. Science, 367, 12521255, https://doi.org/10.1126/science.aax0194.

    • Search Google Scholar
    • Export Citation
  • Moser, C., 2021: Colorado Basin river forecast center. “Water year 2021 early season water supply outlook.” Accessed 5 February 2021, https://www.cbrfc.noaa.gov/present/2021/CBRFC_WY2021_EarlyOutlook.pdf.

  • Musselman, K. N., A. Nans, J. Vano, and N. P. Molotch, 2019: Reconsidering the utility of the April 1st snow water equivalent metric for water resource applications. Accessed 25 October 2022, https://westernsnowconference.org/files/PDFs/2019Musselman.pdf.

  • NCEI, 2022: Climate at a glance: Statewide time series, average temperature. Accessed 27 September 2022, https://www.ncdc.noaa.gov/cag/.

  • Oyler, J. W., S. Z. Dobrowski, A. P. Ballantyne, A. E. Klene, and S. W. Running, 2015: Artificial amplification of warming trends across the mountains of the western United States. Geophys. Res. Lett., 42, 153161, https://doi.org/10.1002/2014GL062803.

    • Search Google Scholar
    • Export Citation
  • Runyon, L., 2020: Colorado River Basin winter forecast signals dry times ahead. National Public Radio Northern Colorado, accessed 26 May 2023, https://www.watereducation.org/aquafornia-news/colorado-river-basin-winter-forecast-signals-dry-times-ahead.

  • Sackett, H. D., 2021: Drought and dry soils again will diminish Colorado’s spring runoff. Aspen Journalism, accessed 28 September 2022, https://aspenjournalism.org/drought-and-dry-soils-again-will-diminish-colorados-spring-runoff/.

  • Sakas, M. E., 2021: Dry soils and drought mean even normal snowpack can’t keep up with climate change in the west. Colorado Public Radio News, accessed 28 September 2022, https://www.cpr.org/2021/05/17/dry-soils-and-drought-mean-even-normal-snowpack-cant-keep-up-with-climate-change-in-the-west/.

  • Schmidt, R. A., 1982: Vertical profiles of wind speed, snow concentration, and humidity in blowing snow. Bound.-Layer Meteor., 23, 223246, https://doi.org/10.1007/BF00123299.

    • Search Google Scholar
    • Export Citation
  • Seager, R., M. Ting, C. Li, N. Naik, B. Cook, J. Nakamura, and H. Liu, 2013: Projections of declining surface-water availability for the southwestern United States. Nat. Climate Change, 3, 482486, https://doi.org/10.1038/nclimate1787.

    • Search Google Scholar
    • Export Citation
  • Sengupta, A., B. Singh, M. J. DeFlorio, C. Raymond, A. W. Robertson, X. Zeng, D. E. Waliser, and J. Jones, 2022: Advances in subseasonal to seasonal prediction relevant to water management in the western United States. Bull. Amer. Meteor. Soc., 103, E2168E2175, https://doi.org/10.1175/BAMS-D-22-0146.1.

    • Search Google Scholar
    • Export Citation
  • Siirila-Woodburn, E. R., and Coauthors, 2021: A low-to-no snow future and its impacts on water resources in the western United States. Nat. Rev. Earth Environ., 2, 800819, https://doi.org/10.1038/s43017-021-00219-y.

    • Search Google Scholar
    • Export Citation
  • Skiles, S. M. K., T. H. Painter, J. Belnap, L. Holland, R. L. Reynolds, H. L. Goldstein, and J. Lin, 2015: Regional variability in dust-on-snow processes and impacts in the Upper Colorado River Basin. Hydrol. Processes, 29, 53975413, https://doi.org/10.1002/hyp.10569.

    • Search Google Scholar
    • Export Citation
  • Talsma, C. J., K. E. Bennett, and V. V. Vesselinov, 2022: Characterizing drought behavior in the Colorado River Basin using unsupervised machine learning. Earth Space Sci., 9, e2021EA002086, https://doi.org/10.1029/2021EA002086.

    • Search Google Scholar
    • Export Citation
  • Udall, B., and J. Overpeck, 2017: The twenty-first century Colorado River hot drought and implications for the future. Water Resour. Res., 53, 24042418, https://doi.org/10.1002/2016WR019638.

    • Search Google Scholar
    • Export Citation
  • USBR, 2013: Managing water in the west Colorado River overview. 16 pp., https://usea.org/sites/default/files/event-/Final_Colorado_Overview.pdf.

  • USBR, 2022: Colorado River Basin water supply and demand study executive summary. 34 pp., https://www.usbr.gov/watersmart/bsp/docs/finalreport/ColoradoRiver/CRBS_Executive_Summary_FINAL.pdf.

  • USDA NRCS, 2021: Snowpack Telemetry (SNOTEL) historical data. USDA Natural Resources Conservation Services, accessed October 2021, https://wcc.sc.egov.usda.gov/nwcc/tabget.

  • USDA NRCS, 2022a: Snowpack Telemetry (SNOTEL) interactive map. USDA Natural Resources Conservation Services, accessed October 2022, https://www.nrcs.usda.gov/wps/portal/wcc/home/quicklinks/imap.

  • USDA NRCS, 2022b: Snowpack Telemetry (SNOTEL) report generator. USDA Natural Resources Conservation Services, accessed 8 December 2022, https://wcc.sc.egov.usda.gov/reportGenerator/.

  • Vitart, F., and A. W. Robertson, 2018: The sub-seasonal to seasonal prediction project (S2S) and the prediction of extreme events. npj Climate Atmos. Sci., 1, 3, https://doi.org/10.1038/s41612-018-0013-0.

    • Search Google Scholar
    • Export Citation
  • Vrugt, J. A., H. V. Gupta, S. C. Dekker, S. Sorooshian, T. Wagener, and W. Bouten, 2006: Application of stochastic parameter optimization to the Sacramento Soil Moisture Accounting model. J. Hydrol., 325, 288307, https://doi.org/10.1016/j.jhydrol.2005.10.041.

    • Search Google Scholar
    • Export Citation
  • White, C. J., and Coauthors, 2017: Potential applications of subseasonal-to-seasonal (S2S) predictions. Meteor. Appl., 24, 315325, https://doi.org/10.1002/met.1654.

    • Search Google Scholar
    • Export Citation
  • Xia, Y., J. Sheffield, M. B. Ek, J. Dong, N. Chaney, H. Wei, J. Meng, and E. F. Wood, 2014: Evaluation of multi-model simulated soil moisture in NLDAS-2. J. Hydrol., 512, 107125, https://doi.org/10.1016/j.jhydrol.2014.02.027.

    • Search Google Scholar
    • Export Citation
  • Xiao, M., B. Udall, and D. P. Lettenmaier, 2018: On the causes of declining Colorado River streamflows. Water Resour. Res., 54, 67396756, https://doi.org/10.1029/2018WR023153.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    Time series of the standardized precipitation evapotranspiration index (SPEI) for Colorado Climate Division II (1895–2021). SPEI is calculated using the average monthly temperature and precipitation over the climate division in the NOAA nClimGrid dataset (Adams 2017), a Thornthwaite potential evapotranspiration estimate, and a Pearson distribution trained on data from 1901 to 2020.

  • Fig. 2.

    (a) The 2020 percent of 1991–2020 normal 1 Apr snowpack, (b) forecasted April–July runoff, and (c) observed April–July runoff measured at basin scale (Hydrological Unit Code 6). Black outlines are HUC 6 boundaries and blue lines are HUC 4 boundaries. Source: USDA NRCS (2022b).

  • Fig. 3.

    Locations of United States Geological Survey stream gages (blue circles) and SNOTEL snowpack/precipitation observations (black circles) used. The yellow-shaded area represents Colorado west of the Continental Divide, blue lines are rivers, and black lines are watershed boundaries.

  • Fig. 4.

    Fraction of variance in April–August streamflow explained by statistical hindcasts (1981–2021) with no future precipitation and temperature data (black), and precipitation + spring temperature data for the remainder of the water year (blue). Gray shaded area represents improvements in streamflow hindcasts to be gained from foresight of observations (FO). The blue shaded area represents error from other sources.

  • Fig. 5.

    Mean absolute error in April–August streamflow, represented in standard deviations, explained by statistical hindcasts (1981–2021) with no future precipitation and temperature data (black), and precipitation + spring temperature data for the remainder of the water year (blue). Gray shaded area represents improvements in streamflow hindcasts to be gained from foresight of observations (FO). The blue shaded area represents error from other sources.

  • Fig. 6.

    Heidke skill score for April–August streamflow explained by statistical hindcasts (1981–2021) with no future precipitation and temperature data (black), and precipitation + spring temperature data for the remainder of the water year (blue). Gray shaded area represents improvements in streamflow hindcasts to be gained from foresight of observations (FO). The blue shaded area represents error from other sources.

All Time Past Year Past 30 Days
Abstract Views 177 177 0
Full Text Views 751 751 102
PDF Downloads 521 521 85