As of early February, accumulation of snowpack in the Snake River Basin has reached about two-thirds of its annual maximum, which makes this a good time to assess the current water situation and make some projections for the upcoming spring.
In this blog, Christina and I analyze snow-water equivalent (SWE) at five sites in and near the upper Henry’s Fork and Fall River watersheds, translate SWE projections into projected spring-time streamflow and reservoir content at Island Park, and conclude with analysis of long-term trends in SWE. In addition to providing actual data and predictions, we provide some insight into the statistical methods used to make these predictions.
The Short Version
If you don’t want to read the whole blog, here are the main points:
- Current snowpack in the Henry’s Fork watershed is near the long-term mean and expected to remain there through the remainder of the winter.
- Winter inflow to Island Park Reservoir can be predicted very accurately from knowledge of flow during the previous spring and summer.
- Spring-time inflow is harder to predict, but we developed a model that explains 80% of the variability in April-June inflow to Island Park Reservoir. Its inputs are previous winter streamflow, April 1 snow-water equivalent (SWE) at the Crab Creek site in the Centennial Mountains and April 1 SWE at the Black Bear site on the Madison Plateau.
- Using these predictive models, we project that inflows to Island Park Reservoir over the remainder of the winter and spring will not be sufficient to reach the fill target of 127,000 acre-feet (94% of capacity) by April 1 at the current outflow of 130 cfs. However, outflow decreases will not be necessary, as the reservoir will reach this target by April 11, and under reasonable outflow management, it will be close to capacity by the end of April, thanks to this year’s average snowpack.
- Although we predict that spring-time inflow to Island Park Reservoir will be only 77% of average due to cumulative effects of three years of drought, it will be 38% higher than that in 2015 due to average snowpack this year. Our predictions for inflow, reservoir storage, and outflow management indicate that there will be enough flexibility to fill water rights, allow a small amount of flood-control space, and achieve outflows of 500-800 cfs during the early part of the fishing season, in comparison to 350-450 cfs during 2013 and 2014. Again, this improvement is attributable to an average snowpack. Improvements in winter flow from this year’s average snowpack will not be realized until the winter of 2016-2017.
- We found no evidence of long-term declines in April 1 SWE at four of the five SnoTel sites in and near the upper Henry’s Fork watershed. However, we observed a statistically significant decrease in April 1 SWE at the Crab Creek SnoTel site of about 4.4% per decade, which indicates the potential for decreases in inflow to Island Park Reservoir and associated decreases in winter outflows.
Current and Projected Snow-Water Equivalent (SWE)
There are five Snow Telemetry (SnoTel) sites that are the primary predictors of water supply in the upper Henry’s Fork and Fall River watersheds: 1) Crab Creek, elevation 6880 feet, located in the Centennial Mountains just to the east of Interstate 15, 2) White Elephant, elevation 7710 feet, located on the east side of Mt. Sawtelle in Island Park, 3) Island Park, elevation 6290, located near Ponds Lodge in Island Park, 4) Black Bear, elevation 8170 feet, located just over the continental divide in Montana, between Big Springs and West Yellowstone, and 5) Grassy Lake, elevation 7265 feet, located near Grassy Lake near the south entrance to Yellowstone National Park. As of February 5, SWE at these five sites ranges from 90% of the 1981-2010 median for this date at Black Bear to 122% of median at Crab Creek. Averaged over the whole Henry’s Fork watershed, including the Teton subwatershed, SWE is currently 102% of the 30-year median.
Long-range weather forecasts from the National Weather Service Climate Prediction Center give relative probabilities of below-normal, normal, and above-normal precipitation for the upcoming months. Equal chances would be a 33.3% probability each of below-normal, normal, and above-normal precipitation. The 1-month outlook predicts a 40% probability of below-normal precipitation for our region for the month of February. The 3-month outlook calls for equal chances. Thus, for our projections, we assumed probabilities of 40% below-normal, 40% normal, and 20% above-normal for February, and equal chances of each for March. We used the 25th percentile of historic snow-accumulation as “below normal”, the median as “normal”, and the 75th percentile as “above normal.” Using these probabilities and historic data, we projected snow accumulation at each of the five sites between now and April 1, which is close to the average date of maximum snow accumulation. Current SWE to date and our projects are shown in the following graphs.
You can see that our April 1 projections fall very close to the long-term mean at all five sites. This occurs for two reasons: 1) SWE accumulation from here on depends greatly on current values, which are very close to long-term averages, and 2) the probability of below-average precipitation for February is not much greater than 33.3%, so with an average March in the forecast, SWE accumulation between now and April 1 is projected to remain very close to average.
You can also see from the graphs above how strongly SWE is dependent on elevation. All of the graphs are on the same scale, so visual comparison immediately indicates how much more snow falls at elevations above 7000 feet than at elevations below 7000 feet.
Predicting inflow to Island Park Reservoir
As you know, there is a streamflow gage on the Henry’s Fork immediately downstream of Island Park Reservoir, which measures reservoir outflow. There is also a streamflow gage in the Henry’s Fork at Coffee Pot, but this measures only a portion of the inflow to the reservoir. A substantial amount of inflow, especially during spring-time runoff, comes from streams that drain the Centennial Mountains. These streams are not gaged. Thus, we calculate inflow to Island Park Reservoir using the equation
Inflow = Outflow - change in reservoir storage + reservoir evaporation
Change in storage is measured by a gage on the reservoir itself. Positive change-in-storage occurs when the reservoir is filling, so calculated inflow is greater than outflow. Negative change-in-storage occurs when the reservoir is being drained, so calculated inflow is less than outflow. Evaporation from the reservoir surface occurs when the reservoir is not covered with ice. This term in the equation is small compared to the others, but nonetheless, it is water that flows into the reservoir that is not reflected in outflow, so we include it to make sure the total water budget balances. We usually subtract the contribution of the Henry’s Lake basin from inflow to Island Park Reservoir, for several reasons. First, during storage season, most if not all inflow to Henry’s Lake is stored in Henry’s Lake and so does not contribute to Island Park Reservoir fill. Second, the Henry’s Lake basin contributes less than 10% to the annual water supply of the Henry’s Fork at Island Park. Lastly, the hydrogeology of the Henry’s Lake basin is quite a bit different than that of the mainstem Henry’s Fork and the Centennial Mountains tributaries. Thus, our predictive models are more precise when the Henry’s Lake basin is removed from the calculations.
We use statistical models to predict inflow to Island Park Reservoir over two critical time periods: winter (October-March) and spring (April-June). The winter inflow prediction is critical to setting reservoir outflow appropriately to achieve reservoir storage of around 127,000 acre-feet (94% of capacity) on April 1. The spring-time prediction is important for setting outflow during the spring to fill the reservoir to capacity (135,000) acre-feet prior to high irrigation demand in the upper Snake River basin as a whole, but while also leaving a small amount of space in the reservoir to capture a flood, should extremely warm weather or rain-on-snow produce excessive runoff. Spring-time operations have become quite challenging during the past two years, as evidenced by low outflows at the beginning of fishing season. In both 2014 and 2015, inflows much lower than average resulted in the need to lower outflows in order to top off the reservoir during late May and early June.
Prediction of winter inflow is easy. Statistical models show that winter inflow to Island Park Reservoir is very strongly predicted by inflow during the previous spring and summer (April-August). The graph below shows the dependence of winter inflow on spring/summer inflow. It also shows the winter inflow value we estimated last October 1 for the winter of 2015/2016. We predicted a mean inflow of 341 cfs. Including statistical uncertainty, we predicted inflow to fall between 300 cfs and 387 cfs. Mean winter inflow through February 4 was 358 cfs, and accounting for the observation that minimum inflow occurs each year during late February and early March, our current projection for the October-March mean is 353 cfs, only 3.5% from our prediction and well within the margin of error.
Prediction of winter flow is very precise because winter flow is made up entirely of what we call “baseflow,” the component of flow derived from groundwater. Baseflows are very stable and predicted well by streamflow over previous time periods. Prediction of spring-time flow in the Henry’s Fork watershed is much more difficult and imprecise because it consists of three components: 1) baseflow, 2) snowmelt, and 3) spring-time rain. Indeed, the best statistical model of spring-time streamflow depends on all three of these inputs, but unfortunately, at the end of the winter or beginning of the spring, when we need to make the prediction, we have no idea of how much rain will fall during the spring. Thus, for our predictive model to be of any use at the time when we need to use it, we omitted spring-time precipitation as a predictor.
Extensive and sophisticated statistical modeling produced a reasonably good predictive model of spring-time (April-June) inflow to Island Park Reservoir. This model requires three inputs: 1) inflow during the previous winter, 2) April 1 SWE at the Crab Creek site, and 3) April 1 SWE at the Black Bear site. The most important of these predictors was winter inflow, which represents the baseflow component of spring-time streamflow. This predictor turned out to be more important than either of the snowpack predictors because the Henry’s Fork is primarily fed by groundwater springs, not by snowmelt. The second-most important predictor was April 1 SWE at Crab Creek, representing the contribution of runoff from the Centennial Mountains to Island Park inflow. The weakest predictor—still necessary for a good model but not as critical as the other two—was SWE at the Black Bear site, which represents contribution of runoff from the Yellowstone Plateau. However, as you know, most of the snowmelt on the Plateau recharges the deep aquifers that feed Big Springs and other springs, which is why SWE at Black Bear was less important to runoff than the other two predictors. Most of the effect of snowmelt at the Black Bear site is realized in subsequent years as increased baseflow in the upper Henry’s Fork Watershed.
You might be surprised that neither the Island Park nor the White Elephant SnoTel sites contributed to this model. SWE at Island Park does not physically contribute Island Park Reservoir inflow because the site actually sits downstream of the reservoir’s watershed. Snowmelt from the White Elephant site contributes to runoff into Island Park Reservoir, but it sits half-way between the Crab Creek and Black Bear sites and hence represents a spatial average of those two sites. This averaging eliminates small-scale variations in weather patterns that produce different snow accumulation patterns in the Centennial Mountains versus the Yellowstone Plateau. Including both Crab Creek and Black Bear as distinct sites that reflect the two primary sources of snowmelt in the upper Henry’s Fork watershed produced a much better model than when SWE from the White Elephant site was included.
The following graph shows the model-fitted values on the vertical axis and the observed values on the horizontal axis. Although the fit is good—the model explains 80% of the variability in the data—it is obviously not as tight as the model for winter flow, which explains 90% of the variability in the data.
Next, we used our projected winter flow, April 1 Crab Creek SWE, and April 1 Black Bear SWE values to estimate mean April-June inflow to Island Park Reservoir for 2016. The actual numerical values were 353 cfs winter inflow (76% of long-term mean), 34.2 inches SWE at Black Bear (88.7% of long-term mean), and 14.9 inches SWE at Crab Creek (108% of long-term mean). The resulting prediction was 672 cfs, which is 77% of the 1979-2015 mean. Statistical error around this prediction is large for reasons stated above; with 95% certainty, the true value of mean spring-time inflow lies between 481 cfs and 938 cfs.
The following graph shows dependence of spring-time inflow on winter inflow, the strongest of the three predictors in the model. This graph also shows our prediction for 2016.
This graph clearly shows the reason why our inflow prediction is so far below average, despite what appears right now to be abundant snow on the ground. Remember that the most important predictor of spring-time runoff is winter baseflow, which is at near-record lows due to a three-year drought. You may recall that in earlier blogs, we reported that 2013-2015 was the driest three-year sequence since 1940-1942, which led to low inflows this winter, which, in turn, leads to a prediction of low spring-time runoff. However, last year’s runoff averaged a paltry 485 cfs, so our projection for 2016 beats this by 187 cfs—a whopping 39% increase—due to this year’s average snowpack.
Projected reservoir contents and streamflows
Current outflow from Island Park Reservoir is around 130 cfs, and current inflow is around 335 cfs. At this rate, the reservoir is filling at 406 acre-feet per day, which is not sufficient to achieve the 127,000 acre-foot target by April 1. However, given the average snowpack, there does not appear to be any reason to reduce outflows from Island Park. Using all of the projections above, and assuming normal timing for snowmelt this spring, the reservoir will reach this target on April 11. The following graphs show observed-to-date and predicted inflow, outflow, and reservoir content—along with long-term means for comparison.
Our projected outflow scenario will fill the reservoir to near capacity around the end of April, allowing flexibility to leave some room to accommodate flood-control space while maximizing retention of physical water in the reservoir at the time when Fremont-Madison Irrigation District’s water rights will be accruing on paper. Projected inflows during late May and June, while still below average, are much higher than those in 2014 and 2015. This will provide much more flexibility than has existed the past two years to align physical reservoir contents and paper water rights prior to high irrigation demand, while providing outflows in the 500-700 cfs range at the beginning of fishing season. This will be a large improvement over flows in 2014 and 2015, which were in the 350-450 cfs range. It is nearly impossible right now to predict when irrigation demand will begin and how much storage delivery it will require, but we will be able to take a stab at that at the end of March.
Long-term Trends in Snowpack
As a final topic for this blog, we analyzed long-term trends in April 1 SWE at the five SnoTel sites. We found strong statistical evidence of a decline in April 1 SWE at the Crab Creek site but no statistical evidence for declining SWE at four of the five sites. The best statistical description of SWE at these four sites was simply random variation around the mean, except that there was a statistically significant pattern of alternating above-average and below-average years at the Island Park site. The graphs are shown below.
The good news is that in contrast to other mountainous areas in the western U.S., most of our SnoTel sites are not displaying long-term declines in April 1 SWE. This is most likely due to our location in a geographic “sweet spot.” Our snow-accumulating mountains are higher than those in northern California, northern Idaho, and northwestern Montana; farther north than those in the southwestern states; and farther from the ocean than the Cascades and Sierra Nevada. The bad news is that SWE at the Crab Creek site is the second-most important predictor of spring-time runoff into Island Park Reservoir, and it is experiencing a long-term decline of 0.6 inches (around 4.4%) per decade. All other factors being equal, this will reduce spring-time inflow to Island Park Reservoir by about 2342 acre-feet—around 1.5% of the 1979-2015 mean spring-time inflow—over the next decade. If this reduction in spring-time inflow results in an equal reduction in end-of-season reservoir storage carryover, winter outflows from Island Park will be reduced by 13 cfs, about half of the gains we have accomplished through collaborative Drought Management Planning over the past ten years.
One last observation is that in the online SnoTel reporting system, current SWE values are compared to 30-year averages. These 30-year averages are updated every 10 years. The current 30-year averaging period is 1981-2010. At some SnoTel sites, averages over this time period are quite a bit lower than those over the previous 30-year time period. Thus, “100% of average” in the most recent period is actually below average when compared to previous periods. An example of this is given by the Black Bear site. Even though our statistical analysis showed no significant trend, it is easy to see that mean maximum SWE for the 1981-2010 period is over an inch lower than that for the 1972-2001 period. In this example, 100% of the 1981-2010 mean is only 97.5% of the 1972-2001 mean. This is something to keep in mind when reading and interpreting SnoTel or any other hydrologic or climatic data.