To provide streamflow information for all river stakeholders, we have constructed a computer simulation model of the Henry’s Fork watershed stream, reservoir, and irrigation system. Using early-April water-supply conditions and long-term temperature trends as inputs, we expect streamflow conditions to be generally better than average and much better than last year across the watershed. Focusing specifically on Island Park Reservoir and the river immediately downstream, we predict:
- Streamflow during the second half of June at Island Park Dam will be roughly equal to the river’s natural flow. With 90% probability, this natural streamflow will range between 400 and 750 cfs.
- Irrigation delivery from Island Park Reservoir will begin around July 1 and peak in mid-July. With 95% probability, releases from the reservoir during July will be lower than 1,400 cfs, and with over 95% probability will be much lower than 2016 releases between the middle of June and the first week of August.
- With 95% probability, Island Park Reservoir contents at the end of the September will remain above 58,000 ac-ft (43% full), very close to the long-term average.
- Because of lower outflows and higher reservoir contents, turbidity (how “dirty” the water appears) in the river immediately downstream of Island Park Dam is expected to remain lower than the high values observed in 2016 during the mid-July cyanobacteria bloom and late-summer reservoir drawdown.
What, when, where and why?
We have constructed a dynamic, randomized (the technical term is stochastic) computer simulation model of the Henry’s Fork water system, including natural streamflows, reservoir storage and delivery, river reach gains and losses, and irrigation diversion. In addition, because water clarity (measured by “turbidity”) downstream of Island Park Dam is critically important to the early- and mid-summer fishing experience, we have used statistical analysis of turbidity data collected in 2016 by our continuous-recording instruments to develop a simulation model of turbidity. A more detailed description of model inputs, methods, and assumptions is given in the “How the model works” section at the bottom of the blog.
The outputs are a series of hydrographs of the various system components over the period April 1 through September 30, 2017. Our predictions for 2017, including the range of values that will occur with 90% probability, are compared to conditions in 2016 and to long-term averages. A guide to interpreting the graphs is given in a separate section following the graphs. These predictions are designed to provide timely and relevant information to help anglers, fishing outfitters and guides, other river recreationists, water users, and water managers anticipate river, reservoir, and irrigation-delivery conditions during the upcoming summer. Model outputs, in the order we report them in this blog, are:
- Island Park Reservoir contents, and streamflow and turbidity in the Henry’s Fork immediately downstream of Island Park Dam.
- Henry’s Lake contents and outflow.
- Streamflow in the Henry’s Fork at Ashton, Fall River at Chester, Henry’s Fork at St. Anthony, and Teton River immediately upstream of the Crosscut Canal inflow point.
- Grassy Lake contents and outflow.
- Diversion of water into the Crosscut Canal at Chester Dam to be delivered to the Teton River to meet irrigation demand.
The reservoir fills around May 20, natural inflow is passed through the full reservoir until around July 1, modest irrigation delivery is required through July and August, and the reservoir reaches its minimum volume by the middle of September. With 95% confidence, ending reservoir content is very near or above its long-term mean of around 58,000 ac-ft (43% full). Expected ending content is around 110,000 ac-ft (81% full). Outflow is expected to remain below 1,000 cfs for the entire spring and summer, although flows up to 1,400 cfs during runoff in late May and again during irrigation season are possible within the 90% prediction interval.
One important thing to note is that between date of reservoir fill and the need for irrigation delivery, outflow from Island Park will be equal to the river’s natural flow (as if neither Henry’s Lake nor Island Park Reservoir existed). Because natural streamflow in the upper Henry’s Fork watershed is predicted to be only 85-90% of average, and because runoff is expected to occur slightly earlier than average, these natural streamflows during June may be as low as around 400 cfs and are unlikely to be much higher than 750 cfs—below average in any case. These values are completely determined by the natural water supply of the river, since the Island Park Reservoir management objective for all stakeholders is to keep as much water in the reservoir as possible. Delivering water during late June to keep river flows higher for recreational purposes when it is not needed for irrigation delivery will result in decreased water quality later in the summer and in decreased outflow during the subsequent winter.
Turbidity is expected to be a little higher than last year during spring and the early part of the summer, due to higher inflow during snowmelt. However, the high spikes in turbidity observed last year during mid- to late-summer are very unlikely to occur this year because the majority of the outflow will pass through the power plant rather than the west-side gates and because the reservoir will remain much higher than last year.
Henry’s Lake is expected to fill around its normal time in mid-June and end up pretty close to its long-term average by the end of the summer. Outflow is expected to increase to almost 200 cfs during runoff but could get much higher if runoff is higher than expected and occurs in a short time window in early June.
With 90% probability, streamflow in the Henry’s Fork at Ashton will be lower than the long-term average from the middle of May until the middle of August. This reflects the prediction of below-average streamflow and slightly earlier runoff timing for the upper Henry’s Fork watershed. However, streamflow during April and May will nearly certainly exceed last year’s streamflow. Late-summer flow at Ashton is expected to be near the long-term average and much higher than last year.
Streamflows in Fall River, Teton River, and Henry’s Fork at St. Anthony are expected to be near average to slightly above average during April, May and June, but fall below average in July, due mostly to slightly earlier runoff timing. The biggest discrepancy between projected 2017 flows and the long-term average occurs at St. Anthony, where flow will fall below average with 95% probability for most of July. With high probability, streamflow at St. Anthony will remain at the target minimum of 1,000 cfs for most of July, with release of water from Island Park Reservoir necessary to maintain this flow. However, with 95% probability, streamflows at these three locations will be higher than last year for the majority of the spring and summer. Late-summer flows are expected to rebound to long-term averages, reflecting increase baseflows as a result of above-average precipitation dating back to last October.
Grassy Lake will fill in late May, and outflow in the range of 40-70 cfs will be required to pass inflow once it is full. No irrigation delivery will be needed from Grassy Lake, which will essentially remain full through the irrigation season. In practice, if no Grassy Lake delivery is needed to meet irrigation demand, some water is delivered later in the summer to reduce its volume to around 12,000 before the start of winter. This provides space to capture inflow the following spring. We did not incorporate this delivery in the simulations.
Coincident with receding flows in the Teton River, delivery of water through the Crosscut Canal will be necessary starting on about July 1, which, in turn, is the date on which additional delivery will be needed from Island Park Reservoir. Expected need for Crosscut delivery is below average, but the 90% prediction interval contains both average and 2016 delivery during the month of August.
Each graph shows four pieces of information:
- 2017 prediction,
- 2016 actual value,
- long-term mean, and
- 90% interval for the prediction.
In statistical terms, this is the expected value or mean outcome, given the early-April starting conditions and uncertainties associated with streamflow, runoff timing, weather, etc. between now and September 30. This expected value is obtained by first running 5,000 independent, unique, random simulations of the summer, and then computing the arithmetic average (mean) of the resulting 5,000 output values.
2016 actual value
This is the actual data from the summer of 2016, which was one of the hottest, driest water years on record in the Henry’s Fork watershed (see my blog on runoff timing).
For all streamflow and reservoir outputs, the long-term mean is the arithmetic average over water years 1978-2016 (39 years). This period of record was chosen because it is the longest common record over which sufficient electronic data are available to analyze all of the relevant stream reaches and reservoirs. For the Crosscut Canal delivery to the Teton River, the long-term mean is the arithmetic average over water years 1988-2016 (29 years). This somewhat shorter period of record is used here because this parameter is a water-rights accounting measure, not a physical flow measure, and the electronically available water-rights accounting data for Water District 01 date back only to the 1988 irrigation season. There is no long-term mean for the turbidity record, because we have been continuously measuring turbidity only since 2014, when we established our water-quality monitoring network.
90% prediction interval
The gray shaded region in each figure shows the range of values that will occur with 90% probability, given the starting conditions and uncertainties. The top of this band is the 95th percentile of all values produced by the 5,000 simulations, and the bottom is the 5th percentile. In other words, the band contains the middle 90% of all possible outcomes. In some cases, this band is quite wide, reflecting greater uncertainty, and in other cases, the band is much narrower. For example, if the bottom and top of the 90% prediction interval on a given day are for streamflows of 400 cfs and 750 cfs, respectively, then we say that with 90% probability, streamflow on that day will be between 400 cfs and 750 cfs. There is still a 10% chance that streamflow will fall outside of that range due to very rare events that could occur (a large thunderstorm or 30 consecutive days of hot, dry weather, for example).
If we are concerned about low flows, we can also interpret this interval as meaning that with 95% probability, streamflow will be at least 400 cfs, since the lower band is the 5th percentile. In other words, there is only a 5% chance that streamflow will be lower than 400 cfs. If we are concerned about high flows, we can say that with 95% probability, streamflow will be less than 750 cfs, since the upper band is the 95th percentile. In other words, there is only a 5% chance that streamflow will be higher than 750 cfs on that day.
The model simulates each day between April 18 and September 30, starting with streamflow, reservoir, and diversion conditions as of April 17. Streamflow volumes and runoff timing were predicted using statistical models relating streamflow to winter baseflow and April 1 snow-water equivalent (see streamflow prediction blog for details) and runoff center-of-mass to April 1 snow-water-equivalent and projected April-June temperature (see runoff timing blog for details). The relevant streamflow and timing predictions are given in the tables below, for reference.
These statistical models all contain uncertainty due to unexplained variability in the data. This uncertainty is quantified by well-defined statistical distributions, from which random numbers are drawn in each simulation. Because streamflow volumes and timing in the five subwatersheds needed for the simulations are correlated with one another, these statistical distributions are multivariate, accounting for variability within each subwatershed as well as correlations across the subwatersheds. Thus, the random selections are realistic—if high streamflow and late runoff timing is selected for Fall River, there is a greater chance that streamflow will be higher and runoff later in the other subwatersheds, too. Furthermore, to make sure that hydrographs are realistic, the hydrograph shapes (runoff timing) are selected randomly from actual years that have occurred in the 1978-2016 record, according to the probability that each given year could have resulted from the early-April conditions observed this year.
Once streamflow magnitude and timing are selected for each simulation, the rest of the simulation is driven by a set of management rules and assumptions that are based on the way the irrigation system is currently managed. The basic rules and assumptions are listed below.
We assumed that irrigation diversions from the Henry’s Fork, Fall River, and Teton River will be equal to 1988-2016 averages. This assumption will slightly overestimate diversions, because surface-water diversions in the Henry’s Fork watershed have been steadily declining since the late 1970s due to increased irrigation efficiency. Furthermore, use of the mean diversion rate will simulate a much greater diversion rate than in 2016, when supply was limited and many users ran out of both natural flow and storage by late July. Physical and administrative (i.e., water-rights) shortages are very unlikely to occur in 2017.
River reach gains
The lower reaches of the mainstem Henry’s Fork and Teton River interact with the aquifer and either gain or lose water as a result of this interaction. When and where the river gains water, this is additional streamflow over and above the amounts predicted to flow into the irrigation system in the Henry’s Fork at Ashton, Fall River at Chester, and Teton River upstream of the Crosscut Canal. When and where the river loses water, these losses are subtracted from the amount available to support diversions. The simulations assume 1988-2016 mean gains/losses in the Henry’s Fork between Ashton and St. Anthony and in the Crosscut Canal itself between its point of diversion on the Henry’s Fork at Chester Dam and its injection point into the Teton River. The simulations use 2003-2016 mean gains/losses in the Teton River downstream of the Crosscsut Canal injection point. In an above-average water-supply year like 2017, actual gains are likely to be slightly higher and actual losses slightly lower than the mean values.
A number of groundwater wells can be used to inject water into Fall River, Teton River, and lower Henry’s Fork to offset diversion associated with certain water rights. The model assumes mean 1988-2016 injection of water from these wells. These are generally small amounts of water that are well within model uncertainties in overall water supply.
Outflow from Henry’s Lake is set at its current value of 70 cfs until Henry’s Lake fills. Once the lake is full, outflow matches inflow to hold the lake steady at its full capacity. Once inflow drops below 70 cfs, outflow reverts back to 70 cfs for the remainder of the summer.
Outflow from Grassy Lake is set at its current value of 0 until the lake fills. Once the lake is full, outflow matches inflow to hold the lake steady until release is needed to meet downstream demand on Fall River. In accordance with customary operation of Grassy Lake, this need is triggered when flow in Fall River at the Chester gage falls below 75 cfs, at which time 50 cfs is released from Grassy Lake.
Crosscut Canal delivery to Teton River
The Crosscut Canal diverts water from the Henry’s Fork at Chester Dam for two purposes: 1) to supply water to the Fall River canal system and 2) to deliver Island Park storage water from the Henry’s Fork to the Teton River. The model assumes average diversion rate into the Crosscut Canal for delivery into the Fall River canal system. Need for delivery to the Teton River is triggered when Teton River inflow, less reach losses, is not sufficient to meet diversions on the Teton River, exclusive of the lower North Fork Teton diversions, which are met with return flows and river reach gains. When supply cannot meet diversion rates on the Teton River, water is delivered from the Crosscut to the Teton River. The model assumes that this diversion rate is equal to the shortfall on the Teton River plus any channel losses (or minus any gains) in the Crosscut Canal between Chester Dam and the Teton River.
Island Park Reservoir
Outflow from Island Park Reservoir is initially set at its current value of 800 cfs and then adjusted slightly up or down from this value to ensure that the reservoir fills by May 20, given the simulated inflows. Once the reservoir is full, outflow matches inflow to hold the reservoir steady at its full capacity unless additional release is needed to meet diversion demand downstream. This need was determined by computing actual streamflow in the Henry’s Fork at St. Anthony, after accounting for all irrigation diversions, inflow between Island Park and Ashton, supply in Fall River and Teton River, reach gains/losses, delivery from Grassy Lake, and delivery from the Crosscut Canal to the Teton River. When streamflow in the model is projected to drop below 1,000 cfs at St. Anthony on a given day, Island Park outflow on the preceding day (to account for one-day travel time) was set to the minimum value necessary to keep flow in the Henry’s Fork at St. Anthony at 1,000 cfs, but never lower than inflow to Island Park Reservoir. This last condition assumes that once extra release from Island Park is no longer needed to maintain 1,000 cfs at St. Anthony, the reservoir is held constant for the remainder of the water year. In 2016, a very dry year, the St. Anthony target flow was reduced to 850 cfs, and Island Park outflow was reduced below inflow during the second half of September in order to store as much water as possible before winter. In this model, we did not assume such drastic management to store extra water during September.
Island Park turbidity model
Analysis of our 2016 continuous record of turbidity in the Henry’s Fork immediately downstream of Island Park Dam showed that turbidity is an increasing function of reservoir inflow and proportion of total outflow delivered through the west-side dam gates (versus the power plant). That is, higher inflow and a higher fraction of outflow passing through the west-side gates increased turbidity. Turbidity is a decreasing function of reservoir volume, meaning that as reservoir volume drops, turbidity increases. In addition, turbidity on a given day was highly correlated with that on the previous day. This models the persistence of random events such as cyanobacteria and algae blooms over spans of 3-7 days. These statistical relationships, including the random events, were combined with simulated inflow, fraction of outflow through the gates, and reservoir volume to simulate turbidity. With regards to the fraction of flow passing through the gates, we assumed that the Island Park hydroelectric plant will function properly all summer and use any and all outflow up to its capacity 960 cfs. Thus, the only flow that will pass through the west-side gates is that in excess of 960 cfs.