Water Supply Forecasts

A Field Office Guide for Interpreting Streamflow Forecasts

Section 2 - Question and Answer

How are snow survey data used in water supply forecasting?

About 75 percent of the spring and summer streamflow produced in the West originates from mountain snowpacks. The other 25 percent is primarily a result of spring and summer rains. Because of the relationship between snowpack and streamflow volumes, measurements of snow water content provide a dependable basis for estimating streamflows. SNOTEL data and manually measured snow course information are available to the hydrologists through CFS for use in preparing water supply forecasts.

What is snow water content (snow water equivalent)? How is this different from snow depth?

Snow water content is the depth of water that would be obtained by melting the snowpack. Snow depth is the measurement from the top of the snow to ground level. The amount of water produced by a snowpack of a given depth varies depending on the density of the snowpack.

Why do the forecasts change from month to month?

Forecasts change from month to month because snow and precipitation do not occur at a uniform rate throughout the year. As snowpack and precipitation accumulate, the forecasts change to reflect the current conditions. Monthly updates usually result in more accurate forecasts as the year progresses.

How are the NRCS forecasts developed? How do the hydrologists at the National Water and Climate Center (NWCC) determine which sites to use in the forecast equations?

Forecast procedure development begins with data site selection and an assessment of data quality and availability. The streamflow data for the forecast point are reviewed for correctness and must have a long enough record to provide an adequate statistical sample. Then, data collection sites (usually snow and precipitation) within hydrologic proximity to the basin are reviewed for completeness, correctness, length of record, and availability at forecast time. Once this is done, a set of candidate variables is selected, and the data are prepared for the regression analysis.

In the past, standard multiple regression analysis was used to develop a forecasting equation. Selection of sites to enter the equation from the list of candidates was done by professional judgement and trial and error. To make the development of a regression equation more objective and to be more assured of obtaining the optimal combination of variables, the hydrologists at the NWCC have developed a computer program that identifies the best combinations of variables for forecasting from a set of candidate variables. It employs advanced techniques not commonly used in forecast procedure development until now. These techniques are as follows:

Pseudo-all Combinations. The only completely foolproof way to obtain the best combinations of variables is to compute regressions for all possible combinations of the variables. The time and effort needed to do this makes it unfeasible. Instead, a pseudo-all combinations technique was developed to find the best combinations of variables that evaluates only a fraction of the possible combinations. The technique is not perfect, but experience has shown that it does a very good job of identifying optimal or near-optimal combinations of variables.

Principal Components Analysis. This is a statistical technique that objectively accounts for the intercorrelation among the input variables. Standard regression without principal components does not behave well when the input variables are highly correlated among themselves. Yet, it is often desirable to include several such variables in an equation (for example, several snow courses instead of just one or two.)

Principal components analysis restructures the input data to account for this intercorrelation. When the regression analysis is performed, the resulting equation is generally more stable and performs better than an equation produced by standard regression with the same variables. Principal components regression extracts the information relevant to streamflow more efficiently than standard regression.

Cross-Validation or "Jackknife Test". Standard error is an indication of a regression equation's accuracy during the development period, but it can be an overly optimistic indication of the equation's forecasting performance. A better test of its performance in a forecast situation is to use a portion of the data to develop the equation and use the remaining data to test its performance.

Cross-validation does this in a systematic way. One year is withheld from the set of years used for a given combination of variables, and a regression equation is calculated. Using this equation, a forecast is made for the withheld year. The withheld year is placed back into the data set, the next year is withheld, an equation is calculated, and a forecast is made. This process is repeated until a forecast has been made for all the years available. A standard error can then be calculated from the errors of the forecasts compared to the observed flows.

This process simulates an actual forecasting situation, and is considered a good way to evaluate the forecasting potential of an equation. Cross-validation is done for each combination of variables examined in the pseudo-all combination procedure.

The use of these techniques represents a major step forward in water supply forecasting. They provide (1) more accurate forecasts, (2) a truer evaluation of forecast error, and (3) a clear picture of the data sites required for forecasting.

There are five forecasts for each point; which forecast should I tell my cooperators to use?

You shouldn’t tell your cooperators which forecast to use. Explain what each of the five forecasts represents, then let them decide which forecast to use, based on operational needs and the degree of acceptable risk. You should, however, remind them that, given current conditions, the most probable forecast is still the best estimate.

What do I do when a user requests information for a station we don’t forecast?

At the current time, there is no hard-and-fast policy on the establishment of new forecast points. Written requests from cooperators to the local conservation district for new forecast points should be forwarded to the state water supply staff through the field office. Until you’ve received approval from the state water supply staff and the Snow Survey Program Manager, commitments to forecast a new point should not be made. Responsibility for forecasting new points has to be reviewed with the National Weather Service.

If a user needs to relate forecast information to a nearby delivery point, contact the state water supply staff to develop a "forecast localization" technique. This will involve collecting historical flow data for that delivery point and developing a relationship between the delivery point and an upstream forecast point.

What is the difference between natural (adjusted) flows and observed flows?

Observed flows are the flows measured at a given point on a stream, regardless of the effect of upstream water management on streamflows. Natural (adjusted) flows are the best estimate of flows that would have occurred without human influence. Natural (adjusted) flow is calculated by adjusting observed flows for changes in storage and gaged diversions that affect streamflow volumes. Unless otherwise noted, streamflow forecasts are for natural (adjusted) flows.

What is a drought?

According to Webster's "New Collegiate Dictionary", a drought is a "prolonged period of dryness". One of the most common questions asked during a dry year is, "Are we in a drought?" How do you respond to that question? Unless a drought has been officially declared, you don’t. Drought conditions can mean one thing to one person and something completely different to another. For example, unusually low winter snow and minimal spring and summer precipitation in a wheat-growing area might be considered a drought by a dryland farmer, but to someone with an irrigated operation and an adequate supply of water in storage, it may not be.

The governor is the elected official who declares a drought in a state or portion of a state. Within Federal projects, the management agency will declare a drought. Because of the serious economic and social consequences involved, be careful when using the "d------" word. Instead, encourage users to concentrate on the information provided in streamflow forecasts, indices such as the Surface Water Supply Index (SWSI) and Palmer Drought Index (PDI), and comparisons with previous years.

What is the role of the National Weather Service in water supply forecasting?

The National Weather Service (NWS), like the Natural Resources Conservation Service, is responsible for providing streamflow volume forecasts. Generally, the NWS is responsible for providing forecast information on the downstream reaches of larger streams, while the NRCS, because of its relationship with the agricultural community, forecasts points in the smaller basins and the upstream reaches of larger streams. Hydrologists from both agencies compare their forecasts to provide a single "coordinated" forecast to water users.

Along with streamflow volume forecasts, the NWS also provides flood forecast information. The responsibility for flood forecasting belongs solely to the NWS; any questions you get concerning that subject should be referred to them.

Why are stream gages important for water supply forecasting?

Streamflow gages are very important because the measurements taken at these sites are needed to develop and verify the accuracy of forecast equations. Without historical streamflow information from stream gages, hydrologists could not determine the relationship between streamflow volume and hydrologic and weather-related variables. Even if an equation existed, hydrologists could not evaluate the effectiveness of the forecast values without some type of measurement at the forecast point. Most stream gages are operated by the U.S. Geological Survey.

How can I find out more about the Centralized Forecast System?

Contact the state water supply staff, have them send you any materials you may need to introduce you to CFS, and then arrange for them to provide any additional training you may need. Use the CFS tutorial to simulate sessions on the system and access the system as needed--there is no substitute for "hands on" experience when dealing with any computer facility. Much of the system is menu driven and can be very easy to use once you are familiar with it. Use the Problem/Enhancement Logs ("Bug Sheets") to let state office and NWCC water supply staffs know of any problems you encounter on the system or to suggest any improvements.

What is a Reservoir Operation Guide?

Reservoir Operation Guides (ROG) are a decision support tool used to help reservoir operators manage their facilities by using streamflow forecasts. Using a process called Reservoir Storage Volume Planning (RSVP), the guides provide a means by which the operator can optimize water use, while keeping flood damages to a minimum. A ROG can be prepared for any appropriate reservoir upon receipt of a written request from the reservoir operator or owner. Requests are to be reviewed by the local soil conservation district before forwarding them on to the state conservationist for approval. For more information on ROGs, ask the state water supply staff for a copy of Technical Release (TR) 75, which outlines the RSVP process.

How can I answer questions about comparing the current year’s conditions to previous years?

Information comparing the current year’s conditions to last year’s conditions is available in many of the basin analysis routines in the WYFOR (Water Year FORecast) section of CFS. Another routine, called COLSUM (COLumnar SUMmary of monthly data with frequency analysis), provides the opportunity to rank monthly data in ascending or descending order for a particular site. The COLSUM routine can be found in the DBQ (Data Base Query) section of CFS.

The example in Figure 4 shows an analysis of seasonal precipitation (October through March) at a SNOTEL (SNOwpack TELemetry) site. Water Year 1988 was the driest in the 19-year record. Water Year 1983 was the wettest. NWS climate stations usually have a longer period of record.

 

Gage No. Station : S042,MARLETTE LAKE,NV -------- ------- for the months Oct - Mar Ranks Water Value Weibull Value # Year inches plot pos. log10 ----- ---- --------- --------- ------- 1 88 14.40 0.050 1.158 2 87 15.30 0.100 1.185 3 91 16.20 0.150 1.210 4 92 16.90 0.200 1.228 5 90 17.10 0.250 1.233 6 94 18.00 0.300 1.255 7 81 18.40 0.350 1.265 8 79 24.20 0.400 1.384 9 85 24.60 0.450 1.391 10 89 26.90 0.500 1.430 11 84 33.20 0.550 1.521 12 96 34.10 0.600 1.533 13 93 37.60 0.650 1.575 14 80 38.10 0.700 1.581 15 95 40.10 0.750 1.603 16 97 40.70 0.800 1.610 17 86 43.50 0.850 1.638 18 82 45.70 0.900 1.660 19 83 46.50 0.950 1.667 number of values = 19 arithmetic average = 29.03 log average = 1.428 std. deviation = 11.52 log std. dev. = 0.183 coeff of variation = 0.40 log cof. var. = 0.128 arithmetic skew = 0.141 log skew = -0.136 Exceed Return Normal Log Normal LP3 % Prob Period Probability Probability Probability ------ ------ ----------- ----------- ----------- 99.00 1.010 2.23 10.06 9.65 95.00 1.053 10.08 13.40 13.19 90.00 1.111 14.26 15.61 15.52 80.00 1.250 19.33 18.79 18.85 50.00 2.000 29.03 26.77 27.03 20.00 5.000 38.73 38.16 38.25 10.00 10.000 43.79 45.92 45.61 5.00 20.000 47.98 53.50 52.61 1.00 100.000 55.82 71.26 68.32

Figure 4. Example of COLSUM output for October-March precipitation at Marlette Lake, NV.

If I know what the snowpack conditions are like, why do I need streamflow forecasts to evaluate my future water supplies?

Although snow water content is a major contributing factor, other components in the watershed runoff process, such as rainfall and soil moisture conditions, affect the amount of spring and summer streamflows. For example, in 1984 and 1989, the Truckee River Basin reported 100% of the average snowpacks during April, yet the resulting streamflows were 104% and 73% of average, respectively, a difference of over 30%.

Streamflow forecasts, which consider snowpack plus other factors related to runoff, provide a better overall picture of streamflow conditions than just using snowpack. This is true whether the user wants specific flow values or an indication of general conditions.

How does NRCS determine the quality of forecasts being produced?

An analysis routine called Forecast Error Analysis Routine (FEAR) is used to determine the accuracy of seasonal streamflow volume forecasts. The routine verifies the performance of the forecasts and helps the hydrologists determine if any adjustments are needed.

Some states hold annual reviews with water users to review snow survey and water supply related activities and discuss forecast performance and development in order to better address their needs.