Recent Analysis and Improvements of the Statistical Water Supply Forecasts for the Upper Klamath Lake Basin, Oregon and California, USA Jolyne Lea and David Garen, USDA/NRCS, Adam Kennedy, Portland State University, and John Risley, USGS WORKSHOP ON MANAGEMENT OF NATURAL AND ENVIRONMENTAL RESOURCES FOR SUSTAINABLE AGRICULURAL DEVELOPMENT
USDA NRCS Snow Survey and Water Supply Forecasting The Snow Survey and Water Supply forecasting program began in the early 19s. This was an effort to predict spring and summer water supplies from the winter snowpack. Early snow surveys were done in the Sierra Nevada and Cascade mountains and engineers determined the relationship between snowpack and streamflow using graphical and regression techniques.
Surprise Lakes SNOTEL site. A remote climate and snow station.
History of Forecasting USDA Natural Resources Conservation Service (NRCS) formerly the Soil Conservation Service (SCS) has been forecasting water supplies in the Klamath Basin since the 193s. We also forecast at 75 other stream gage stations. Due to the seasonal nature of precipitation in the Western US, forecasts were initialized to determine the amount of streamflow runoff an area would have from the snowpack water content, or snow water equivalent (SWE). Forecasts in the West have traditionally been requested and used by water managers for irrigation, power generation, municipal use and more.
What are Statistical Water Supply Forecasts? A water supply forecast has traditionally been a volume of water expected into a lake, reservoir or past a stream gage in a multi-month time step or season. Statistical Forecasts are also made for annual events such as peak flow and date, and recession (low flow) date and stage height. Seasonal Water Supply Forecasts are used for water management decision making for irrigation, municipal use, wildlife and fish, hydropower, and recreation. And is often an input into daily water management models.
Klamath Basin Competing Water Needs The Klamath Basin forecasts have been provided to the Bureau of Reclamation and PacifiCorp Hydropower Company for many years. In the 199s, Suckers in Upper Klamath Lake were determined to be endangered species (USFWS). Coho in the Klamath River was listed as threatened (NOAA). Water management plans were developed to improve these fish populations. In 21, the 5 th driest year on record, there was very limited water for the irrigation, power and endangered fish needs. Based on the April 1 water supply forecast, to comply with the endangered species laws, no irrigation water would be available for the farmers in the Bureau of Reclamation Project, though others farmers in the area were able to irrigate. This decision caused many protests throughout the local, state and national agricultural community and received the President of the United State s attention. The resulting Federal funds were used for emergency well drilling, loans and disaster declarations.
Impacts to Klamath Basin Water Supply Forecasting Funding for many water conservation projects. Increased funding for hydrologic gages and climate data collection. Study on the accuracy of the statistical Water Supply Forecasts and ways to improve them. Project Funding for improved forecast models. Increased education of the many stakeholders on water supply forecasts.
Photos courtesy of the Bureau of Reclamation.
Threatened and Endangered Fish Coho Salmon Photo: NOAA Shortnose and Lost River Suckers
Klamath Basin Forecast Locations
Klamath Basin Forecast Data 5 - USGS Stream gages 19 - SNOTEL 2 - Enhanced SNOTEL 6 - Snow Courses 4 - Aerial Markers 1 - SCAN Site 5 - NWS COOP stations BOR Agrimet stations and others.
Water Supply Forecast Frequency and Periods Issued monthly, January through June. Mid- month updates December 15 June 15. Forecasted periods are Feb- Jul, Mar- Jul, Apr- Sept, May- Sept, Jun- Sept. The forecast periods developed based on user need, and the US Bureau of Reclamation relies primarily on Apr- Sept. volumes. Other forecasts periods have been requested by a hydropower utility, PacifiCorp.
Water Supply Forecast Statistical Model Calibration Linear regression of monthly hydroclimatic input variables (SWE, PRCP, groundwater, etc.) against observed streamflow volume. Regression coefficients are calculated between selected station input variables and observed streamflow volumes.
Water Supply Statistical Model Principal Components statistical model used since the early 199s by Dr. David Garen. This technique takes into account the intercorrelation among predictor variables. Jackknife testing for calibration fit. An iterative test where each year in turn is removed and then modeled with the remaining years. 2 to 4 years of data for optimum statistical fit.
Water Supply Forecast Equation Variables Data is required to be operationally available by the 3 rd working day of the month. Snow water equivalent (SWE) Snow courses/aerial markers. SNOTEL sites NWS and SNOTEL Precipitation (fall and spring). USGS and BOR Streamflow data and diversion data Reservoir Inflow Data Well level data Temperature Trans-Niño Index (TNI).
Water Supply Forecast Process Only long term observed data are used. No future conditions are included, e.g. no future snow or spring and summer precipitation are assumed. Each forecast month is treated independently, Jan, Feb, Mar, Apr, May. No memory of previous forecast. Implicit with the predictor variables carried throughout the year. Produces statistically representative probability distributions of forecast volume based on calibrations.
Forecast Improvements The updating of forecasts is ongoing. Additional years of data and variables have been researched and incorporated with improved correlations. Additional statistical routines were explored (artificial neural network). To date, they have not provided a significantly improved statistical model for forecasting in the Klamath Basin.
USGS Artificial Neural Network Models A flexible mathematical structure capable of describing complex nonlinear relationships between input and output data sets that are typically found in natural systems
Hidden layer feed forward ANN h j = tanh [ X i 1 w ij + 1 b j ] i where h j is the computed output from hidden layer node, j is the hidden-layer node index, tanh is the hyperbolic tangent, i is the input layer node index, X i is the input variable, 1 w ij is the hidden layer weight, and 1 b j is hidden-layer bias Y = h j 2 w j + 2 b j where Y is the output variable 2 w j is the output layer weights, and 2 b is the output layer bias.
New Variables Improved Accuracy Variables limited to current climate (post 1977 climate shift). 1981-24. New Variables added: Groundwater components: Spring flow data at Fall River. Well level data. Mean monthly temperature from Crater Lake N.P. Areal mean monthly precipitation Trans Niño Index.
Well and Spring Data Groundwater a big component in the basin hydrology due to its volcanic nature. Well data not affected by pumping. Spring data from US Geological Survey Oregon Water Resources Dept. Stream gage.
State of Oregon Well locations Long Term Observation Well location
46 6 6 5 Preliminary Water-Table Map of the Upper Klamath Basin 5 46 45 5 42 47 434 42 43 42 5 5 5 42 42 5 48 42 44 41 41 35 4 44 5 46 4 41
Principal Ground-Water Discharge Areas.28-1.4 m3/s 1.4-2.8 m3/s 2.8-5.7 m3/s 5.7-8.5 m3/s About 71 m3/s of ground - water discharge has been identified, 51 m3/s is to the lake and its tributaries.
Upper Klamath Basin Annual Precipitation (The principal source of ground- water recharge) > 127 mm 635-127 mm 38-635 mm < 38 mm Total average precipitation: about 12 m 3 /yr
Water-table elevation fluctuations follow a pattern similar to the cumulative departure from average precipitation. 3 Depth to water, in meters 1 2 3 4 5 Depth to Water in Well 36S/14E-25BCB Cumulatuve Departure from Average Precipitation at Crater Lake 2 1-1 Cumulative departure from average, in millimeters 6 196 1965 197 1975 198 1985 199 1995 2 25-2
Climate Indices Standard Southern Oscillation Index (SOI) not correlated well. Pacific Decadal Oscillation (PDO) not correlated well. Trans-Niño Index (TNI) - Since 1977, a good correlation between this index and the Klamath Basin.
Winter sea surface temperature, wind pattern (arrows), and sea level pressure (contour) anomalies for the warm (left) and cool (right) phase of the Pacific Decadal Oscillation. (Courtesy of Nate Mantua, JISAO Univ. of Washington.)
Trans-Niño Index (TNI) The TNI is the standardized equatorial sea surface temperature (SST) gradient between Niño 1+2 and Niño 4 regions (Trenberth, 2).
Mean Areal Precipitation PRISM Derived Mean Areal Precipitation and temperature. Used 5 to 6 SNOTEL and NWS COOP stations.
Mean Areal Precipitation Obtain monthly gridded base data from Oregon Climate Service s ftp site. Define area of interest with a zonal grid. Employ automated script for multi-step processing of base layers. Gridded data available at: http://www.ocs.oregonstate.edu/index.html
Mean Areal Precipitation Upper Klamath Lake Williamson Sprague Arithmetic Average Processing March 23 Sprague = 54.87 mm Williamson = 51.14 mm UKL = 75.33 mm
Temperature and Groundwater Variables Standard Error Improvement
No Improvement is shown in the Sprague R. 45 Sprague River Standard Error Improvement with Temperature and Groundwater Jacknife Standard Error (1 3 m 3 ) 4 35 3 25 2 15 1 5 JAN FEB MA R A PR MAY JUN avg NEW SE + NEW SE - Old SE+ Old SE -
Williamson River Improvement with Groundwater 7 Williamson River Standard Error Improvement with Temperature and Groundwater Jacknife Standard Error (1 3 m 3 ) 6 5 4 3 2 1 JAN FEB MAR APR MAY JUN avg NEW SE + NEW SE - Old SE+ Old SE -
Upper Klamath Lake Improvement with Temperature 9 Upper Klamath Lake Inflow Standard Error Improvement with temperature and Groundwater Jacknife Standard Error (1 3 m 3 ) 8 7 6 5 4 3 2 1 JAN FEB MAR APR MAY JUN avg NEW SE + NEW SE - Old SE+ Old SE -
TNI and Areal Precipitation Standard Error Improvement
The Sprague River forecast with TNI reduces forecast uncertainty. 12 1. Jackknife Standard Error (1 6 m 3 ) 1 8 6 4 JSE w/ TNI JSE w/o TNI r-val. w/ TNI r-val. w/o TNI.95.9.85.8.75.7 Jackknife r-value 2 January Febuary March April May.65 Forecast Month
The Upper Williamson with TNI reduces early season forecast uncertainty. Jackknife Standard Error (1 6 m 3 ) 4 35 3 25 2 15 1 JSE w/ TNI JSE w/o TNI r-val. w/ TNI r-val. w/o TNI 1..95.9.85.8 Jackknife r-value 5.75 January Febuary March April May Forecast Month
Forecast Improvement Results TNI alone provided great improvement in early season accuracy. Crater Lake temperatures provided improvements in the Upper Klamath Lake late season forecasts. Well data and spring data improved overall accuracy in some forecasts.
Additional Variables Under Consideration. Soil Moisture data from SNOTEL when more years of data are available. Additional Spring Temperature data from SNOTEL sites. Other climate variables. GIS data Applying PRISM Mean Areal Precipitation to additional forecast points. Total Basin SWE from NWS NOHRSC.
March 24 Continued difficulty in forecasting the chaotic component
Forecast Limitations. Statistical forecasts do not provide daily products. Limit to how far in advance forecasts can be made with current science and data.
Future Statistical procedures will continue to be utilized to forecast water supplies in the Western US. Provide fairly accurate seasonal volume forecasts. Use monthly and sub-monthly quality controlled data. Optimal for limited program resources. Limited by available long term data.
Upper Klamath Lake Reservoir is a principal source of water for the Klamath project. The reservoir has a capacity of 1,76,829,66 cubic meters and is operated by Pacific Corp., subject to Klamath Project rights. Upper Klamath Lake (photo courtesy of the Oregon Department of Environmental Quality)
Crater Lake - Crater Lake National Park.
TNI = (Niño 1+2 N - Niño 4 N ) S,N where N indicates normalized and S smoothed Regression of global SST anomalies with TNI for 19-1976 in C. Values exceeding.1 C are hatched and less than -.1 C are stippled. The contours are ±.5 C, ±.1 C, ±.15 C, etc. http://www.cgd.ucar.edu/cas/catalog/climind/tni_n34/
Fig. 1. Time series of SSTs from Niño 3.4 (N3.4 top), and TNI (bottom) from 1871 to 2. Both time series are normalized by the standard deviation s, as given. In the top panel, shading is included to show where thesholds of ±.4 C are exceeded, indicating ENSO events (see Trenberth, 1997). A low pass spline fit is given to highlight interdecadal variations. http://www.cgd.ucar.edu/cas/catalog/climind/tni_n34/
Pacific Decadal Oscillation (PDO) PDO reflects slow decadal variability in the North Pacific Ocean PDO Index = Coastal SST - mid ocean SST In the Pacific Northwest: Positive PDO--warm and dry periods Negative PDO--cool and wet periods Positive PDO Phase Sea-surface Temperatures
PDO has no correlation with Klamath Basin Precipitation or Streamflow La Nina/PDO combined influences on past US Precipitation, from Gershunov, Barnett, and Cayan 1999 (EOS)
TNI Computations (revised for operational use) Obtain Niño 1+2 and Niño 4 area average monthly SST series. Compute monthly means over the complete period of record (1951-24). Standardize both series by subtracting the monthly means and dividing by the standard deviation of the complete period of record. Subtract standardized Niño 4 from standardized Niño 1+2 to obtain monthly TNI series. Perform correlations with the variables of interest to identify months which contain the predictive signal. Compute a multi- month TNI by averaging months well correlated to selected variables. Divide multi- month TNI by its respective multi- month standard deviation. TNI=(Niño 1+2 std - Niño 4 std ) avg Revised from: Trenberth and Stepaniak, 21