University of Washington Dept. of Civil and Environmental Engineering Using CPC long lead climate outlooks for ensemble streamflow forecasting Andy Wood and Dennis P. Lettenmaier University of Washington Dept. of Civil and Environmental Engineering Session A24A 2006 Joint Meeting of the AGU Baltimore, MD May 23, 2006
Climate forecast importance: temporal variability In Western US: Jan – April forecasts of summer streamflow are critical for decision-making related to: agriculture environmental flows hydropower navigation water supply Western US Water Cycle Climate Forecasts Importance Monthly Timestep
Climate forecast importance: spatial variability Most basins east of the Sierras and Cascade Mtns are heavily influenced by spring precipitation. Water supply forecasts there have unavoidably high uncertainty because spring precipitation is relatively unknown. Wet spring 65% Apr-Jun Oct-Jun PCP Dry spring 15% Courtesy of Tom Pagano, NRCS
Climate forecast importance: spatial variability Example: Climate forecasts relatively unimportant by late Winter Areas with dry spring …. Summer flow forecast skill Courtesy of Tom Pagano, NRCS Wet spring 65% Forecast Skill Precip Apr-Jun Oct-Jun Low High Dry spring 15%
Climate forecast importance: spatial variability Example: Climate Forecasts very important through Spring Areas with wet spring …. Summer flow forecast skill Courtesy of Tom Pagano, NRCS Wet spring 65% Forecast Skill Precip Apr-Jun Oct-Jun Low High Dry spring 15%
Background Current Practice for Western US Streamflow Forecasting combine: (1) estimate of current hydrologic state (2) forecast of historical climate…usually* produce: streamflow forecast with uncertainty information UPPER HUMBOLDT RIVER BASIN Streamflow Forecasts - May 1, 2003 <==== Drier === Future Conditions === Wetter ====> Forecast Pt ============ Chance of Exceeding * =========== Forecast 90% 70% 50% (Most Prob) 30% 10% 30 Yr Avg Period (1000AF) (% AVG.) MARY'S R nr Deeth, Nv APR-JUL 12.3 18.7 23 59 27 34 39 MAY-JUL 4.5 11.3 16.0 55 21 28 29
Research Objective Current Practice for Western US Streamflow Forecasting combine: (1) estimate of current hydrologic state (2) forecast of historical climate CPC Outlook produce: streamflow forecast with uncertainty information ICs Spin-up Forecast obs recently observed meteorological data ensemble of met. data to generate forecast ESP-type forecast method hydrologic state We use a hydrologic model-based approach similar to the NWS River Forecast Center’s Ensemble Streamflow Prediction (ESP)
NWS Climate Prediction Center (CPC) Seasonal Outlooks e.g., precipitation
CPC Seasonal Outlook Use Challenge: Seasonal (3-month) probabilities must be converted to daily meteorological values at the scale of the hydrology model
CPC Seasonal Outlook Use spatial unit for raw forecasts is the Climate Division (102 for U.S.) CDFs defined by 13 percentile values (0.025 - 0.975) for P and T, and μ and σ
Hydrologic Prediction using CPC Seasonal Outlooks CD scale CPC climate outlooks variables: mean temperature (Tavg) total precipitation (Ptot) scales: 102 climate division (CD) / US overlapping 3-month timestep information: forecast (μ, σ) at each timestep normal (μ, σ) at each timestep disaggregate spatially climate division unit --- becomes --- 1/8 degree (~12-13 km) disaggregate to a daily timestep 1/8 degree monthly Tavg and Ptot --- becomes --- 1/8 degree daily Ptot, Tmin and Tmax create Tavg & Ptot ensemble forecasts (μ, σ) at each timestep/CD generate seasonal ensemble data disaggregate temporally overlapping 3-month timestep --- becomes --- non-overlapping 1-month timestep Use CPC forecasts as inputs to a hydrologic model to produce streamflow forecast ensembles link Tavg & Ptot ensembles Associate monthly variables spatially & temporally
disaggregate temporally overlapping 3-month timestep --- becomes --- non-overlapping 1-month timestep Several methods of doing this work well but not perfectly. Schneider et al., Weather & Forecasting (2005) – applied monthly/seasonal mean correction factors – approach being adopted by CPC We are trying multiple linear regression: monthly values = f(seasonal values)
disaggregate temporally overlapping 3-month timestep --- becomes --- non-overlapping 1-month timestep Sample Results ML regression approach appears to yield better variance, but is not markedly superior ML regression approach CPC approach std dev Schneider et al. (2005)
disaggregate temporally overlapping 3-month timestep --- becomes --- non-overlapping 1-month timestep Sample Results ML regression approach CPC approach R = 0.80 Schneider et al. (2005)
link Tavg & Ptot ensembles Associate monthly variables spatially & temporally Challenge: Given monthly distributions for a climate variable, how do you associate the values in time to yield a single sequence of one variable? Of two variables?
link Tavg & Ptot ensembles Associate monthly variables spatially & temporally Challenge: Given monthly distributions in adjacent cells, how might sequences in one climate division be associated with those in another?
link Tavg & Ptot ensembles Clark et al., J. of Hydromet (2004) Schaake Shuffle link Tavg & Ptot ensembles Associate monthly variables spatially & temporally Clark et al., J. of Hydromet (2004)
Spatial and Temporal Downscaling disaggregate spatially climate division unit --- becomes --- 1/8 degree (~12-13 km) Spatial sampling of anomalies within climate divisions disaggregate to a daily timestep 1/8 degree monthly Tavg and Ptot --- becomes --- 1/8 degree daily Ptot, Tmin and Tmax Re-sampling of daily patterns Scaling/shifting to reproduce CPC forecast anomalies ‘OBS’ Another new feature is that we’re now plotting up several analyses of snow observations, and these update on a daily basis. We’ve been automatically downloading the data for a long time for use in our assimilation, and the goal here was to show the west-wide conditions at a single glance, something that’s hard to find elsewhere. Note, in addition to the NRCS snotel points, we also have the California DWR snow pillows, and the Env. Canada snow pillows in the Columbia R. drainage. There are about 5 plots – some of which are for changes during the last week or two. downscaled
University of Washington Forecast System Website project led by Dennis Lettenmaier funded by NOAA, NASA Another new feature is that we’re now plotting up several analyses of snow observations, and these update on a daily basis. We’ve been automatically downloading the data for a long time for use in our assimilation, and the goal here was to show the west-wide conditions at a single glance, something that’s hard to find elsewhere. Note, in addition to the NRCS snotel points, we also have the California DWR snow pillows, and the Env. Canada snow pillows in the Columbia R. drainage. There are about 5 plots – some of which are for changes during the last week or two.
Streamflow Forecast Results: Westwide at a Glance This bubble plot shows the streamflow outlook for summer runoff for about 90 locations in the domain. The anomalies are consistent with those shown in the spatial plots, with the lowest outlooks for the SW streams nearest that very low SM pattern we saw 2 slides back, and normal outlooks in the PNW. Note, this west-at-a-glance display, with both mouse-overs that show various anomalies for the locations, and clickable points that launch more details, is something we have only recently added.
Streamflow Forecast Details Clicking the stream flow forecast map also accesses current basin-averaged conditions Streamflow Forecast Details Flow location maps give access to monthly hydrograph plots, and also to raw forecast data. In addition to the streamflow hydrographs that we’ve had for a while, the clickable streamflow map now brings up the current water year conditions for P,T,SM,SWE, RO – which are helpful in showing where we are with respect to climatology. These are averaged over the drainage basin contributing to streamflow at each location.
Streamflow Forecast Results: Spatial Precip Temp SWE Runoff Soil Moisture Apr-06 May-06 Jun-06
UW Real-time Daily Nowcast SM, SWE (RO) ½ degree VIC implementation Free running since last June Uses data feed from NOAA ACIS server “Browsable” Archive, 1915-present Another area of current research relates to the surface water monitor developed last year by A. Wood. This system, applied at coarse (1/2 degree) resolution over the entire CONUS, is completely automated (free-running) and updates every day. It’s just a prototype, demo project that have been unable to get funding to extend, and the main products are maps of current soil moisture & SWE, and an monthly archive that extends back to 1915, that also has SM & SWE maps. Anyway, we are now adapting the daily update approach for use in the westwide forecast system, and should have the first basin (PNW) land surface conditions updating daily (at 1/8 degree) within the month. After that we’ll move on to other basins, and probably extend the 1/8 nowcast eastward to the Mississippi R. We are currently migrating the CPC forecast approach to a national US implementation
http://www.hydro.washington.edu / forecast / westwide / Conclusions Our current approach for downscaling CPC seasonal outlooks is adequate from hydrologic perspective. Simple temporal disaggregation approaches are sufficent, although it’s possible that slightly higher performance can be achieved via more elaborate disaggregation methods Ensemble formation step bears further analysis at the monthly to seasonal time scale. Translation of CPC outlooks to ensembles for hydrologic forecasting should not be an obstacle for their use. For more information: http://www.hydro.washington.edu / forecast / westwide /
Thank You