Transcript Yakima R. Application
Real-time seasonal hydrologic forecasting for the Western U.S.: Recent Progress
Andrew Wood
University of Washington JISAO weekly seminar Seattle, WA June 10, 2003
Topics
1. Overview of Approach
a) climate model based forecasts b) ESP (and ESP-composite) forecasts
2. Current Columbia River basin forecasts
a) spin-up approach b) snow initialization
3. Retrospective West-wide forecast skill analysis 4. Next Steps
Overview: Background
•
Previously
, demonstrated an approach for combining seasonal climate model forecasts with hydrologic simulation to create hydrologic forecasts • Climate Model: NCEP Global Spectral Model (GSM), for • Hydrology Model: VIC (at 1/8 or 1/4 degree resolution) • Real-time experimental applications: • East Coast (Spring/Summer 2000, November 1997) (paper: Wood et al. (2001), JGR) • Columbia R. basin (Spring/Summer 2001) • Alternate applications: • DOE Parallel Climate Model (PCM) climate change analyses for CRB, California and Colorado; as well as a downscaling method comparison (papers: Wood et al. (2003) and 3 others, Climatic Change) •
Currently
, supported by several funding sources to implement seasonal forecast approach for the Western U.S., in real-time, “pseudo-operationally”. Support: • NASA NSIPP • IRI/ARCS • NOAA GCIP/GAPP
Overview: Project Goals
• Implement the hydrologic forecasting approach over the western U.S. domain. • Move from aperiodic experimental forecasts to quasi-operational forecast products.
• Calibrate streamflow forecast points throughout the domain, and identify potentially associated uses and users.
• Use the NCEP GSM retrospective ensemble climatology to assess streamflow forecast accuracy • For routine implementation over the large domain, automate various processing steps through enhancements to existing software.
• Identify and evaluate potential real time data sources for use in hydrology model initialization (spin-up), a critical factor for forecast accuracy. • Standardize ongoing retrospective efforts to verify recent forecasts and diagnose prediction accuracy and identify sources of error. • Enhance our existing web site for disseminating forecasts and forecast retrospective evaluation results, and will forge links to interested operating agencies.
• Investigate approaches to making the forecast results available to these communities other than water management.
1-2 years back
Overview: VIC Simulations
start of month 0 end of mon 6-12 forecast ensemble(s) model spin-up climatology ensemble data sources NCDC met. station obs. up to 2-4 months from current LDAS/other real-time met. forcings for remaining spin-up climate forecast information snow state information
Forecast Products streamflow soil moisture runoff snowpack derived products
Overview: Hydrologic Forecast Approach
1.
climate forecast
• ~2-3 degree resolution (T42-T62) • monthly total P, avg T
hydrologic model inputs climate model ensemble outputs
• 1/8-1/4 degree resolution • daily P, Tmin, Tmax
streamflow, soil moisture, snowpack, runoff, derived 2.
products
Use 2 step approach: 1) statistical bias correction 2) downscaling
Extended (“Ensemble) Streamflow Prediction (ESP)
• 1/8-1/4 degree resolution • daily P, Tmin, Tmax No bias-correction or downscaling needed Met. traces can be composited before/after ensemble is run to represent ENSO, PDO conditions, etc.
Overview: Global Spectral Model (GSM) ensemble forecasts from NCEP/EMC
• forecast ensembles available near beginning of each month, extend
6 months
beginning in following month • each month: •
210
ensemble members define GSM
climatology
for monthly Ptot & Tavg •
20
ensemble members define GSM
forecast
Overview: VIC Hydrologic Model
Overview: Bias Correction
Overview:
climate model forecast processing sequence 30 25 20 15 10 5 0 0
Probability
1
T OBS T GSM a.
b.
c.
a) bias correction: climate model climatology observed climatology b) spatial interpolation: e.g., GSM (1.8-1.9 deg.) VIC (1/8 deg) c) temporal disaggregation (via resampling of observed patterns): monthly daily
Current Forecasts
Updates Dec 28, 2002 Jan 15, 2003 Feb 1 Feb 15 Mar 1 Mar 16 Apr 1
Current Forecasts: Spin-up approach
Problem:
For most recent 2-2.5 months, meteorological data availability is poor 1-2 years before start date: 2-2.5 months prior to start: dense station network (~1000), forcings consistent with those used in model calibration, etc. (mostly COOP stations) coarse reporting network (~150 stns) density lower in Canada. (some COOP stns)
Solutions
: > LDAS 1/8 degree real-time forcings from NOAA/NASA > Index station method: combine coarse network signals with dense station climatology
Current Forecasts:
Spin-up approach, Index Stn Method
1. interpolate daily anomalies from index stations over domain 2. apply the result to daily averages taken from the dense station-derived climatology
Current Forecasts:
Spin-up approach, Index Stn Method
Example for daily precipitation Index stn pcp pcp anomaly gridded to 1/8 degree 1/8 degree pcp 1/8 degree dense station daily average
Current Forecasts: Initial snow state assimilation
Problem:
index station method incurs some systematic errors, but snow state estimation is critical
Solution:
use SWE observations (from the 600+ station USDA/NRCS SNOTEL network and several ASP stations in BC, Canada, run by Environment Canada) to adjust snow state at the forecast start date
Current Forecasts: Initial snow state assimilation
Assimilation Method
• weight station obs.’ influence over VIC cell based on distance and elevation difference • number of stations influencing a given cell depends on specified influence distances • distances “fit”: obs. weighting increased throughout season • obs. anomalies applied to vic long term, combined with vic simulated SWE • adjustment specific to each VIC snow bands spatial weighting function elevation weighting function SNOTEL/ASP VIC cell
Current Forecasts: Initial Conditions
Dec 28, 2002 Jan 15, 2003 This past winter, alarmingly low December snowpacks mostly recovered by April, although some locations are still well off their long term averages Feb 1, 2003 Mar 1, 2003 Apr 1, 2003
Jan 1
Current forecasts: streamflow (ESP)
Feb 1 Mar 1 Apr 1
Current forecasts: streamflow (GSM)
GSM forecasts were somewhat similar to the ENSO/PDO composite forecasts, relative to unconditional ESP: slightly greater spring flows, slightly drier summer … and yet… Feb 1 Mar 1 Apr 1
Current Forecasts: GSM verifications
GSM forecasts this winter were muddled. e.g., February forecast:
Current Forecasts: Statistics
A limited set of statistics were calculated for the forecasts this spring, partly for comparison with official streamflow forecasts from NWS/NRCS. One example: Forecast flow anomaly (percent) at three percentiles, for 2003 APR-SEP average flow percentile # NAME ------------------------------------------------------------------- 1 MICAA 0.1 5 0.5 -7 0.9 -22 2 REVEL 3 ARROW 4 DUNCA 5 LIBBY 6 CORRA 7 HHORS 8 COLFA 9 KERRR 10 WANET 11 CHIEF 12 PRIES 13 DWORS 14 ICEHA 15 DALLE 5 -1 23 -18 -5 12 7 6 12 -9 -8 38 32 -0 -9 -7 1 -28 -28 -30 -32 -35 -21 -18 -19 -11 -15 -19 -21 -25 -24 -37 -36 -38 -38 -39 -36 -31 -32 -41 -32 -31
Current forecasts: UW/NRCS comparison
UW results to date are comparable to the official streamflow forecasts of the National Resources Conservation Service (NRCS) streamflow forecast group (one location shown).
90 80 70 60 50 1-Jan Apr-Sep Streamflow Forecasts Columbia River at the Dalles, OR 1-Feb 1-Mar
forecast date
UW NRCS Best Estimate 1-Apr 1-May (the best estimate to date is given by the NRCS May 1 forecast) computer disk failure halted UW forecasts
90 80 70 60 50 1-Jan
Current Forecasts: Results
ensemble median flow Apr-Sep Streamflow Forecasts Snake River "near mouth" 90 Apr-Sep Streamflow Forecasts Libby Reservoir Inflow 80 1-Feb 1-Mar
forecast date
UW NRCS Best Estimate 1-Apr 1-May 70 60 50 1-Jan 1-Feb 1-Mar
forecast date
UW NRCS Best Estimate 1-Apr 1-May 100 90 80 70 60 50 1-Jan Apr-Sep Streamflow Forecasts Dworshak Reservoir Inflow 1-Feb 1-Mar
forecast date
UW NRCS Best Estimate 1-Apr 1-May 90 80 70 60 50 1-Jan Apr-Sep Streamflow Forecasts Waneta Dam Inflow 1-Feb 1-Mar
forecast date
UW NRCS Best Estimate 1-Apr 1-May
Retrospective Assessment
Objective: Quantify skill of GSM based forecasts for western U.S.
• Used 1979-99 GSM climatology ensemble as forecasts • Four start dates (JAN,APR,JUL,OCT) • Five basins/sub-domains • 20 streamflow locations • Compared results to 2 baselines: • climatology ensemble (CLIM) • unconditional ESP ensemble • Assessed significance of results with Monte Carlo experiments • Results calculated for: • monthly basin-wide averages of P, T, RO, SM, SWE • Average flow in forecast months 1-3 and 1-6
Retrospective Assessment:
GSM forecast and climatology ensembles 10 member climatology ensembles 35 25 15 35 25 35 (21 sets) from 1979 SSTs from 1980 SSTs from 1981 SSTs 25 5 15 -5 5 15 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 -5 5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 -5 35 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 25 from 1999 SSTs 20 member forecast ensemble 35 15 5 25 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 15 5 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 from current SSTs
Retrospective Assessment: Results
General finding is that NCEP GSM climate forecasts do not add to skill of ESP forecasts, except… April GSM forecast with respect to climatology (left) and to ESP (right)
Retrospective Assessment: Results
during strong ENSO events, for some river basins: California, Pacific Northwest RO forecasts improved with strong-ENSO composite; but Colorado River, upper Rio Grande River basin RO forecasts worsened.
October GSM forecast w.r.t ESP: unconditional (left) and strong-ENSO (right)
Retrospective Assessment: Results
Additional findings: • P/T skill doesn’t guarantee RO skill • RO forecast skill doesn’t automatically result in streamflow forecast skill for the statistics evaluated thus far (e.g., 3 and 6 month flow averages)
Next Steps
Add NSIPP forecasts (retrospectively) to climate model forecast set (and web site) http://www.ce.washington.edu/pub/HYDRO/aww/w_fcst/w_fcst.htm
More complete evaluation of results and comparison with official NRCS/NWS products Increasing contact with NRCS/NWS forecasting groups Improved spin-up process Development of hydrologic ensembles from official NCEP seasonal forecasts
Next Steps: Official forecast product
Consensus forecasts based on a number of tools: CANONICAL CORRELATION ANALYSIS COMPOSITE ANALYSIS OPTIMAL CLIMATE NORMALS METHOD CONSTRUCTED ANALOG ON SOIL MOISTURE SCREENING MULTIPLE LINEAR REGRESSION Released as Probability of Exceedence (POE) curves/tables by Climate Division, separately for precipitation and temperature
Next Steps: Official forecast product
Consensus forecasts based on a number of tools: CANONICAL CORRELATION ANALYSIS
COMPOSITE ANALYSIS OPTIMAL CLIMATE NORMALS METHOD CONSTRUCTED ANALOG ON SOIL MOISTURE SCREENING MULTIPLE LINEAR REGRESSION
Next Steps: Official forecast product
Problem
: how to create meteorological ensembles of [P,T]?
Solution
: “Shaake Shuffle” approach (Martyn Clark et al., JHM, submitted) Pull out P,T values that span forecast distribution, then associate them using randomly sampled rank structure from appropriate historical distributions
A. Ranked ensemble output
Ens # (5) (7) (3) (6) (10) (9) (2) (4) (8) (1)
Stn 1
7.5
8.3
8.8
9.7
10.1
10.3
11.2
11.9
12.5
15.3
Ens # (2) (9) (4) (3) (7) (1) (6) (10) (5) (8)
Stn 2
6.3
7.2
7.5
7.9
8.6
9.3
11.8
12.2
13.5
17.7
Ens # (9) (3) (4) (7) (2) (6) (10) (1) (5) (8)
Stn 3
12.4
13.5
14.2
14.5
15.6
15.9
16.3
17.6
18.3
23.9
B. Randomly selected historical observations Ens #
1 4 5 2 3 6 7 8 9 10
Date
8 th Jan 1996 17 th Jan 1982 13 th Jan 2000 22 nd Jan 1998 12 th Jan 1968 9 th Jan 1976 10 th Jan 1998 19 th Jan 1980 16 th Jan 1973 9 th Jan 1999
Stn 1
10.7
9.3
6.8
11.3
12.2
13.6
8.9
9.9
11.8
12.9
Stn 2
10.9
9.1
7.2
10.7
13.1
14.2
9.4
9.2
11.9
12.5
Stn 3
13.5
13.7
9.3
15.6
17.8
19.3
12.1
11.8
15.2
16.9
END
blank
Bias: Developing a Correction
10 member climatology ensembles 35 25 15 35 25 35 (21 sets) from 1979 SSTs from 1980 SSTs from 1981 SSTs 25 5 15 -5 5 15 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 -5 5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 -5 35 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 25 from 1999 SSTs 20 member forecast ensemble 35 15 5 25 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 15 5 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 from current SSTs
Bias: Developing a Correction
10 member climatology ens.
35 25 35 15 25 5 35 15 25 1979 SSTs etc.
-5 5 15 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 -5 5 35 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 -5 25 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 15 5 from 1999 SSTs -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 July Tavg, for 1 GSM cell 30 25 20 15 10 0 GSM Observed 0.2
0.4
0.6
percentile (wrt 1979-99)
0.8
*
1 * for each month, each GSM grid cell and variable
Bias-Correction: Spatial Perspective
raw GSM output bias-corrected
shown 1 month, 1 variable (T), 1 ens-member
35 25 15 5 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 35 25 15 5 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6
Bias: Spatial Perspective
bias-corrected express as anomaly 35 25 15 5 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 8 4 0 -4 -8 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6
Downscaling: step 1 is interpolation
(bias corrected) anomaly anomaly at VIC scale 8 4 0 -4 -8 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 8 4 0 -4 -8 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6
Downscaling: step 2 adds spatial VIC-scale variability to smooth anomaly field
anomaly VIC-scale monthly forecast mean fields note: month m, m = 1-6 ens e, e = 1-20
Lastly, temporal disaggregation…
VIC-scale monthly forecast 35 25 15 5 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6
Lastly, temporal disaggregation…
VIC-scale monthly forecast 35 25 15 5 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6
Downscaling Test
1.
2.
3.
4.
Start with GSM-scale monthly observed met data for 21 years Downscale into a daily VIC-scale timeseries Force hydrology model to produce streamflow Is observed streamflow reproduced?
GSM climatology: use #2
10 member climatology ens. (21 sets) 35 25 35 from 1979 SSTs etc.
15 25 35 5 15 25 -5 5 15 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 -5 5 35 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 -5 25 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 15 from 1999 SSTs 5 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 sample: 21 member climatology ensemble 35 25 15 5 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6
GSM climatology: use #2
10 member climatology ens. (21 sets) 35 25 35 from 1979 SSTs etc.
15 25 35 5 15 25 -5 5 15 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 -5 5 35 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 -5 25 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 15 from 1999 SSTs 5 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 sample: 21 member climatology ensemble 35 25 15 5 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6 20 member forecast ens.
35 25 15 5 -5 Mon 1 Mon 2 Mon 3 Mon 4 Mon 5 Mon 6