11

Objective Drought Monitoring andObjective Drought Monitoring andPredictionPrediction

Recent efforts at Climate Prediction Ct.Recent efforts at Climate Prediction Ct.

Kingtse Mo &

Jinho YoonClimate Prediction Center

22

ObjectivesObjectives

Develop objective drought monitoring and prediction based on drought indices

Support drought monitor and outlook operation

Develop regional applications with users and the River Forecast Centers

33

Outline of presentationOutline of presentation

Current operation Drought briefing each month (8-11 of the

month, dial in is available)If you would like to participate, emailKingtse.mo@noaa.govUncertainties of drought indicesPrediction of SPIFuture plan

44

Define drought based on Drought IndicesDefine drought based on Drought Indices

Meteorological droughtMeteorological drought: : Precipitation deficit. Precipitation deficit.

Index: Standardized Precipitation IndexIndex: Standardized Precipitation Index Hydrological droughtHydrological drought: Runoff deficit: Runoff deficit Index: Standardized runoff indexIndex: Standardized runoff index Agricultural droughtAgricultural drought: S: Soil moisture deficit oil moisture deficit

Index: SM anomaly percentileIndex: SM anomaly percentile

55

SPISPISPI3:

SPI3 shows dryness over the Great Lake area

Wetness over AZ, New Mexico and western Texas.

For longer terms

A very wet picture over the Southeast and eastern central United States

D3 D2 D1

66

Multi model SM percentilesMulti model SM percentiles

Univ of Washington

NCEPU. Washington

Uncertainties in the NLDASUncertainties in the NLDAS

Feb 2010

77

•Differences between two systems are larger than the spread among members of the same system

•The differences are not caused by one model. They are caused by forcing.

• In general, extreme values from the UW (Green) are larger than from the NCEP (red) NCEP(red),UW(green)NCEP(red),UW(green)

standardized SM anomalies for area 38-42N,110-115W

88

Number of station P reports averaged over a year

Reports dropped for real time operation

Historical period Real time

99

Southeast & CBRFC pilot projectsSoutheast & CBRFC pilot projectsEnsemble hydrologic Forecasts in Ensemble hydrologic Forecasts in

support of NIDISsupport of NIDIS

CPC: Kingtse Mo, Jinho YoonCPC: Kingtse Mo, Jinho Yoon

EMC: Michael Ek, Youlong XiaEMC: Michael Ek, Youlong Xia

Princeton University : Eric WoodPrinceton University : Eric Wood

SERFC: John Schmidt, John Feldt ,Jeff DoburSERFC: John Schmidt, John Feldt ,Jeff Dobur

OHD: John Schaake and D. J. Seo OHD: John Schaake and D. J. Seo

CBRFC: Kevin WernerCBRFC: Kevin Werner

Funded by TRACS programFunded by TRACS program

1010

A wet regionA wet region

droughtdrought

6 mo running mean black line6 mo running mean black line

3 mo running mean (black line)3 mo running mean (black line)

No smoothing No smoothing

Red line: monthly mean, no smoothing

75-85W,31-35N

1111

A dry region6 mo running mean

1212

Hydrologic predictionHydrologic prediction

Develop hydrologic forecasts out to 6 months for drought indices

P , Tmin and Tmax from the CFS P , Tmin and Tmax from the CFS forecasts=> downscaling (from 250km to forecasts=> downscaling (from 250km to 50km) and error correction=> Vic 50km) and error correction=> Vic model=> SM % and runoffmodel=> SM % and runoff

Based on the Princeton System developed Based on the Princeton System developed by Eric Wood’s groupby Eric Wood’s group

Corrected P => append P time series=> Corrected P => append P time series=> SPI indices SPI indices

1313

Linear Interpolation Linear Interpolation Linear Interpolation : correct meanLinear Interpolation : correct mean Correct the model climatology and bilinear Correct the model climatology and bilinear

interpolation to a high resolution gridinterpolation to a high resolution grid

For variable A ensemble fcst: assume normal For variable A ensemble fcst: assume normal distributiondistribution

anomaly A’ rwt to model climatologyanomaly A’ rwt to model climatology

AA’’= A-model climatology = A-model climatology Corrected A = A’+ observed Corrected A = A’+ observed

climatologyclimatology

1414

Bias correction & Downscaling Bias correction & Downscaling (BCSD) (BCSD)

Probability mapping based on distributionsProbability mapping based on distributions

• Get probability distribution PDFs for A (coarse) and Get probability distribution PDFs for A (coarse) and A(fine)A(fine)

• From A’ (coarse) get percentile based on PDF (coarse)From A’ (coarse) get percentile based on PDF (coarse)• => assume the same percentile for the fine grid and => assume the same percentile for the fine grid and

work backward based on the PDF fine get A’ fine work backward based on the PDF fine get A’ fine (anomaly)(anomaly)

• If normally distributed , time ratio of stdIf normally distributed , time ratio of std

)(

)(*)(')('

coarse

finecoarseAfineA

Ref Wood et al (U. Washington 2002,2006)

1515

Schaake’s linear regressionSchaake’s linear regression

Schaake’s linear regression – Schaake’s linear regression – calibrate P ensemble forecasts calibrate P ensemble forecasts based on the historical performance based on the historical performance

)(

)(),(*)(')('

coarse

fineobshindcastcoarseAfineA

Ref: Wood and Schaake (2008)

Schaake et al. (2007)

Do no harm

1616

Bayesian merging & bias correction Bayesian merging & bias correction

Bayesian correction – calibrate P Bayesian correction – calibrate P forecast based on the historical forecast based on the historical performance and spread of members performance and spread of members in the forecast ensemble in the forecast ensemble

use all members in the ensembleuse all members in the ensemble

Ref: Luo et al. (2007); Luo and Wood (2008)

1717

Standardized Precipitation Index Standardized Precipitation Index Forecasts Forecasts

Append the bias corrected and downscaled P to the observed P time series

Calculate SPI from extended time series The advantages are (1) no need of

hydrologic model and (2) can use any base period.

P :time series : Jan1950-oct1981 append fcsts with ICs in Oct lead 1 f1 lead

2 f2 etcJan1950-oct 1981 (obs) Nov 1981 (fcst)

1818

Seasonal dependence of skillSeasonal dependence of skill

1.For the first 3 months, 1.For the first 3 months, AC>0.6 and RMS < 0.8 AC>0.6 and RMS < 0.8

2.Overall, Bayesian wins2.Overall, Bayesian wins

3. Skill is higher for Nov ICs 3. Skill is higher for Nov ICs and low for May ICsand low for May ICs

LI

BCSD

Schaake

Bayesian

1919

NovNov

JanJan

DecDec

FebFeb

LI Bayesian SPI6 T62 RMSE

2020

LI Bayesian SPI6 T62 rmse

May

June

July

Aug

2121

Over the Southeast,

SPI6 out to 3 months

2222

SPI6 fcst (contours) /ana (colored)SPI6 fcst (contours) /ana (colored)

32-40N Hovmoller

2323

3 month later3 month later

2424

High resolution run (T382)High resolution run (T382)and dynamic downscaling (RSM)and dynamic downscaling (RSM)

1.High resolution T382 run from April 19-23 ICs run through Nov(5 members) (Thanks Jae Schemm)

2.RSM (regional spectral model) downscaling from the CFS forecasts (April 28-May 3) ICs (50 km resolution) (Thanks, Henry Juang)

2525

T62 LI 5 members vs T382T62 15ensm LI

T62 vs RSM 5 mem LI, T62 15 member LI

5 member T382 & T62 Bayesian 15 members

5 member RSM & T62 Bayesian 15 members

2626

JJA P anom over the Northern Plains JJA P anom over the Northern Plains

(34-42N, 100-85W(34-42N, 100-85W))

BCSD, RSM or T382, Obs

mm

/day

mm

/day

2727

Future plan &our needsFuture plan &our needs Develop real time prediction of SPI’s based on the

CFSRR hindcasts The bias corrected P and T will drive VIC model to

produce hydrologic forecasts of soil moisture and runoff

Develop regional applications with the RFCs What do we need? Better station reporting of real time P Better P analyses Better global model prediction of P Better model physics

2828

Discussion questionsDiscussion questions

What current activities (monitoring and What current activities (monitoring and forecasts) can we build on?forecasts) can we build on?

Regional vs entire United StatesRegional vs entire United States How can we network and coordinate How can we network and coordinate

drought related information such as drought related information such as drought impact, planning and information drought impact, planning and information exchange?exchange?

What gaps do we need to fill?What gaps do we need to fill? What issues are important to you, but What issues are important to you, but

have not been discussed?have not been discussed?

2929

Streamflow Streamflow fcstsfcsts

the binary event for observed monthly mean

Row 2-6 represents the exceedence probability for forecasts initialized from Nov 2006

Luo and Wood