1 developing objective climate drought monitoring and prediction – a ctb project kingtse mo team...

28
1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

Upload: edwin-picker

Post on 31-Mar-2015

243 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

1

Developing objective climate drought monitoring and

prediction – A CTB project

Kingtse Mo

Team Leader

Drought NIDIS

Page 2: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

2

Drought monitoring:http://www.cpc.ncep.noaa.gov/products/Drought

Drought briefing:

Second Thursday each month

Call in is available

[email protected]

Atmospheric variables: NARR

Hydrological variables: ensemble NLDAS: Mosac, Noah, VIC and NARR

Prediction: NAEF, (ESP, CFS downscaling)

Page 3: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

3

Partners and contributors

CPC: Kingtse Mo, Wanru Wu, Muthuvel Chelliah, Wei Shi,Yun Fan,Huug van den Dool

EMC: the NAEF forecast group,

Ken Mitchell, Jesse Ming, Youlong Xia

GSFC/NASA: Brian Cosgrove and Chuck Alonge

University of Washington: Andy Wood, Dennis Lettenmaier

Princeton University: Eric Wood, Lifeng Luo, Justin Scheffield

Page 4: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

4

For drought assessment

• Are we able to use the North American Land data Assimilation Systems (NLDAS) to develop early drought warning system?

• Do we need more than one index to assess drought?

Page 5: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

5

Conditions

• Reliability: Agreement among the NLDAS systems.

• Consistency: All different indices/NLDAS should be able to select strong drought events.

• Long term data to obtain representative probability distribution functions

• Availability : Operational in near real time.

Page 6: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

6

Drought Indices

• Meteorological drought: Precipitation deficit. Index: Standardized Precipitation Index• Hydrological drought: Streamflow or runoff

deficit Index: Standardized runoff index• Agricultural drought: soil water storage deficit Index: SM percentile (to be determined)

Page 7: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

7

Data sets

• VIC - 0.5 degrees from 1915-2003 (Maurer et. al. 2002, Thanks! Andy Wood)

• Noah- 0.125 degrees from 1948-2001 from Fan and van den Dool

• Time scales: Monthly means from 1950-2000

Page 8: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

8

What is the soil moisture probability distribution function ?

• Bi-modal (D’Odorico and Porporato 2004)

• Beta distribution (Schiffield 2004)

A non parametric method will be used to determine the soil moisture distribution function based on the monthly mean VIC data from 1915-2003

Page 9: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

9

There are 4 types of distribution

Unimodal: Type 1 and 3 :and type 3 is close to the Gamma distribution

Bimodal: Type 2 and 4: Type 2 has two peaks and the second peak for type 4 is a shoulder

Type 1 Type 3

Type 2Type 4 To get smooth

distribution function, more points are needed.

Page 10: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

10

SM total SM anomalies

SM PDF for selected RFC: For SM anomalies, PDF is Gaussian

Type 1:

Southeast RFC

Type 2:

Mid Atlantic RFC

Type 3:

Colorado RFC

Type 4:

Ohio RFC

Red :winter

Blue summer

Green Spring

red crosses Fall

Page 11: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

11

Uncertainties among the NLDAS

• Models: VIC and Noah with the common period from 1950-2000.

• Compute SM percentiles for each model

A) Form SM standardized anomalies with respect to its own monthly climatology.

B) Obtain percentiles based on Gaussian probability distribution function for each month

Page 12: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

12

SM percentile difference between VIC and Noah

•Differences are regional dependent

•Over the areas east of 90W, differences are small.

•Over the areas west of 90W, differences are large.

•The RMS error is larger than 25%: the difference between one drought class to another

CorrRMS

Page 13: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

13

Two reasons:

a) SM is more persistent over the west region so SM at deeper levels play a role. That depends on model soil structure & parameters

b) Difference in precipitation. Less stations over the western mountains and different ways to grid data

Corr SPI3 VIC,noah RMS SPI3 VIC,Noah

Page 14: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

14

The need for ensemble NLDAS

Total SM percentile for selected River Forecast Center areas

Vic(Blue), Noah (black)

From 1950-2001

For RFC lower Mississippi, the VIC and the Noah agree well

For Missouri basin, there are large differences

3 month running mean soil moisture percentiles

Page 15: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

15

More than one index is needed

SM % are more smooth. SM has longer memory and events occur 2-3 months later than P and last longer

SPI has higher frequency: more events and shorter duration

Corr(SPI,SRI)=0.87

Corr(SRI,SM)=0.72

Corr(SPI,SM)=0.52

Longer record is needed

Colorado RFC

SM

SPI6

SRI6

Page 16: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

16

RFC: Southeast

All indices are similar.

They are likely to pick up same events

Corr(spi,sri)=0.91

Corr(sri,SM)=0.73

Corr(spi,sm)=0.63

SM

SPI6

SRI6

Page 17: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

17

Precipitation anom for the water year 2006- 2007

The dipole with wet Great -Plains & Dry SE persists from Jan 2007 to JAS

2. A weak to normal NA monsoon

Page 18: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

18

Jan 2007

Droughts as measured by the SPI6SPI<-0.8 enter drought conditions

3-cell pattern

Page 19: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

19

Conclusions

• The uncertainties of NLDAS are larger over the western region than areas east of 90W.

• Over areas east of 90W, different indices based on P, SM or runoff are likely to pick up same drought events.

• Over the west region, uncertainties are too large to select drought events for less than 3 months over 0.5 degree boxes

Page 20: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

20

What do we need?

NLDAS data : from different models & forcing

• How many samples are needed?

• Are differences in the NLDAS caused by model or forcing ?

• What is the best way to consolidate them to form ensemble?

Better precipitation analyses

Page 21: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

21

1. Overall, all indices pick up major drought events

2. SM (Black) droughts have longer duration

SM%

SRI

SPI

SM%

SRI

SPI

SM%

SRI

SPI

SM%

SRI

SPI

SM%

SRI

SPI

SM%

SRI

SPI

Page 22: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

22

Precipitation and SPI6

Drought

Page 23: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

23

SM percentiles for

the Colorado River Fcst Ct

1. NLDAS over the western region differs too much to analyze “low flow ‘ cases on the scales less than 3 months

2. VIC has more high frequency components than the Noah.

3. For droughts on the time scales 6-months or longer, the differences are smaller

Black-Noah

Blue VIC

Page 24: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

24

Runoff

• Runoff obeys Gamma distribution similar to precipitation.

• For drought on different time scales, they can be represented by standardized runoff indices computed the same way as the SPIs. We will have 3, 6, 12 and 24 month SRIs

Page 25: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

25

Areas with bi modal distribution coincide with areas with strong seasonal cycle

Page 26: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

26

Probability Distribution function

For each grid point : • Pool all months together and use 9 grid points in

the 1x1 deg. box centered at that grid point : there are 9X89X 12= 9612 data points

• From 9612 points, we determined the histogram and normalized to 1.

• Group grid points into different types using simple cluster analysis. (All points in the same type correlate with each other with correlation greater than 0.8)

Page 27: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

27

Precipitation and SM (32-36N)

3 cell pattern persisted for 2007

Can we declare drought for the SE? YES

Page 28: 1 Developing objective climate drought monitoring and prediction – A CTB project Kingtse Mo Team Leader Drought NIDIS

28

Probability Distribution function

For each grid point : • Pool all months together and use 9 grid points in

the 1x1 deg. box centered at that grid point : there are 9X89X 12= 9612 data points

• From 9612 points, we determined the histogram and normalized to 1.

• Group grid points into different types using simple cluster analysis. (All points in the same type correlate with each other with correlation greater than 0.8)