GYC108 Climate and society Regional climate downscaling – in practice.

GYC108 Climate and societyRegional climate downscaling – in practice

-40-30-20-10

0102030405060

CGCM3 CNRM CSIRO GFDL GISS IPSL MPI

Chan

ge (%

)

Djibouti annual PRCP2050s

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

CGCM3 CNRM CSIRO GFDL GISS IPSL MPI

Chan

ge (d

egC)

Djibouti annual TMAX 2050s

Projected changes in annual rainfall (PRCP) and maximum temperature (TMAX) in Djibouti by the 2050s downscaled from seven GCMs under SRES A2 emissions. Data source: Climate Systems Analysis Group, University of Cape Town

Choice of GCM(s): affects all downscaled scenarios

Comparable skill of statistical and dynamical methods

Correlation of modelled and observed precipitation indices for each season for SEE. Bars show inter-quartile range and median with 5th and 95th percentiles indicated by outer range. Source: Haylock et al. (2006)

Spring precipitation scenarios

-80

-60

-40

-20

0

20

40

60

Tanger Mekness Casablanca Beni Mellal Marrakech Oujda Midelt Agadir Ouarzazate

% c

han

ge

UCT-CSIRO UCT-ECHAM4 UCT-HadAM3 SDSM-HadCM3

Choice of downscaling method determines local scenario

Projected changes in spring precipitation totals for the 2080s for two downscaling methods (UCT, SDSM) and three climate models (CSIRO, ECHAM4, HadCM3) under SRES A2 emissions. Source: Wilby & DMN (2007)

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Co

rrel

atio

n c

oef

fici

ent

MSLP QSUR

Predictability varies in space and time

Variations in the strength of the correlation between daily wet–day amounts at Eskdalemuir (55º N, 3º W) and mean sea level pressure (MSLP), and near surface specific humidity (QSUR) over the Scottish Borders region, 1961–1990.

12.5°W 10.0° 7.5° 5.0° 2.5° 0.0° 2.5° 5.0°E32.5°

35.0°

37.5°

40.0°

42.5°

45.0°N

1 0 0

0 0 0

0 0 3

0 3 0

1 5 2

5 4 3

4 1 0

1 5 0

0 0 0

1 0 1

1 4 3

3 2 2

9 3 0

5 3 3

0 0 1

1 1 1

Choice of predictor variable domain

Frequency and location of predictors selected for downscaling daily precipitation occurrence (top) and amounts (bottom) at selected sites across Iberia (crosses)

Site-specific optimum predictor sets for Northern Ireland when using the Republic of Ireland grid boxes. Site pie chart segment size reflects the ranked order of explained variance provided by each of the 5 optimum predictors.

Source: Crawford et al. (2007)

Choice of predictor

variable suite

Low skill at heavy summer rainfall

Percent difference (discrepancy) between grid cells for UKMO and control simulations for the 1-day 5-year return value during summer. Most of the RCMs underestimate the precipitation extremes.

Source: Fowler and Ekstrom (2009)

Extrapolating beyond the calibration data

y = 0.0361x3 - 0.3237x2 + 2.5025x + 4.8157R² = 0.9991

0

5

10

15

20

25

0 1 2 3 4 5 6 7 8

Dai

ly g

ust

(m

/s)

Geostrophic flow (hPa)

Changi GUST v GSUR 1971-2000

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

0 5 10 15 20 25 30

Fre

qu

ency

Daily gust (m/s)

Observed Quantile

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0 2 4 6 8 10 12 14

Fre

qu

ency

Geostrophic flow (hPa)

20CM A2 2050sA2 2080s B1 2050sB1 2080s NCEP

Downscaling daily wind gust (m/s) for Singapore under present (top) and future (lower) climate conditions.

1. GCM boundary conditions are the main source of uncertainty affecting all regional downscaling methods

2. Statistical and dynamical downscaling have similar skill

3. Different downscaling methods yield different scenarios

4. There are no universally “optimum” predictor(s)/domains

5. Downscaling extreme events is highly problematic (summer rainfall predictability is very low)

6. Skilful downscaling of present climate does not imply skilful downscaling of future scenarios of change

Findings from inter-comparison studies

Photo: Richenda Connell

Photo: http://www.flickr.com

When is statistical downscaling problematic?

Photo: Bull (1929)

The global network of the World Weather Watch (WWW) stations colour coded to indicate silence (red dot) or reporting rates in 2008. Source: WMO (2009)

Data sparse where needed most

Local forcing: Cooling by late season snow cover

Niwot C1 winter mean TAVG (r=0.83)

-10

-8

-6

-4

-2

0

1950 1960 1970 1980 1990 2000

Tem

per

atu

re (

deg

C) Observed SDSM

Niwot C1 spring mean TAVG (r=0.82)

-6

-4

-2

0

2

4

6

1950 1960 1970 1980 1990 2000

Tem

per

atu

re (

deg

C) Observed SDSM

Niwot C1 summer mean TAVG (r=0.57)

8

9

10

11

12

13

14

1950 1960 1970 1980 1990 2000

Tem

per

atu

re (

deg

C) Observed SDSM

Niwot C1 fall mean TAVG (r=0.87)

-2

0

2

4

6

8

1950 1960 1970 1980 1990 2000

Tem

per

atu

re (

deg

C) Observed SDSM

Observed and downscaled seasonal mean TAVG at Niwot Ridge, Colorado

1982/83 El Nino

Simpler methods may suffice

Sensitivity of annual mean total chlorophyll-a concentration (mg m-3) to changing water temperature (C) and nutrient load: (a) Changing nitrate and soluble reactive phosphorous (SRP); (b) Changing SRP. Source: Elliott & May (2008)

Simpler methods may suffice

90th percentile daily precipitation (mm/d) in autumn (SON) for OBS (top left), for direct output from the reanalysis (ERA40), for the local intensity scaling (LOCI-E40), and for three regional climate models (CHRM, HADRM, HIRHAM). Results are valid for the period 1979-1993.Source: Schmidli et al. (2006)

Guidance for downscaling in practice

UNDP (2006)IPCC (2007)

Public domain downscaling tools

SDSM UCT

ENSEMBLES

Application: daily precipitation (for ski resorts)

Downscaled and observed daily precipitation for a site above 3000 m in the US Rockies

Niwot C1 PRCP Jan 1983 to May 1983

0

5

10

15

20

25

30

01/01/1983 01/02/1983 01/03/1983 01/04/1983 01/05/1983 01/06/1983

Pre

cip

itat

ion

(m

m) Observed SDSM

Forecast of June-July-August rainfall anomalies (% normal) based on SSTs in April 2009, downscaled from ECHAM4.5 by RSM

Source: FUNCEMEhttp://www.funceme.br/DEMET/index.htm

Application: high resolution seasonal forecasts

West-wide seasonal hydrologic forecast system. Source: http://www.hydro.washington.edu/forecast/westwide/sflow/index.cpc.6mons.shtml#seas_vol

Application: seasonal

streamflow forecasts

10

12

14

16

18

20

22

24

26

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Mean

(te

nth

s d

eg

C)

CGCM2 CSIRO ECHAM4 HADCM3 OBS NCEP

-3

-2

-1

0

1

2

3

4

5

6

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Ch

an

ge in

mean

(te

nth

s d

eg

C)

CGCM2 CSIRO ECHAM4 HADCM3

-1

-0.5

0

0.5

1

1.5

2

2.5

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Ch

an

ge in

ep

iso

des (

days>4 d

eg

C)

CGCM2 CSIRO ECHAM4 HADCM3

0

1

2

3

4

5

6

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Ep

iso

des (

days>4 d

eg

C)

CGCM2 CSIRO ECHAM4 HADCM3 OBS NCEP

Application: intensity of London’s urban heat island

Statistically downscaled nocturnal temperature gradient between central and outer London for the present and 2050s. Source: Wilby (2008)

Utility of downscaling when there is no consensus amongst GCMs?

Source: IPCC AR4 (2007)

Utility of higher resolution with uncertainty?

Changes in wettest day in summer by 2050s. Source: UKCP09

2050s A1FI90th percentile

2050s B110th percentile

Number of ISBN research publications listed on the Web of Science. [Accessed May 2009]

0

20

40

60

80

100

120

1993 1995 1997 1999 2001 2003 2005 2007

Jour

nal p

ublic

ation

s

Year

"downscaling and climate" "statistical downscaling""regional climate model" "downscaling and impact""downscaling and adapt*"

To what extent is downscaling science really entering adaptation practice?