Applying probabilistic scenarios to environmental management

and resource assessment

Rob WilbyClimate Change Science Manager

rob.wilby@environment-agency.gov.uk

Adaptation can involve costly and difficult decisions...

…involving development of major assets

-60

-40

-20

0

20

40

60

80

100

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

Mo

nth

ly t

ota

l (%

ch

ang

e)HadCM3 CGCM2 CSIRO ECHAM4

-40

-20

0

20

40

60

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

imu

m 5

-day

to

tal

(% c

han

ge)

HadCM3 CGCM2 CSIRO ECHAM4

Climate model uncertainty and adaptation response...

Downscaled precipitation scenarios for the River Thames under A2 emissions in the 2050s

Wider policy drivers

Shaping adaptation in the UK

• EU Directives (e.g., Groundwater, Landfill, Water Framework)

• UK Government Adaptation Policy Framework

• Planning (e.g., CFMPs, SMPs, AMP, BAPs, PPS25)

• Guidance (e.g., Defra flood factor)Land use issues are identified as a priority area for embedding climate change in Agency policy and process.Photo: Philip Owen

Talk outline

Derwent reservoir during the 1995 UK drought.Photo courtesy of Nick Jackoby

The plan

• Using a multi-model ensemble to evaluate adaptation options

• Components of uncertainty affecting river flow projections

• Future plans and challenges of using probabilistic information

• Concluding remarks

Using multi-model ensembles: Appraisal of adaptation options

for the River Kennet

Long-term nutrient enrichment of the River Thames, 1930-2000

‘Luxuriant’ macrophyte growth

Statistical DownScaling Model (SDSM) and predictor variable archive

http://www.sdsm.org.uk/

Downscaled daily precipitation for the River Kennet under A2 emissions by the 2080s illustrating the relative aridity of the HadCM3 GCM

-60

-40

-20

0

20

40

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

Wet

-day

s (%

ch

ang

e)

HadCM3 CGCM2 CSIRO

-20

-10

0

10

20

30

40

50

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

95th

per

cen

tile

(%

ch

ang

e)

HadCM3 CGCM2 CSIRO

Changes in daily precipitation downscaled from each GCM

CATCHMOD daily river flow changes under A2 emissions

-100

-50

0

50

100

150

200

250

300

Julian day 2080s

Dis

char

ge

(% c

han

ge)

Daily precipitation and PE downscaled for the River Kennet using three GCMs for the 2080s

INCA-N model of soil and instream nutrient concentrations

INCA-N simulation of Nitrate-Nunder A2 emissions: headwater

Nitrate as Nitrogen, A2 emissions

6

7

8

9

10

11

12

1960 1980 2000 2020 2040 2060 2080 2100

mg

N/l

CGCM2 CSIRO HadCM3

CGCM2S CSIROS HadCM3S

Concentrations exceeded 5% of the time

Appraisal of adaptation options

Five scenarios

• Baseline conditions under HadCM3 A2 emissions

• Fertiliser use reduced by 50%

• Deposition of atmospheric pollutants reduced by 50%

• Water meadow creation (4 x surface area of river)

• Combined approach with 25% reduction in fertiliser and

deposition, water meadow creation (2 x surface area)

Simulation of adaptation outcomes under HadCM3 A2 emissions

Concentrations exceeded 5% of the time

Nitrate as nitrogen, A2 emissions

0

2

4

6

8

10

12

1960 1980 2000 2020 2040 2060 2080 2100

mg

N/l

Baseline Fertiliser Deposition

Meadows Combined

Summary (part 1)

• Large uncertainty in projected (summer) climate due to GCM• HadCM3 : lower river flows, deployable yield, nutrient flushing following

prolonged drought, and general decline in water quality• CGCM2 : wetter summers, more deployable yield, greater reliability, stabilising

water quality beyond 2050s• Differences between A2 and B2 emissions relatively small• Effectiveness of strategies: reduce fertiliser (or land use) > water meadow

creation > reduced atmospheric deposition• Combined approach sustains water quality at 1950s level even under climate

change projected by HadCM3

Probabilistic framework: Assessing uncertainties in low

flows for the River Thames

The ‘cascade’ of uncertainty affecting low flow projections

Future society

Emissionspathway

Climatemodel

Regionalscenario

Impactmodel

Impact

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 100 200 300 400 500

Effective winter rainfall (mm)

CD

F

Observed

NCEP

CGCM2

CSIRO

ECHAM4

HadCM3

Uncertainties due to GCM/ downscaling pair

Summer Winter

Model Bias (%) Weight Bias (%) Weight

CGCM2

CSIRO

ECHAM4

HadCM3

52.1

14.3

49.6

33.6

0.138

0.503

0.145

0.214

42.2

6.0

16.1

14.9

0.074

0.522

0.194

0.210

NCEP 7.6 n/a 4.8 n/a

An Impacts Relevant Climate Prediction Index

This IR-CPI is based on the skill of the GCM/downscaling pair for effective rainfall in the Thames basin

0

5

10

15

20

25

30

35

40

45

50

1960 1965 1970 1975 1980 1985 1990

Q95

(m

3/s)

Observed CATCHMOD REGESSION

Uncertainty due to low flow model structure

Derived from observed daily rainfall and PE

Uncertainty due to waterresource model parameters

0

0.2

0.4

0.6

0.8

1

0 10 20 30 40 50 60 70 80 90 100

DP (%) 1961-1990

Nas

h

0

0.2

0.4

0.6

0.8

1

0 10 20 30 40 50 60 70 80 90 100

Identifiability - high

• 4x GCMs, 2x emissions, 2x downscaling methods, 2x low flows models, 100x parameter sets

• Weight GCMs by modified Climate Prediction Index• Weight low flow model structures by radj statistic• Weight low flow model parameters by N-S score• Emissions and downscaling method unweighted• Monte Carlo simulation (2000+ runs)• Evaluate using (Q95) low-flow index for River Thames

An experimental framework for assessing uncertainties

0.0

0.2

0.4

0.6

0.8

1.0

-50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100

Q95 (% change)

CD

F

A2 B2

Uncertainties due to emission scenario

0.0

0.2

0.4

0.6

0.8

1.0

-50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100

Q95 (% change)

CD

F

CGCM2 CSIRO ECHAM4 HADCM3

Uncertainties due to GCM

0.0

0.2

0.4

0.6

0.8

1.0

-50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100

Q95 (% change)

CD

F

CF SD

Uncertainties due to downscaling method

0.0

0.2

0.4

0.6

0.8

1.0

-50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100

Q95 (% change)

CD

F

CATCHMOD REGMOD

Uncertainties due to low flow model structure

0.0

0.2

0.4

0.6

0.8

1.0

-50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100

Q95 (% change)

CD

F

CATCHMOD REGMOD

Uncertainties due to low flow model parameters (HadCM3, A2)

0.0

0.2

0.4

0.6

0.8

1.0

-50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100

Q95 (% change)

CD

F

2020s 2050s 2080s

Combining all sources of uncertainty

Summary (part 2)

LikelihoodUncertainty component

Min Max

Emissions 82 83

GCM 47 100

Downscaling 66 100

Hydrological model 72 92

All weighted/ unweighted 76 82

Conditional probabilities of lower summer river flows in the Thames by the 2080s

Next steps: Preparing the way for the

UKCIPnext probabilistic scenarios

Impacts assessment usingClimatePrediction.net archive

Frequency distribution of global mean temperature response to doubled CO2 produced by CP.net, compared with IPCC (2001) range.

Reviewing Defra’s 20% allowance for future flood risk

Variations in the 20-year flood by the 2050s under the UKCIP02 Medium-High emissions scenario

Source: Reynard et al. (2004)

Incorporating probabilities in flood maps & planning guidance

Changes in the extent of the 0.5% tidal flood between 2000s (green) and 2050s (red line).Source: Tim Hunt

• Probabilistic climate change information is increasingly becoming available

• Planning and asset management will need to make best use of new forms of climate change information

• Risk-based modelling frameworks must be developed to inform adaptations responses

• Developing guidance on the use of conditional probabilities will be a critical part of this process

Concluding remarks

A few words of wisdom

“If there is even a small risk that your house will burn down, you will take care to install smoke alarms and buy insurance. We can scarcely do less for the well-being of our society and the planet’s ecosystems”

Spencer Weart (2003: 199)