Downscaling as a contributor to developing robust messages of change

Bruce Hewitson

Information framework

Limitations of current offering

A case study example

The multi-model challenge

CORDEX and AR5

A: The Information Framework

Pick as role as a stakeholder trying to accommodate climate change

What has already changed?

Is that any different from variability?

What is the future?

When is the future?

How do you know that?

Where do you get your information?

Do you “believe” it?

How do you know how good it is?

Would you spend your own money based on this information?

At the root of the issue, what level of information do you trust?

DataClimate models, historical

observations, trends,

downscaling, projections, event

frequency, …

InformationMeasures of vulnerability and

risk, threshold exceedence,

combinatory impacts,

uncertainty and confidence,

regional scale variations, …

KnowledgeAssessing options,

understanding consequences,

evaluating responses, informing

decision making, …

A basis for actionPolicy development to balance

competing priorities, strategic

investments in adaptation and

mitigation, new research

avenues, coordination of

response frameworks, …

Generated by models,

analyses, downscaling,

observations …

We are not always sure

when we have “information”

Comes with close coupling

between science and society

Actions are risky, and takes

place within a multi-stressor

context

De

live

red

by s

cie

nce

Need

ed

by

so

cie

ty

WG

1 Obs & past

trend

Circulation

changes

Framework for integrating information on the physical climate system

for regional impacts and adaptation needs

Regional Integration and Understanding

Data products with

articulated uncertainty

Storylines and robust

messages of changeWG

1 /

WG

2-C

h 2

1

Adapted from Hewitson et al., 2010

Contextualization

around real world

questions

WG2 Part A

WG2 Part B

WG

2

Stakeholders

Downscaled

change

GCM

change

General classes of downscaling

Local climate = f (larger scale predictors) + locally forced variance

Dynamical

Two approaches

Empirical-statistical

Three main classesPerturbed observed

RCM Hi-res GCM

Weather Generators Transfer Functions

Trained on long term

time series and

atmospheric re-analysis

data, conditioned by

GCM parameters to

capture low frequency

variance

Trained on time series

that spans range of

variability, residual local

scale variance added

stochastically

Index / analogues

Requires long term data

sets and uses weather

typing or historical

analogues

Even then … it’s not only about the data!

Users face multiple dependencies in seeking value for

adaptation decision making

Articulation

of relevant

thresholds

Understanding

natural

variability

Effective

communication

between

knowledge

provider and

user

Tailored

information

products

Quantified

uncertainty

Iterative and

sustained

re-examination

Synergy between

process change and

local change

Accommodation of

feedbacks and tipping

points

Assessment of

error

Balancing

multi-stressor

factors

Etc …

Downscaling

Information source

(AO)GCMs – CMIP3

Downscaling – RCMs/SD

Process changes

Historical changes

User communities

Research scientists

Policy / mitigation

Vulnerability / Impacts

Adaptation

Cre

atio

n

Tra

nsfo

rma

tio

n

Inte

rpre

tatio

n

Each source has different:

- attributes of signal and noise

- limitations on interpretation

- degrees of uncertainty

- methodologies of evaluation

Each community has different:

- definitions / terminology

- priorities of need

- scales of interest

- access to information

What we would like to accomplish …

B: The Confusion of offerings

A proliferation of portals and data sets, with poorly articulated

uncertainties, weakly explained assumptions and dependencies,

data implied as information, and communicated to a user

community poorly equipped to understand the limitations

GCM data presented for local application

Fig 3: From Tabor & Williams, 2010

The proliferation of change factor “downscaling” as credible local scale information

Often users do not grasp the multiple assumptions and dependencies, and the fact that the data is not robust at fine scales.

Large vulnerability on change factor of precipitation at GCM grid cell resolution

Multiple portals springing up based on the GCM change factor approach

1

2

3 4

1

2

3 4

1

2

3 4

1

2

3 4

Typical statistical

downscaling

products at the

time of the AR4,

for Africa

Usually spatially

discontinuous

with weak spatial

cohesion

Significant degree of “noise” in RCM

downscaling

Robustness of Future Changes in Local

Precipitation Extremes. (Kendon et.al. 2008)

Examines the contribution of natural variability to

the spread of future climate projections of

precipitation extremes

“In general, where climate noise has a significant

component varying on decadal time scales, single

30-yr climate change projections are insufficient to

infer changes in the extreme tail of the underlying

precipitation distribution.”

FIG. 1. Percentage changes

in extreme precipitation in the

three different ensemble

members of HadRM3H

[DJF – left, JJA – right]

C: An example of multimodel downscaling complications

Downscaling may not lead to convergence or information products

of value – new complications can be intrroduced

Zambia 13.55S 32.58E – downscaled control

Examples of where things go right and wrong

Zambia 13.55S 32.58E – downscaled anomaly

Examples of where things go right and wrong

Surat, India – downscaled control

Note the ok-looking downscaled GCM

And some not-ok

Examples of where things go right and wrong

Surat, India – downscaled anomaly

The “ok” downscaled

control climate GCM!

Examples of where things go right and wrong

Calcutta, India – downscaled control

Interesting clustering of

downscaled GCMs – not

so clear in raw GCM data

Examples of where things go right and wrong

Calcutta, India – downscaled anomaly

Examples of where things go right and wrong

D: An case study

Drawing on multiple lines of evidence to build a

message of change

AOGCM multi-model

projected changes in sea

level pressure and surface

winds

Sea level pressure multi-

model median anomaly

Surface wind multi-model median anomaly

Raw GCM projections: Wind speed

and direction90th percentile

Median“Best estimate?”

10th percentile

Raw GCM projections: rainfall

75th percentile

Median“Best estimate?”

25th percentile

AR4 multi-model

median anomaly:

(2045-2064) - control

GCMs perspective

mm/month: max change = ~15%

DownscaledAR4 multi-model

median anomaly:

GCMs downscaled

precipitation change

(2045-2064)

Downscaled

PROJECTION

GCM

Information source Message Discussion

Historical trends Core winter wetting dominantly in the mountains

Shoulder season drying

Marginal indications of a possible wetter summer

The region is spatially inhomogeneous in trend magnitude, although the dominant trends can be seen to greater or lesser degrees across the region

GCM changes in circulation / processes Increased subsidence due to a stronger mid-latitude high pressure inducing drying

Deeper thermal surface trough over the continent increasing west coast pressure gradient and possibly summer convection in the east

Poleward shift in mid-latitude flow decreasing frontal intensity

Increased longshore west coast wind promoting stronger upwelling, colder coastal waters, and consequent drying on the west coast.

The models are in good agreement on these large scale circulation changes, albeit with a range of differing magnitudes of change. The change further is physically consistent with the anticipated first order response of the climate system.

GCM grid cell changes General drying over the region

A weak suggestion of possible summer wetting in the north east

The models are in strong agreement on the drying message for the region, but it is clear that the spatial detail related to local scale topography is absent.

Local scale downscaled changes A general drying in the west with modest wetting to the east, modulated by the topography

Core winter wetting in the important water catchments in the core winter season

Small decreases in rainfall frequency in the west and small increases in the east

Changes in dry spell duration commensurate with the above changes.

The downscaled projected changes across all models are robust in spatial pattern although vary in magnitude, and the projected changes in some regions are too small to be of consequence. Of importance is the drying in regions of non-irrigated agriculture in the west, and while core winter wetting in the key catchments is indicated for the near term, later in the century this reverses. Taken with an increase in temperatures, the indication is for problematic increases in water stress.

Assess, distill, conclude, communicate a message:

E: CORDEX

For Africa at least, the promise of a major advance

Multiple GCMs + process change assessment

Multiple RCMs

Multiple statistical downscaling

Common experiment framework

NARCCAP

CLARIS

ENSEMBLES

RCMIP

+ polar regions From Colin Jones

CORDEX: changing the landscape of information?

Evaluating the first CORDEX results over AfricaColin Jones and Grigory Nikulin, Rossby Centre, SMHI

Closing thoughts

Downscaling is a key (but only one) foundation for

developing regional projections of change

Perhaps the biggest challenge relates to the

integration of multi-model multi-method results