Transcript 20100921101511009-152608 - Isaac Newton Institute for

Theme D: Model
Processes, Errors
and Inadequacies
Mat Collins, College of Engineering, Mathematics and Physical
Sciences, University of Exeter and Met Office Hadley Centre
© Crown copyright Met Office
A Much Used Quote
climate models
• “All models are wrong, some are useful” George Box
• Easy to test when a model is wrong, much
harder to say when it is useful
Uncertainties and Errors in
Climate Models
• Uncertainties in parameters, sampled using
Perturbed Physics Ensembles (PPEs)
• “Structural” uncertainties, at least partially sampled
by Multi-Model Ensembles (MMEs)
• Coding errors!
• Errors common to all models (examples follow)
• Missing processes
Why do Errors Matter
• Predictions and projections
• Detection & attribution
• Understanding climate and climate change
• Risk assessment for natural events (cat modelling)
•…
Types of Errors
& Inadequacies
Incapabilities
Iconic Error 1: Blocking
© Crown copyright Met Office
Tim Hinton, Gill Martin
Iconic Error 2: Double ITCZ
models - MME
observations
Figure 8.5
Source: IPCC Fourth Assessment Report
Errors known to be important for
e.g. climate projections
• Boé J, Hall A, Qu X (2009), September Sea-Ice
Cover in the Arctic Ocean Projected to Vanish
by 2100, Nature Geosci, 2: 341-343
• Hall A, Qu X (2006) Using the current
seasonal cycle to constrain snow albedo
feedback in future climate change. Geophys.
Res. Lett., 33, L03502
Poorly Observed but
Important Variables
Which model is
right?
Figure 10.15
Source: IPCC Fourth Assessment Report
Known Missing Processes
• Land use change (in HadCM3Q)
• Soot/black carbon (in HadCM3Q)
•…
• Dynamic ice sheets
• Methane hydrate release
• Stratosphere
•…
Unknown Missing Processes
Dealing with Model
Errors and
Inadequacies
Dealing with Model Errors and
Inadequacies
• Do nothing
• Do nothing but at least discuss the implications of
doing nothing
• Embark on an ambitious programme of metrics
and intercomparisons
• Use a discrepancy term
•…
• (Continue to improve models, of course)
Metrics, metrics, metrics
MME Error Characteristics
Relative model errors
Reichler and Kim, 2008
HadGEM Traffic Lights
© Crown copyright Met Office
Introduce a Discrepancy e.g. Rougier
(2007)
y  f(x )  d
*
y = {yh,yf} climate variables (vector)
f = Climate model e.g. HadCM3
x* = best point in HadCM3 parameter space – for
observable and non-observable fields
d = discrepancy – irreducible/”structural” model error
(vector)
How to determine d?
Estimating Discrepancy in
UKCP09 (David Sexton Talk)
• Use the multi-model ensemble from IPCC AR4 (CMIP3)
and CFMIP (models from different centres)
• For each multi-model ensemble member, find point in
HadCM3 parameter space that is closest to that member
• There is a distance between climates of this multi-model
ensemble member and this point in parameter space i.e.
effect of processes not explored by perturbed physics
ensemble
• Pool these distances over all multi-model ensemble
members
• Uses model data from the past and the future
© Crown copyright Met Office
Scottish Snow
• Clear sky
SW TOA
flux over
HadCM3
Scotland
grid point
• Most
extreme
discrepancy
found
© Crown copyright Met Office
Theme D: Questions/Topics for
Discussion
• Develop language/classification for errors and
inadequacies
• Explore ways of specifying discrepancy
• Explore alternative strategies for dealing with
errors and inadequacies
• Any more?
Blocking
The frequency of blocking events in
the perturbed physics HadCM3
ensemble (PPE_A1B, red lines) for
winter (DJF, top) and summer (JJA,
bottom) together with that
estimated from ERA40 (thick black
lines). The blocking index is
calculated following Pelly and
Hoskins (2003) and uses a variable
latitude to track the location of the
model storm track (in contrast to
other indices which used a fixed
latitude).
Murphy et al. 2009 UKCP09
Types of Errors and Inadequacies
• Incapabilities – e.g. due to resolution
• Known errors/biases in simulating mean climate and
variability, leading to metrics, iconic errors
• Errors which are known to be important for
projection, D&A, …
• Potential errors in poorly observed variables
• Known missing processes
• Unknown missing processes
Systematic Errors in All
Models
Collins et al. 2010