Transcript Assessing and predicting regional climate change

Assessing and predicting regional
climate change
Hans von Storch, Jonas Bhend and
Armineh Barkhordarian
Institute of Coastal Research, GKSS, Germany
„Assessing and predicting“
• Assessing – detection non-natural ongoing climate
change, and attribution most likely cause for such a
change. (Hasselmann, 1993)
• Assessing – if not possible: determination of
consistency of ongoing change and deflated
projections. (Bhend and von Storch, 2008)
• Predicting – not really possible at this time (except
for first examples), almost all cases are descriptions
of possible, plausible, internally consistent futures
(scenarios).
Needed tools
• Data describing homogeneously past and ongoing
change (reconstructions); when insufficient
observational evidence available, use RCMreconstructions
• Data describing plausible, internally consistent and
possible future states (scenarios) prepared by global
GCMs or (better:) regional RCMs.
• Use scenarios to guide statistical analysis if ongoing
changes are beyond the range of natural variations
(detection) and how they are best “explained“
(attribution)
Research questions
Is the observed change different There is something
from internal variability? predictable
Is anthropogenic forcing a Predictability based on
plausible explanation? plausibility possible.
Is anthropogenic forcing a Predictions for strong
necessary explanation? forcing possible
Methodical options
• Detection:
“Is the observed change different from what we expect due
to internal/natural variability alone?” – not always doable.
• Trends – are there significant trends? – no useful results.
• Consistency:
“Are the observed changes similar to what we expect from
anthropogenic forcing?”
Doable: Plausibility argument using an a priori known
forcing.
Observations
Regional detection and
attribution
Needs
Detection
Attribution
a) Homogeneous observational record
b) Estimates of natural (or internal)
level of variability (based on
observations, proxies or control run
simulations)
c) Assumption of linear overlay of
effect of different drivers
d) Realistic simulations of regional
climate
e) Estimates of the effects of different
drivers generated by model
simulations
„Significant“ trends
Often, an anthropogenic influence is assumed to be in operation
when trends are found to be „significant“.
• In many cases, the tests for assessing the significance of a trend
are false as they fail to take into account serial correlation.
• If the null-hypothesis is correctly rejected, then the conclusion to
be drawn is – if the data collection exercise would be repeated, then
we may expect to see again a similar trend.
• Example: N European warming trend April – July as part of the
seasonal cycle.
• It does not imply that the trend will continue into the future
(beyond the time scale of serial correlation).
• Example. Usually September is cooler than July.
Consistency analysis:
attribution without detection
The check of consistency of recent and ongoing trends with
projections from dynamical (or other) models represents a kind of
„attribution without detection“.
This is in particular useful, when time series of insufficient length
are available or the signal-to-noise level is too low.
The idea is to estimate the driver-related change E from a (series
of) model scenarios (or projections), and to compare this “expected
change” E with the recent trend R.
If R  E, then we may conclude that the recent change is not due to
the suspected driver, at least not completely.
DJF mean precip in the Baltic Sea catchment
Example:
Recent 30-year trend R
Trend of DJF precip
according to different
data sources.
Consistency analysis
Expected signals
• six simulations with regional coupled atmosphere-Baltic Sea
regional climate model RCAO (Rossby-Center, Sweden)
• three simulations run with HadCM3 global scenarios, three with
ECHAM4 global scenarios; 2071-2100
• two simulation exposed to A2 emission scenario, two
simulations exposed to B2 scenario; 2071-2100
• two simulations with present day GHG-levels; 1961-90
• Regional climate change in the four scenarios relatively similar.
Δ=0.05%
Regional DJF precipitation
Consistency analysis: Baltic Sea catchment
All seasons: RCAO-ECHAM B2 scenario
Pattern correlation
Intensities
precipitation
temperature
precipitation
temperature
DJF
0.84* (0.74*)
0.95* (0.73)
2.50 (2.07)
1.33 (0.66)
MAM
0.72* (0.69*)
0.83 (0.79)
3.21 (2.86)
1.15 (1.06)
JJA
-0.28
0.95*
4.42
1.85
SON
-0.59
0.60
2.23
0.71
Consistency analysis: Baltic Sea catchment
1. Consistency of the patterns of model “projections” and recent
trends is found in most seasons.
2. A major exception is precipitation in JJA and SON.
3. The observed trends in precipitation are stronger than the
anthropogenic signal suggested by the models.
4. Possible causes:
- scenarios inappropriate (false)
- drivers other than CO2 at work (industrial aerosols?)
- natural variability much larger than signal (signal-to-noise ratio 
0.2-0.5).
Perspectives
Consistency analysis requires the presence of homogeneous data
extending across several decades.
In case of insufficient observational evidence: use global re-analysis and
regional downscalings thereof.
If large-scale mean values are looked after: use GCMs; if regional details
matter: use RCMs.
Detection requires the presence of homogeneous data
extending across many decades without significant external
influences.
s
If available use multi-century GCM control runs, and
downscalings thereof (not existent so far)
Attribution requires the presence of simulations
of the changes caused several relevant drivers.
Use global modelling efforts like CMIP3; for
regional studies, available simulations insufficient
Regional Modelling
• Dynamical downscaling to obtain highresolution (10-50 km grid; 1 hourly)
description of weather stream.
- use of NCEP or ERA re-analysis allows
reconstruction of regional weather in past
decades (1960-2003)
- when global scenarios are used, regional
scenarios with better description of
space/time detail can be downscaled.
• Problem: Model may be incomplete,
“global driver” inadequate.
• Advantages: Homogeneity (if driver
homoigeneous), and very many
(unobservable) variables available.
Integration area used in GKSS reconstruction and regional
scenarios
variance
regional model
Insufficiently
resolved
Well resolved
Spatial scales
Added value
RCM
3-d vector of state
Physiographic detail
State space equation
Ψ t 1  F(Ψ t ;ηt )  εt
Observatio n equation
d t  G(Ψ t )  δt
Known large scale state
 t ,  t  model and observatio n errors
wit h
F projection
dynamical of
model
full state on largescale
G  observatio scale
n model
Ψ t*1  F(Ψ t ;ηt )
Forward integratio n :
d t*1  G (t*1 )
 Ψ t 1  Ψ t*1  K(d t*1  d t 1 )
with a suitable operator K .
Large-scale (spectral)
nudging
Barkhordarian, 2010
Mediterranean Sea Basin
Downscaling cascade
Take home …
We need to assess current change, and to do so
- by comparing against natural/internal variability (detection), and
- by comparing with (deflated) scenarios of possible futures
(attribution, consistency)
Data to this end may be generated by modelling.
- Sometimes global GCM simulations suffice
- For regional details, RCM simulations are better.
Reconstructions (and scenarios) should (can) be done with large
scale constraining.
Additional material
Hansen’s scenario published in 1988 as a
prediction up to 2010 (Hargreaves, 2010).
Allen et al.’s (2000) forecast of global temperature made in 1999. Solid line shows original
model projection. Dashed line shows prediction after reconciling climate model simulations
with the HadCRUT temperature record, using data to August 1996. Grey band shows 5-95%
uncertainty interval. Red diamond shows observed decadal mean surface temperature for
the period 1 January 2000 to 31 December 2009 referenced to the same baseline.