Transcript PPT - Atmospheric Chemistry Modeling Group

Gauging Uncertainty in Climate Models:
Implications for atmospheric chemistry and health
Loretta J. Mickley
November 29, 2012
School of Public Health
GCM = General Circulation Model,
Global Climate Model
Basic working of climate models
All climate models depend on basic physics to describe motions and
thermodynamics of the atmosphere:
E.g., vertical structure of pressure is described by hydrostatic equation
P( z )  P( z  dz )   a gdz

dP
  a g
dz
Climate models also depend on parameterizations for
many processes.
E.g., microphysics of cloud droplet formation,
vegetation processes.
Input
Tilt of earth,
geography,
greenhouse
gas content
Climate model
Physics +
Parameterized
processes
Output
Weather +
Climate
2
Three ways to run a climate model. There are many
variations!
1. Continually nudge model with
observations from satellites, surface
1950
2000
2010
2050
2. Initialize with observed sea surface
temperatures, then let run free.
Force with observed greenhouse gas trends.
3. Run model forced by
scenarios of greenhouse
gases and aerosols.
4. Start centuries earlier with estimated ocean Ts.
5. Run with carbon cycle online.
Detour: how 3-D chemistry models work.
emissions
transport
dilution
chemistry
particulate matter (PM)
and ozone pollution
population
GEOS-Chem chemical transport model:
Global 3-D model describes the transport and
chemical evolution of atmospheric pollutants
winds
Meteorology from
a climate model
Emissions + chemistry
calculated within box
Winds carry
pollutants to
other boxes.
4
Validation of present-day climate models
Temperature anomalies relative to 1901-1950 mean
Observed global mean
temperature anomalies
14 models,
58 simulations
mean of models
Models allowed to run freely, forced only by observed trends in
greenhouse gases and aerosols.
What causes spread in model response?
IPCC, 2007
What causes spread in model response?
1. Climate chaos = “Butterfly effect” = noise, interannual variability
Starting the very same model with slightly different initial conditions
will yield different day-by-day or year-by-year results.
Temperature anomalies over eastern US
9-year running means
Plot shows regional
warming due to removal of
US aerosol starting in
2010.
Red dotted curves are
results with same A1B
forcing, but different initial
conditions.
Green is same but no US
aerosol.
Modelers run ensembles
of simulations.
Mickley et al., 2011
What causes spread in model response?
2. Differences in parameterizations or model resolutions, which lead to
differences in model sensitivity to changing forcing.
Global mean temperature
response to 1% a-1 increase
in CO2 for ~20 models.
One simulation per model.
Doubled CO2
IPCC, 2001
What causes spread in model response?
3. Unknown processes, lack of understanding of basic processes.
E.g. aerosol indirect effect, aerosols
provide cloud condensation nuclei.
Range of estimates of aerosol
indirect forcing in Wm-2 in presentday atmosphere varies greatly
among many models.
IPCC 2007
By comparison, CO2 forcing ~ +1.6 W m-2
The observed atmosphere also has “noise.”
Temperature anomalies relative to 1901-1950 mean
Signal or noise?
Observed global mean
temperature anomalies
14 models,
58 simulations
mean of models
IPCC, 2007
Even a “perfect” model cannot capture observed temperatures
exactly because of climate chaos.
Hard to tell what is signal and what is noise. How long should noise
last? Years? Decades?
Another source of uncertainty in future simulations is the path of
socio-economic development.
Global mean surface temperature anomalies
Different scenarios
follow different socioeconomic paths for
developed and
developing countries.
A2 = heavy fossil fuel
B1 = alternative fuels
A1B = mix of fossil +
alternative fuels
IPCC 2007
Another source of uncertainty: abrupt climate change
Younger Dryas period=
sudden cooling, followed
by abrupt warming.
Last Ice
Age
Greenland warmed by
7oC in a few decades.
Earth system hits a
tipping point and is
thrown into new state.
Possible triggers:
• Loss of sea ice
• Reversal in ocean
currents
http://www.ncdc.noaa.gov
Future regional predictions for meteorology in A1B 2100
atmosphere show large variation across North America.
Percent change in 2100 precipitation relative to present-day
Annual
DJF
JJA
Number of models showing increasing precipitation
most
models
few
models
IPCC 2007
Exploring the uncertainty in climate models: big field of
research
Can we better characterize the spread of uncertainty in one model?
Global mean Temperature
response to 2x CO2.
Response of model
to abrupt doubling of
CO2 shows large
spread.
Results from 90K simulations, each simulation with varied parameters for
cloud processes.
Large number of simulations needed to capture spread.
Researchers colonized personal laptops across UK (like SETI project).
Stainforth et al., 2005
Climate change and air quality
For the effect of climate change on air quality, we need to think about
changes in episodic phenomena, e.g.:
• Stagnation
• Heat waves
• Wildfires
Probability of ozone exceedance
Reasons for increasing probability of ozone
exceedances at higher max temps:
Probability
• Greater stagnation + clear skies
Northeast/
mid Atlantic
in summer
• Faster chemical reactions.
• Greater biogenic emissions
maximum daily temperature (K)
Lin et al., 2001
Calculation of maximum temperatures in climate models is sensitive
to choice of parameters having to do with land cover/soil.
Lower estimate
Upper estimate
Lower and upper estimates of
JJA maximum temperatures in
2x CO2 atmosphere
Central 80% range of increases
for 44 versions of one climate
model, with varying land cover
parameters.
0
8
Forest roughness
parameter
oC
Vegetation root depth
Percent variability in
Tmax accounted for by
vegetation parameters.
Clark et al., 2010
6%
30%
50%
15
Surface ozone levels are sensitive to cold-front passage.
How will frequency of cold-front passages change in future?
Leibensperger et al., 2008
Stagnation is also strongly correlated with high PM2.5.
Correlations of PM2.5
with key meteorological
variables.
1998-2008 meteorology
+ EPA-AQS observations
Multiple linear regression coefficients for total PM2.5 on
meteorological variables. Units: μg m-3 D-1 (p-value < 0.05)
Increases in total PM2.5 on
a stagnant day vs. a nonstagnant day.
Mean PM2.5 is 2.6 μg m-3
greater on a stagnant day
Tai et al. 2010
17
Dominant meteorological modes driving PM2.5 in much of Midwest and
East are associated with cyclone passage.

Principal component (PC) decomposition of eight
meteorological variables (xk) to identify dominant
meteorological regimes that drive PM2.5 variability:
8
PC j = å a kj
k =1
xk - xk
s xk
Time series for dominant PC and deseasonalized PM2.5: Midwest in Jan 2006
3
0
-1
-3


r = -0.54
PM2.5
0
-6 -3
0
0
2
1
0
3
-2

6
1
-2 -1
PC
PM2.5
PC
6
PC
2
5
10
15
20
Observed
PM2.5
(µg m-3)
-6
25
30
Dominant PC in Midwest consists of
low T, low and rising surface
pressure, strong NW wind.
Meteorology signals the arrival of a
cold front.
Dominant PC in East is cyclone
passage, in West is maritime inflow.
Tai et al., 2011
Jan 28
Jan 30
18
Frequency (d-1)
Frequency of meteorological mode (d! 1)
0.14
0.16
0.18
0.20
Evaluation of present-day meteorological modes in AR4 climate
models reveals differences among models.
N42° W87.5°
Observed
2 sample models
NCEP/NCAR
giss_model_e_r
mpi_echam5
1985
1990
Year
1995
2000
Modeled (2 IPCC models) and observed (NCEP/NCAR) 1981-2000 time series
of frequency of dominant meteorological mode for PM2.5 in U.S. Midwest
 Some models capture both the long-term mean and variability of
meteorological mode frequency well.
 As a first step, we use only those models that capture present-day mean and
variability of frequency to predict future PM2.5
19
0.18
Y1
0.14
0.20
0.16
Y1
0.10
0.12
0.16
0.20
0.10
0.14
0.18
0.16
R2= 0.73
NMB = -8.7%
NME = 11%
giss_model_e_r
0.08
X1
R2= 0.45
NMB = -7.1%
NME = 13%
R2= 0.67
NMB = 3.5%
NME = 9.3%
0.14
0.18
X1
0.12
X1
0.16
R2= 0.85
NMB = -3.4%
NME = 5.9%
R2= 0.85
NMB = -0.47%
NME = 6.1%
iap_fgoals1_0_g
0.08
0.16
giss_aom
0.10
Y1
0.12
0.16
R2= 0.62
NMB = -5.5%
NME = 10%
0.06
0.06
X1
Y1
0.16
0.06
0.12
0.08
0.08
0.08
0.08
csiro_mk3_5
0.12
X1
0.16
R2= 0.55
NMB = 0.5%
NME = 9.4%
0.08
0.12
0.16
0.08
0.12
0.16
0.06
0.10
0.14
0.18
mpi_echam5
0.08
0.12
0.16
0.08
miub_echo_g
0.08
ipsl_cm4
0.06
ingv_echam4
0.08
0.10
0.12
0.16
0.10
0.12
X1
Y1
0.12
0.16
R2= 0.72
NMB = 0.047%
NME = 8.7%
0.08
R2= 0.63
NMB = -2.3%
NME = 9.9%
Y1
0.12
0.16
gfdl_cm2_1
0.18
0.12
X1
0.12
X1
Y1
0.14
gfdl_cm2_0
0.08
R2= 0.68
NMB = -1.3%
NME = 8.6%
Y1
0.12
0.16
0.16
Y1
0.12
0.12
X1
0.08
0.08
0.12
0.16
R2= 0.66
NMB = 0.97%
NME = 8.6%
0.08
0.16
0.16
Y1
0.12
0.08
csiro_mk3_0
0.08
0.20
cnrm_cm3
0.16
0.16
X1
R2= 0.57
NMB = -0.43%
NME = 9.4%
Y1
0.12
0.12
0.18
0.08
cccma_cgcm3_1_t63
0.14
cccma_cgcm3_1
R2= 0.72
NMB = -5.5%
NME = 8%
0.06
0.16
R2= 0.83
NMB = 1.9%
NME = 7.2%
Y1
0.12
0.08
0.08
0.12
0.16
R2= 0.82
NMB = 1.6%
NME = 7%
0.08
frequency (dY1-1)
20-year mean
Modeled
Y1
Y1
0.20
We compare the observed and AR4 modeled frequency of those
meteorological modes driving PM2.5 variability across US.
mri_cgcm2_3_2a
0.08
0.12
0.16
Observed 20-year mean frequency (d-1)
Modeled (IPCC) and observed (NCEP/NCAR) 1981-2000 mean of frequency of
dominant meteorological modes for PM2.5 in U.S. (western, central, eastern)
20
2000-2050 climate change leads to increases in annual mean PM2.5
across much of the Eastern US.
0.3
0.2
0.1
0.0
-0.1
-0.2
We choose the 9 models whose
frequency of the dominant
meteorological modes best
agrees with observations.
-0.3
1981-2065 change in period of dominant
meteorological modes for PM2.5 variability
averaged over 9 IPCC models
day
0.15
0.10
0.05
0.00
-0.05
-0.10
-0.15
Corresponding 1981-2065 change in annual
mean PM2.5 concentrations (unit: µg m-3)
mg m-3
We apply sensitivity of PM2.5 to
changing frequency of
dominant meteorological mode
in the A1B atmosphere.
Models show increased
duration of stagnation, with
corresponding increases in
annual mean PM2.5.
There is huge variation among
models.
21
Another example: wildfires in the Western US in a future climate
Our previous research has shown
that area burned depends largely on
temperature, precipitation, and
relative humidity.
Median change in key variables by
2050s relative to present-day,
calculated by 14 AR4 models.
22
Models show large variation in changes in key wildfire variables
across western US.
Projected changes in
key variables by 2050s,
relative to present-day,
across 6 ecoregions in
the Western US.
JJA only.
PNW, Pacific Northwest
CCS, California Coastal Shrub
DSW, Desert Southwest
NMS, Nevada Mountains /Semi-desert
RMF, Rocky Mountains Forest
ERM, Eastern Rocky Mountains /Great Plains.
23
Wildfire in Western United States in 2050s A1B climate.
Ratio of 2050s / present-day
Ratio of 2050s area burned / present-day area burned
median
Medians for all
regions show
increases in area
burned.
24
Given the uncertainties in climate models, how can atmospheric chemists/
epidemiologists proceed?
1. Compare signal of change to noise (interannual variability).
2. Look for those models that best capture present-day variables of
relevance to atmospheric chemistry / health. Then use projections from
only that subset of models.
3. Calculate the probabilities of specific changes. Most simply, give equal
weight to all models/ensemble members, then calculate the percentage
that show a specific effect.
25
26
Three ways to study chemistry-climate interactions.
Climate model
1. Implement chemistry scheme inside climate model!
But this is computationally expensive.
Chemistry can be simplified.
Physics +
Parameterized
processes
+Chemistry
2. Apply model chemical fields (ozone + aerosols)
to climate model
1950
2000
2010
2050
3. Archive meteorology needed to run chemistry model.