Professional Documents
Culture Documents
Abstract
We compare the cloud droplet number concentration (CDNC) prediction of the cloud processing pa-
rameterization developed by Roelofs e.a for the global aerosol-climate model ECHAM5-HAM, to a
microphysical cloud parcel model. We investigate under which circumstances their parametrization
needs some improvement. We find a good fit for aerosols in coarse mode, some deviation for accu-
mulation mode and more for Aitken mode. Updraft speeds below 0.25 m/s account for most of the
deviations, whereas aerosol radius and concentration or mass of the different components does not.
We also perform a sensitivity analysis of the cloud parcel model and suggest an alternative method
to create a parameterization.
1
cumulation and coarse mode, sea salt concentra- a plot of the ratio and the differences in radius
tion in accumulation and coarse mode and num- between two modes (figure 4) shows no obvious
ber concentrations in Aitken, accumulation and correlation.
coarse mode). For each parameter we determined To see whether the parametrization is sensi-
its range by taking the minimum and maximum tive to a particular chemical component, we plot-
values produced by ECHAM5-HAM. For this we ted the fractions of chemical components in each
used all 271 samples and we ignored 5% of the mode againt the ratio. No obvious distribution of
highest and lowest values. We varied each param- data is found, except for the sea salt fraction in the
eter in 100 exponentially increasing steps between accumulation mode. Sea salt fractions above 0.12
this minimum and maximum value, while keep- mainly yield ratios in the 0.5 - 2 range. When the
ing all other parameters in the sample constant. actual mass is plotted instead of fractions of the
In this way we obtained for each sample 13 cross chemical components, no clear distribution can be
sections of the parameter space. seen.
2
Color Red Blue Green Yellow Brown Cyan Orange Purple Pink Light blue
Zrho 1.222 1.104 1.055 1.118 1.208 1.245 1.219 1.181 1.124 1.076
Updraft 0.827 0.04 0.001 0.371 0.321 0.144 0.223 0.451 0.424 0.032
Sulfur Ait. 11.4 1.67 1.42 5.51 0.036 7.02 4.27 0.27 0.73 0.88
Sulfur Acc. 643 461 715 368 278 891 115 187 613 238
Sulfur Crs. 130 71 416 88 141 66 15 29 38 49
OC Aitken 23 5 10 44 18 68 0.8 24 19 4
OC Acc. 184 81 60 450 617 819 30 319 318 188
OC Coarse 42 28 51 41 74 64 13 59 28 64
SS Acc. 13 32 8 28 88 8 316 163 102 31
SS Coarse 1.38 1.55 0.7 1.03 5.43 1.77 1.53 6.62 5.07 5.0
Aitken conc 610 249 493 872 365 1576 32 350 429 174
Acc. conc. 159 121 101 362 275 626 48 165 201 173
Coarse conc 0.9 1.69 1.1 1 8.78 2.3 5.39 6.05 2.02 2.26
Table 1: 10 samples. Zrho is the air density in kg air m−3 . Updraft speed in ms−1 . Sulfur concentrations
are given per mode in 10−12 kg N H4 HSO4 kg−1 air. Organic carbon concentrations are given per mode
in 10−12 kg organic carbon kg−1 air. Sea salt concentrations are given in 10−12 kg sea salt kg−1 air for
accumulation mode and 10−9 kg sea salt kg−1 air for coarse mode. Number concentrations are given for each
mode in # aerosols cm−3 air.
Table 3: Number of deleted data points because of zero values for each mode and all modes together in
each season.
Table 4: Mean ratio for each mode and all modes together for each season.
3
CDNC_CPM CDNC_CPM
100
1000
10
500
1
200
0.1 100
50
0.01
CDNC_GCM 20 CDNC_GCM
0.5 1.0 5.0 10.0 50.0 100.0 1 5 10 50 100 500
CDNC_CPM CDNC_CPM
20.0
1000
10.0
5.0 500
2.0 200
1.0
100
0.5
50
0.2
CDNC_GCM CDNC_GCM
0.5 1.0 5.0 10.0 10 20 50 100 200 500
Figure 1: CDNC output from ECHAM5-HAM (GCM) versus the CPM. The dashed blue line is the 1:1
ratio.
Ratio CDNC
10.0
5.0
2.0
1.0
0.5
4
Ratio CDNC Ratio CDNC
5.0 5.0
2.0 2.0
1.0 1.0
0.5 0.5
2.0 2.0
1.0 1.0
0.5 0.5
3.0 3.0
2.0 2.0
1.5 1.5
1.0 1.0
(a) Aitken versus accumulation mode (b) Accumulation versus coarse mode
Figure 4: The differences in aerosol radius between modes versus the CDNC ratio.
5
Ratio CDNC Ratio CDNC Ratio CDNC
(a) Sulfur in aitken mode (b) Sulfur in accumulation mode (c) Sulfur in coarse mode
(d) Organic carbon in aitken mode (e) Organic carbon in accumulation (f) Organic carbon in coarse mode
mode
Ratio CDNC Ratio CDNC
5.0 5.0
3.0 3.0
2.0 2.0
1.5 1.5
1.0 1.0
Frac. SS Frac. SS
0.02 0.05 0.10 0.20 0.50 1.00 0.02 0.05 0.10 0.20 0.50 1.00
(g) Sea salt in accumulation mode (h) Sea salt in coarse mode
Figure 5: Chemical composition versus the CDNC ratio. Fractions sea salt greater than 0.12 give ratios
between 0.5 and 2.
6
Ratio CDNC Ratio CDNC Ratio CDNC
(a) Sulfur in Aitken mode (b) Sulfur in accumulation mode (c) Sulfur in coarse mode
(d) Organic carbon in Aitken mode (e) Organic carbon in accumulation (f) Organic carbon in coarse mode
mode
Ratio CDNC Ratio CDNC
5.0 5.0
3.0 3.0
2.0 2.0
1.5 1.5
1.0 1.0
Mass SS Mass SS
-10 -9 -8 -7 -10 -9 -8 -7
10 10 10 10 10 10 10 10
(g) Sea salt in accumulation mode (h) Sea salt in coarse mode
700
10.0
100.0 500
50.0 5.0
300
ðaerosols m-3 air
10.0
200
2.0
5.0
150
100 1.0
1.0
0.5
0.5
1.05 1.1 1.15 1.2 1.25 1.05 1.1 1.15 1.2 1.25 1.05 1.1 1.15 1.2 1.25
Air density Hkg air m-3 L Air density Hkg air m-3 L Air density Hkg air m-3 L
Figure 7: Air density versus CDNC for ten different parameter values.
7
700
10.0
100.0 500
50.0 5.0
300
100 1.0
1.0
0.5
0.5
0.001 0.005 0.010 0.050 0.100 0.500 1.000 0.001 0.005 0.010 0.050 0.100 0.500 1.000 0.001 0.005 0.010 0.050 0.100 0.500 1.000
Updraft speed HmsL Updraft speed HmsL Updraft speed HmsL
Figure 8: Updraft speed versus CDNC for ten different parameter values.
4 Conclusions and discussion coarse mode ) appear to have no effect and two pa-
rameters (organic carbon and sea salt concentra-
From the plots it is difficult to conclude which pa- tion in accumulation mode) little effect on CDNC
rameters affect the accuracy of the parametriza- in any of the modes. Also several parameters only
tion. Actually only one parameter, the updraft effect the CDNC in one mode. In the next analysis
speed, clearly correlates to a lower CDNC ra- one should take a closer look at the inner workings
tio deviation for higher wind speeds. This does of the model and the actual physics involved, to
not mean that the deviation is caused by just get a better estimate of the dimensions of param-
one parameter. Probably many parameters have eter space.
their specific influence on the CDNC, some more
We found many discontinuities in CDNC, but
than others. But a good start for improving
they may just be due to discretization within the
the parametrization would be to look at the low
cloud parcel model. If that is the case, it may be
wind speed range. Ultimately the full parameter
possible to create a parameterization of the entire
space should be searched for sensitive parameter
parameter space with a Taylor series. However
regions, but that will take a lot of time.
the parcel model may need to be modified to use
One must be careful not to overrate the com-
a more fine grained discretization.
pletely arbitrary 0.5 - 2 ratio range that we used
as a criterion in this research. How large a devia- We would like to see the creation of a parame-
tion from the 1:1 ratio is acceptable, depends on terization that is based on a state of the art cloud
two factors. First it depends on what one whishes parcel model, either the one studied in this re-
to know from the eventual ECHAM5-HAM model search or another one. It should make as little
runs. Second it depends on the, likely non linear, computation time saving assumptions as possible.
sensitivity of ECHAM5-HAM to inaccuracies in Next a large and well choosen set of parameter
the cloud parcel model parameterization. We rec- value combinations should be choosen. A very
ommend that future research attempts to answer large computer grid can be used to simulate each
these questions in a quantitative sense, so as to of these combination and report back the result.
create a clear criterion for the quality of a param- Based on these results, the next parameter value
eterization. combinations can be choosen, etc. The result can
Several things can learned from the sensitivity be compressed in something like a Taylor series
analysis of the parcel model. We believe that it and ECHAM5-HAM should use this series in stead
may be possible to create a better parametriza- of the current parameterization.
tion if one performs a large scale semi automatic We believe there is a large potential for im-
study of the full parameter space in a more effi- provement of cloud simulation and parameteri-
cient manner than presented above. We provide zation in distributed computing. Anderson and
several suggestions. Fedak (2006) show that for one large existing
We investigated thirteen parameters, but we distributed computing project the host pool of
did not look at the level of independence between 330.000 computers provides processing at a sus-
them. Therefore we can not state that the param- tained rate of 95.5 TFLOPS and a potential stor-
eter space is 13 dimensional. In fact, two param- age space of 7.74 Petabytes that can be accessed at
eters (sulfur and organic carbon concentration in 5.27 Terabytes per second. Because of the present
8
700
10.0
100.0 500
50.0 5.0
300
10.0
200
2.0
5.0
150
100 1.0
1.0
0.5
0.5
-13 -13 -12 -12 -12 -13 -13 -12 -12 -12 -13 -13 -12 -12 -12
2 ´ 10 5 ´ 10 1 ´ 10 2 ´ 10 5 ´ 10 2 ´ 10 5 ´ 10 1 ´ 10 2 ´ 10 5 ´ 10 2 ´ 10 5 ´ 10 1 ´ 10 2 ´ 10 5 ´ 10
kg HNH4 L HSO4 kg air HAitkenL kg HNH4 L HSO4 kg air HAitkenL kg HNH4 L HSO4 kg air HAitkenL
(a) Sulfur concentration in Aitken (b) Sulfur concentration in Aitken (c) Sulfur concentration in Aitken
mode versus CDNC in Aitken mode. mode versus CDNC in accumulation mode versus CDNC in Coarse mode.
mode.
700
10.0
100.0 500
50.0 5.0
300
ðaerosols m-3 air
10.0
200
2.0
5.0
150
100 1.0
1.0
0.5
0.5
-10 -10 -10 -10 -10 -10 -9 -10 -10 -10 -10 -10 -10 -9 -10 -10 -10 -10 -10 -10 -9
1.0 ´ 10 1.5 ´ 102.0 ´ 10 3.0 ´ 10 5.0 ´ 10 7.0 ´ 10 1.0 ´ 10 1.0 ´ 10 1.5 ´ 102.0 ´ 10 3.0 ´ 10 5.0 ´ 10 7.0 ´ 10 1.0 ´ 10 1.0 ´ 10 1.5 ´ 102.0 ´ 10 3.0 ´ 10 5.0 ´ 10 7.0 ´ 10 1.0 ´ 10
kg HNH4 L HSO4 kg air HAcc.L kg HNH4 L HSO4 kg air HAcc.L kg HNH4 L HSO4 kg air HAcc.L
(d) Sulfur concentration in accumu- (e) Sulfur concentration in accumula- (f) Sulfur concentration in accumu-
lation mode versus CDNC in Aitken tion mode versus CDNC in accumu- lation mode versus CDNC in Coarse
mode. lation mode. mode.
700
10.0
100.0 500
50.0 5.0
300
ðaerosols m-3 air
10.0
200
2.0
5.0
150
100 1.0
1.0
0.5
0.5
-11 -11 -11 -11 -11 -10 -10 -10 -11 -11 -11 -11 -11 -10 -10 -10 -11 -11 -11 -11 -11 -10 -10 -10
1.5 ´ 10
2.0 ´ 10 3.0 ´ 10 5.0 ´ 107.0 ´ 101.0 ´ 10 1.5 ´ 10
2.0 ´ 10 1.5 ´ 10
2.0 ´ 10 3.0 ´ 10 5.0 ´ 107.0 ´ 10 1.0 ´ 10 1.5 ´ 10
2.0 ´ 10 1.5 ´ 10
2.0 ´ 10 3.0 ´ 10 5.0 ´ 107.0 ´ 10 1.0 ´ 10 1.5 ´ 10
2.0 ´ 10
kg HNH4 L HSO4 kg air HCoarseL kg HNH4 L HSO4 kg air HCoarseL kg HNH4 L HSO4 kg air HCoarseL
(g) Sulfur concentration in Coarse (h) Sulfur concentration in Coarse (i) Sulfur concentration in Coarse
mode versus CDNC in Aitken mode. mode versus CDNC in accumulation mode versus CDNC in Coarse mode.
mode.
Figure 9: Sulfur volume fraction of each mode versus CDNC for ten different parameter values.
9
700
10.0
100.0 500
50.0 5.0
300
10.0
200
2.0
5.0
150
100 1.0
1.0
0.5
0.5
-12 -11 -11 -11 -12 -11 -11 -11 -12 -11 -11 -11
5 ´ 10 1 ´ 10 2 ´ 10 5 ´ 10 5 ´ 10 1 ´ 10 2 ´ 10 5 ´ 10 5 ´ 10 1 ´ 10 2 ´ 10 5 ´ 10
-1 -1 -1
kg OC kg air HAitkenL kg OC kg air HAitkenL kg OC kg air HAitkenL
(a) Organic carbon concentration in (b) Organic carbon concentration in (c) Organic carbon concentration in
Aitken mode versus CDNC in Aitken Aitken mode versus CDNC in accu- Aitken mode versus CDNC in Coarse
mode. mulation mode. mode.
700
10.0
100.0 500
50.0 5.0
300
ðaerosols m-3 air
10.0
200
2.0
5.0
150
100 1.0
1.0
0.5
0.5
-10 -10 -10 -10 -10 -10 -10 -10 -10 -10 -10 -10 -10 -10 -10 -10 -10 -10
1.0 ´ 10 1.5 ´ 102.0 ´ 10 3.0 ´ 10 5.0 ´ 10 7.0 ´ 10 1.0 ´ 10 1.5 ´ 102.0 ´ 10 3.0 ´ 10 5.0 ´ 10 7.0 ´ 10 1.0 ´ 10 1.5 ´ 102.0 ´ 10 3.0 ´ 10 5.0 ´ 10 7.0 ´ 10
-1 -1 -1
kg OC kg air HAcc.L kg OC kg air HAcc.L kg OC kg air HAcc.L
(d) Organic carbon concentration in (e) Organic carbon concentration in (f) Organic carbon concentration in
accumulation mode versus CDNC in accumulation mode versus CDNC in accumulation mode versus CDNC in
Aitken mode. accumulation mode. Coarse mode.
700
10.0
100.0 500
50.0 5.0
300
ðaerosols m-3 air
10.0
200
2.0
5.0
150
100 1.0
1.0
0.5
0.5
-11 -11 -11 -11 -11 -11 -10 -11 -11 -11 -11 -11 -11 -10 -11 -11 -11 -11 -11 -11 -10
1.0 ´ 10 1.5 ´ 10 2.0 ´ 10 3.0 ´ 10 5.0 ´ 10 7.0 ´ 10 1.0 ´ 10 1.0 ´ 10 1.5 ´ 10 2.0 ´ 10 3.0 ´ 10 5.0 ´ 10 7.0 ´ 10 1.0 ´ 10 1.0 ´ 10 1.5 ´ 10 2.0 ´ 10 3.0 ´ 10 5.0 ´ 10 7.0 ´ 10 1.0 ´ 10
-1 -1 -1
kg OC kg air HCoarseL kg OC kg air HCoarseL kg OC kg air HCoarseL
(g) Organic carbon concentration in (h) Organic carbon concentration in (i) Organic carbon concentration in
Coarse mode versus CDNC in Aitken Coarse mode versus CDNC in accu- Coarse mode versus CDNC in Coarse
mode. mulation mode. mode.
Figure 10: Organic carbon volume fraction of each mode versus CDNC for ten different parameter values.
10
700
10.0
100.0 500
50.0 5.0
300
10.0
200
2.0
5.0
150
100 1.0
1.0
0.5
0.5
-11 -11 -11 -11 -11 -11 -10 -11 -11 -11 -11 -11 -11 -10 -11 -11 -11 -11 -11 -11 -10
1.0 ´ 10 1.5 ´ 10
2.0 ´ 10 3.0 ´ 10 5.0 ´ 107.0 ´ 101.0 ´ 10 1.0 ´ 10 1.5 ´ 10
2.0 ´ 10 3.0 ´ 10 5.0 ´ 107.0 ´ 101.0 ´ 10 1.0 ´ 10 1.5 ´ 10
2.0 ´ 10 3.0 ´ 10 5.0 ´ 107.0 ´ 101.0 ´ 10
-1 -1 -1
kg Sea Salt kg air HAcc.L kg Sea Salt kg air HAcc.L kg Sea Salt kg air HAcc.L
(a) Sea salt concentration in accumu- (b) Sea salt concentration in accumu- (c) Sea salt concentration in accumu-
lation mode versus CDNC in Aitken lation mode versus CDNC in accu- lation mode versus CDNC in Coarse
mode. mulation mode. mode.
700
10.0
100.0 500
50.0 5.0
300
ðaerosols m-3 air
10.0
200
2.0
5.0
150
100 1.0
1.0
0.5
0.5
-9 -9 -9 -9 -9 -9 -9 -9 -9 -9 -9 -9 -9 -9 -9 -9 -9 -9
1.0 ´ 10 1.5 ´ 10
2.0 ´ 10 3.0 ´ 10 5.0 ´ 107.0 ´ 10 1.0 ´ 10 1.5 ´ 10
2.0 ´ 10 3.0 ´ 10 5.0 ´ 107.0 ´ 10 1.0 ´ 10 1.5 ´ 10
2.0 ´ 10 3.0 ´ 10 5.0 ´ 107.0 ´ 10
-1 -1 -1
kg Sea Salt kg air HCoarseL kg Sea Salt kg air HCoarseL kg Sea Salt kg air HCoarseL
(d) Sea salt concentration in Coarse (e) Sea salt concentration in Coarse (f) Sea salt concentration in Coarse
mode versus CDNC in Aitken mode. mode versus CDNC in accumulation mode versus CDNC in Coarse mode.
mode.
Figure 11: Sea salt volume fraction of each mode versus CDNC for ten different parameter values.
11
700
10.0
100.0 500
50.0 5.0
300
10.0
200
2.0
5.0
150
100 1.0
1.0
0.5
0.5
150 200 300 500 700 1000 150 200 300 500 700 1000 150 200 300 500 700 1000
ðaerosols cm-3 air HAitkenL ðaerosols cm-3 air HAitkenL ðaerosols cm-3 air HAitkenL
(a) Number concentration in Aitken (b) Number concentration in Aitken (c) Number concentration in Aitken
mode versus CDNC in Aitken mode. mode versus CDNC in accumulation mode versus CDNC in Coarse mode.
mode.
700
10.0
100.0 500
50.0 5.0
300
ðaerosols m-3 air
10.0
200
2.0
5.0
150
100 1.0
1.0
0.5
0.5
100 150 200 300 500 100 150 200 300 500 100 150 200 300 500
ðaerosols cm-3 air HAccL ðaerosols cm-3 air HAccL ðaerosols cm-3 air HAccL
(d) Number concentration in accu- (e) Number concentration in accu- (f) Number concentration in accu-
mulation mode versus CDNC in mulation mode versus CDNC in ac- mulation mode versus CDNC in
Aitken mode. cumulation mode. Coarse mode.
700
10.0
100.0 500
50.0 5.0
300
ðaerosols m-3 air
10.0
200
2.0
5.0
150
100 1.0
1.0
0.5
0.5
1.0 1.5 2.0 3.0 5.0 1.0 1.5 2.0 3.0 5.0 1.0 1.5 2.0 3.0 5.0
ðaerosols cm-3 air HCoarseL ðaerosols cm-3 air HCoarseL ðaerosols cm-3 air HCoarseL
(g) Number concentration in Coarse (h) Number concentration in Coarse (i) Number concentration in Coarse
mode versus CDNC in Aitken mode. mode versus CDNC in accumulation mode versus CDNC in Coarse mode.
mode.
Figure 12: Number concentration in each mode versus CDNC for ten different parameter values.
12
popularity of climate change, the easy to install Proceedings of the IEEE International Sympo-
software for distributed computing and the large sium on Cluster Computing and the Grid (CC-
percentage of people who have not heard about it GRID06), 2006.
yet, we believe it is feasable to harnass a lot of
computer power at a very low cost for this pur- G. Roelofs, P. Stier, J. Feichter, E. Vignati,
pose. and J. Wilson. Aerosol activation and cloud
processing in the global aerosol-climate model
ECHAM5-HAM. Atmospheric Chemistry and
References Physics, 6(9):2389–2399, 2006.
13