resolution on chemical transport modeling€¦ · Using SEAC4RS observations to understand impact...
1
Using SEAC 4 RS observations to understand impact of grid resolution on chemical transport modeling K. Yu 1 ([email protected]), D. J. Jacob 1 , P. S. Kim 1 , K. R. Travis 1 , L. Zhu 1 , J. A. Fisher 2 , R. Yantosca 1 , and the SEAC 4 RS Science Team 1 Harvard University, Cambridge, Massachusetts; 2 University of Wollongong, Australia MOTIVATION GEOS-Chem recently introduced the capability to run at 0.25 o x0.3125 o resolution over several nested domains. Given the computational cost of running a high resolution model, it is important to quantify the benefits of increased resolution and identify applications where a high resolution model is necessary and applications where running at coarse resolution is sufficient. SEAC 4 RS offers an opportunity to constrain resolution related uncertainties in the GEOS-Chem model and quantify the added value of a high resolution chemical transport model. Overall model performance Differences in model performance for the low and high resolution simulations are generally small when measured in terms of Pearson’s correlation coefficient and root mean square error (RMSE). As computational power increases to allow higher resolution atmospheric models, it is important to quantify the benefits of increasing model resolution. The necessary resolution is highly dependent on the application of the model: modeling the continental background appears to not require as high resolution as urban areas. Commonly used model evaluation metrics are unable to distinguish between displacement and amplitude errors and therefore may undervalue high resolution models. Future work will provide more quantitative evaluations of model performance at different grid resolutions. Figure 1: Taylor Diagram of model performance for various species normalized to observations. The 4 o x5 o simulation generally captures less standard deviation than the 0.25 o x0.3125 o simulations. The 0.25x0.3125 simulation often does not improve correlations because we are comparing points averaged over smaller areas. When the 0.25 o x0.3125 o output is averaged to the 4 o x5 o grid, there are small improvements in model performance for some species. Isoprene and MVK+MACR measurements from Wisthaler. CH2O measurements from CAMS (PI: Fried), PAN measurements from GTCIMS (PI: Huey). Methodology for comparisons Differences in model performance are larger when looking at urban plumes separately from the background. Ozone concentrations in regions of high NOx emissions will be more impacted by resolution because a coarse resolution model is unable to resolve intense point sources. Commonly used model evaluation metrics suggest that increased resolution improves model performance for a non-linear species like ozone, but not for CO which has simpler chemistry.. How does grid resolution affect a model? Atmospheric processes span a wide range of scales, many of which cannot be resolved on typical grid sizes used in global atmospheric chemistry models. Sub- grid interactions between species can induce biases in the model when the model only has access to the grid volume mean values. Previous work found that a higher resolution model produces less ozone in polluted boundary layers and exports more ozone precursors 1 , and that peak ozone can be underestimated in coarse resolution models 2 . In addition to non-linear chemistry, grid resolution can also impact numerical diffusion as well as emissions and other inputs. Left: Spatial scale of NO X emissions in GEOS-Chem. Right: Figure 1 from Finlayson-Pitts and Pitts (1993). O 3 isopleths for different levels of NO x and VOCs, showing non-linearity of O 3 response to emissions. Background vs. urban Figure 4: Model curtain along flight track overplotted with in-situ observations. Ozone and NO observations from ESRL (PI: Ryerson). Ability to capture tails of distribution For certain modeling applications, it is important to capture not only the mean, but also the tails of the distribution of the actual values (for example, modeling ozone exceedances). Even if a high resolution model and a coarse resolution model have similar ability to capture the center of a distribution, the high resolution will be better able to capture the tail. Detailed features in high resolution model Although when compared on a point by point basis, a higher resolution model doesn’t always perform better than a coarse resolution model, it undeniably produces more smaller scale features that cannot be resolved in a coarse resolution model. If a small-scale feature in the model is displaced spatially or temporally from its actual location, the model will not compare favorably to observations using traditional model evaluation metrics. Future will work use LIDAR data to evaluate the model’s ability to reproduce small scale features. Correlations between species Correlations between different chemical species can be used to constrain sources and processes contributing to the burden of a species. Even if two models produce similar concentrations of a species, it may be the result of different processes. References Finlayson-Pitts, B. J., and J. N. Pitts Jr. "Atmospheric chemistry of tropospheric ozone formation: scientific and regulatory implications." Air & Waste 43.8 (1993): 1091-1100. 1 Wild, Oliver, and Michael J. Prather. "Global tropospheric ozone modeling: Quantifying errors due to grid resolution." Journal of Geophysical Research: Atmospheres (1984–2012) 111.D11 (2006). 2 Kumar, Naresh, and Armistead G. Russell. "Multiscale air quality modeling of the northeastern United States." Atmospheric Environment 30.7 (1996): 1099-1116. Taylor Diagram Standard deviation 0 1 2 3 0 1 2 3 0.5 1 1.5 2 2.5 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.95 0.99 Correlation NO CO PAN O3 NO2 CH2O ISOP SO2 SO4 MVK+MACR ISOPN 0.25x0.3125 0.25x0.3125 averaged to 4x5 4x5 Data type DC8 60s merges GEOS-Chem global GEOS-Chem nested Grid None 4 o x5 o 60 min time- step 0.25 o x0.312 5 o 10 min time-step GEOS-Chem was run at both global 4x5 resolution and N. America nested 0.25x0.3125 resolution for the SEAC4RS campaign period with data saved out at 60s timesteps along the flight track. Flight data is averaged to the GEOS-Chem grids for computing correlations. In all figures, ‘avg’ denotes GEOS-Chem output at 0.25x0.3125 resolution averaged to the 4x5 grid. Metric Resolution O3 CO NOx SO2 SO4 Urban Bkgd Urban Bkgd Urban Bkgd Urban Bkgd Urban Bkgd NMB (%) 4x5 22.7 17.6 -25.7 -8.14 -62.0 30.7 -29.7 67.3 -20.9 -25.3 0.25x0.3125 7.72 17.1 -37.4 -6.01 91.7 32.2 -42.4 48.4 -25.4 -13.1 Avg 17.9 21.0 -17.7 -3.17 32.4 32.8 -23.9 60.6 -18.1 -13.1 NME (%) 4x5 38.2 25.4 35.2 24.9 77.0 67.0 77.5 125 38.9 59.2 0.25x0.3125 32.4 26.9 46.7 26.5 213 76.4 76.6 119 49.1 56.6 Avg 35.4 27.2 36.9 24.9 115 66.8 72.4 123 39.2 56.2 RMSE (ppb, ppt, ug/ m 3 ) 4x5 25.4 18.4 120 32.8 2.38 0.192 711 431 1.10 1.32 0.25x0.325 29.3 20.2 303 36.1 8.45 0.291 1040 482 1.28 1.38 Avg 24.6 19.2 121 33.8 4.20 0.194 651 481 1.03 1.36 Table 1: NMB, NME, and RMSE for various species compared to observations. Urban refers to data points where NO 2 > 4 ppbv or NO 2 /No y > 0.4. Background (Bkgd) refers to data after filtering for fire and urban plumes. O 3 and NOx measurements from ESRL (PI: Ryerson), SO 2 from GTCIMS (PI: Huey) and SO 4 from SAGA (PI: Dibbs). 0-2km 2-4km 4-6km 6-8km 8-10km 10-12km 50 100 150 200 O3 (ppbv) Altitude range (km) 4x5 0.25x0.3125 avg field 0-2km 2-4km 4-6km 6-8km 8-10km 10-12km 0 5 10 SO4 (ug/m^3) Altitude range (km) 4x5 0.25x0.3125 avg field Figure 2: Vertical profile for ozone and SO 4 binned to 2 km ranges. The 4x5 model does not reproduce the high end of the values seen in the data. O3 measurements from ESRL (PI: Ryerson) and SO4 measurements from SAGA (PI: Dibbs). 0 50 100 150 200 250 0 50 150 250 CO (ppbv) O3 (ppbv) GEOS-Chem 4x5 r = 0.791 slope = 0.334 0 50 100 150 200 250 0 50 150 250 CO (ppbv) O3 (ppbv) GEOS-Chem 0.25x0.3125 r = 0.702 slope = 0.303 0 50 100 150 200 250 0 50 150 250 CO (ppbv) O3 (ppbv) DC8 60s merge r = 0.522 slope = 0.439 Figure 3: Ozone-CO correlations for data points in the planetary boundary layer. CO observations from DACOM (PI: Diskin). O 3 - 20130916 NO - 20130916 O 3 - 20130916 NO - 20130916 ppbv ppbv ppbv ppbv
Transcript of resolution on chemical transport modeling€¦ · Using SEAC4RS observations to understand impact...
Using SEAC4RS observations to understand impact of grid resolution on chemical transport modeling
K. Yu1 ([email protected]), D. J. Jacob1, P. S. Kim1, K. R. Travis1, L. Zhu1, J. A. Fisher2, R. Yantosca1, and the SEAC4RS Science Team 1Harvard University, Cambridge, Massachusetts; 2University of Wollongong, Australia
MOTIVATION GEOS-Chem recently introduced the capability to run at 0.25ox0.3125o resolution over several nested domains. Given the computational cost of running a high resolution model, it is important to quantify the benefits of increased resolution and identify applications where a high resolution model is necessary and applications where running at coarse resolution is sufficient. SEAC4RS offers an opportunity to constrain resolution related uncertainties in the GEOS-Chem model and quantify the added value of a high resolution chemical transport model.
Overall model performance
Differences in model performance for the low and high resolution simulations are generally small when measured in terms of Pearson’s correlation coefficient and root mean square error (RMSE).
As computational power increases to allow higher resolution atmospheric models, it is important to quantify the benefits of increasing model resolution. The necessary resolution is highly dependent on the application of the model: modeling the continental background appears to not require as high resolution as urban areas. Commonly used model evaluation metrics are unable to distinguish between displacement and amplitude errors and therefore may undervalue high resolution models. Future work will provide more quantitative evaluations of model performance at different grid resolutions.
Figure 1: Taylor Diagram of model performance for various species normalized to observations. The 4ox5o simulation generally captures less standard deviation than the 0.25ox0.3125o simulations. The 0.25x0.3125 simulation often does not improve correlations because we are comparing points averaged over smaller areas. When the 0.25ox0.3125o output is averaged to the 4ox5o grid, there are small improvements in model performance for some species. Isoprene and MVK+MACR measurements from Wisthaler. CH2O measurements from CAMS (PI: Fried), PAN measurements from GTCIMS (PI: Huey).
Methodology for comparisons
Differences in model performance are larger when looking at urban plumes separately from the background. Ozone concentrations in regions of high NOx emissions will be more impacted by resolution because a coarse resolution model is unable to resolve intense point sources. Commonly used model evaluation metrics suggest that increased resolution improves model performance for a non-linear species like ozone, but not for CO which has simpler chemistry..
How does grid resolution affect a model? Atmospheric processes span a wide range of scales, many of which cannot be resolved on typical grid sizes used in global atmospheric chemistry models. Sub-grid interactions between species can induce biases in the model when the model only has access to the grid volume mean values. Previous work found that a higher resolution model produces less ozone in polluted boundary layers and exports more ozone precursors1, and that peak ozone can be underestimated in coarse resolution models2. In addition to non-linear chemistry, grid resolution can also impact numerical diffusion as well as emissions and other inputs.
Left: Spatial scale of NOX emissions in GEOS-Chem. Right: Figure 1 from Finlayson-Pitts and Pitts (1993). O3 isopleths for different levels of NOx and VOCs, showing non-linearity of O3 response to emissions.
Background vs. urban
Figure 4: Model curtain along flight track overplotted with in-situ observations. Ozone and NO observations from ESRL (PI: Ryerson).
Ability to capture tails of distribution For certain modeling applications, it is important to capture not only the mean, but also the tails of the distribution of the actual values (for example, modeling ozone exceedances). Even if a high resolution model and a coarse resolution model have similar ability to capture the center of a distribution, the high resolution will be better able to capture the tail.
Detailed features in high resolution model Although when compared on a point by point basis, a higher resolution model doesn’t always perform better than a coarse resolution model, it undeniably produces more smaller scale features that cannot be resolved in a coarse resolution model. If a small-scale feature in the model is displaced spatially or temporally from its actual location, the model will not compare favorably to observations using traditional model evaluation metrics. Future will work use LIDAR data to evaluate the model’s ability to reproduce small scale features.
Correlations between species
Correlations between different chemical species can be used to constrain sources and processes contributing to the burden of a species. Even if two models produce similar concentrations of a species, it may be the result of different processes.
References Finlayson-Pitts, B. J., and J. N. Pitts Jr. "Atmospheric chemistry of tropospheric ozone formation: scientific and regulatory implications." Air & Waste 43.8 (1993): 1091-1100. 1Wild, Oliver, and Michael J. Prather. "Global tropospheric ozone modeling: Quantifying errors due to grid resolution." Journal of Geophysical Research: Atmospheres (1984–2012) 111.D11 (2006). 2Kumar, Naresh, and Armistead G. Russell. "Multiscale air quality modeling of the northeastern United States." Atmospheric Environment 30.7 (1996): 1099-1116.
Taylor Diagram
Sta
ndar
d de
viat
ion
0 1 2 3
01
23
0.5
1
1.5
2
2.5
0.1 0.20.3
0.40.5
0.6
0.7
0.8
0.9
0.95
0.99
Correlation
NO
CO
PAN
O3
NO2
CH2O
ISOPSO2
SO4
MVK+MACR
ISOPN
0.25x0.31250.25x0.3125 averaged to 4x54x5
Data type
DC8 60s merges
GEOS-Chem global
GEOS-Chem nested
Grid None
4ox5o 60 min time-step
0.25ox0.3125o 10 min time-step
GEOS-Chem was run at both global 4x5 resolution and N. America nested 0.25x0.3125 resolution for the SEAC4RS campaign period with data saved out at 60s timesteps along the flight track. Flight data is averaged to the GEOS-Chem grids for computing correlations. In all figures, ‘avg’ denotes GEOS-Chem output at 0.25x0.3125 resolution averaged to the 4x5 grid.
Metric Resolution O3 CO NOx SO2 SO4 Urban Bkgd Urban Bkgd Urban Bkgd Urban Bkgd Urban Bkgd
NMB (%)
4x5 22.7 17.6 -25.7 -8.14 -62.0 30.7 -29.7 67.3 -20.9 -25.3 0.25x0.3125 7.72 17.1 -37.4 -6.01 91.7 32.2 -42.4 48.4 -25.4 -13.1 Avg 17.9 21.0 -17.7 -3.17 32.4 32.8 -23.9 60.6 -18.1 -13.1
NME (%)
4x5 38.2 25.4 35.2 24.9 77.0 67.0 77.5 125 38.9 59.2 0.25x0.3125 32.4 26.9 46.7 26.5 213 76.4 76.6 119 49.1 56.6 Avg 35.4 27.2 36.9 24.9 115 66.8 72.4 123 39.2 56.2
RMSE (ppb, ppt, ug/m3)
4x5 25.4 18.4 120 32.8 2.38 0.192 711 431 1.10 1.32 0.25x0.325 29.3 20.2 303 36.1 8.45 0.291 1040 482 1.28 1.38 Avg 24.6 19.2 121 33.8 4.20 0.194 651 481 1.03 1.36
Table 1: NMB, NME, and RMSE for various species compared to observations. Urban refers to data points where NO2 > 4 ppbv or NO2/Noy > 0.4. Background (Bkgd) refers to data after filtering for fire and urban plumes. O3 and NOx measurements from ESRL (PI: Ryerson), SO2 from GTCIMS (PI: Huey) and SO4 from SAGA (PI: Dibbs).
0-2km
2-4km
4-6km
6-8km
8-10km
10-12km
50 100 150 200O3 (ppbv)
Alti
tude
rang
e (k
m)
4x5
0.25x0.3125
avg
field
0-2km
2-4km
4-6km
6-8km
8-10km
10-12km
0 5 10SO4 (ug/m^3)
Alti
tude
rang
e (k
m)
4x5
0.25x0.3125
avg
field
Figure 2: Vertical profile for ozone and SO4 binned to 2 km ranges. The 4x5 model does not reproduce the high end of the values seen in the data. O3 measurements from ESRL (PI: Ryerson) and SO4 measurements from SAGA (PI: Dibbs).
0 50 100 150 200 250
050
150
250
CO (ppbv)
O3
(ppb
v)
GEOS-Chem 4x5
r = 0.791slope = 0.334
0 50 100 150 200 250
050
150
250
CO (ppbv)
O3
(ppb
v)
GEOS-Chem 0.25x0.3125
r = 0.702slope = 0.303
0 50 100 150 200 250
050
150
250
CO (pbbv)
O3
(ppb
v)
GEOS-Chem 0.25x0.3125 averaged to 4x5 grid
r = 0.779slope = 0.315
0 50 100 150 200 250
050
150
250
CO (ppbv)
O3
(ppb
v)
DC8 60s merge
r = 0.522slope = 0.439
0 50 100 150 200 250
050
150
250
CO (ppbv)
O3
(ppb
v)
GEOS-Chem 4x5
r = 0.791slope = 0.334
0 50 100 150 200 250
050
150
250
CO (ppbv)
O3
(ppb
v)
GEOS-Chem 0.25x0.3125
r = 0.702slope = 0.303
0 50 100 150 200 250
050
150
250
CO (pbbv)
O3
(ppb
v)
GEOS-Chem 0.25x0.3125 averaged to 4x5 grid
r = 0.779slope = 0.315
0 50 100 150 200 250
050
150
250
CO (ppbv)
O3
(ppb
v)
DC8 60s merge
r = 0.522slope = 0.439
Figure 3: Ozone-CO correlations for data points in the planetary boundary layer. CO observations from DACOM (PI: Diskin).
O3 - 20130916 NO - 20130916
O3 - 20130916 NO - 20130916
ppbv ppbv
ppbv ppbv
