An evaluation of the CMAQ reproducibility of satellite troposph ric NO 2 column observations at di ff erent local time ver East Asia

Introduction Conclusions References


Conclusions References
Tables Figures

Introduction
Accurate modeling of atmospheric chemistry is needed for the current understanding and future prediction of the atmospheric environment from the perspectives of not only air quality but also climate.In recent years, East Asian countries are recognized to have brought a continuous increase in the emissions of nitrogen oxides (NO x = NO + NO 2 ) (e.g., Ohara et al., 2007), which play a central role in tropospheric chemistry.Thus, an evaluation of the ability of NO x simulations in East Asia can provide us with a critical constraint for improving atmospheric chemistry modeling.On regional to global scales, Introduction

Conclusions References
Tables Figures

Back Close
Full  2006) compared tropospheric NO 2 vertical column densities (VCDs) from 17 different international global atmospheric chemistry models with observations from the GOME (Global Ozone Monitoring Experiment) satellite instrument and found that the models tend to underestimate the satellite data in industrial regions, especially over Central Eastern China (CEC).Over the same region, GOME tropospheric NO 2 VCDs were compared with those simulated by the Community Multi-scale Air Quality (CMAQ) model using the Regional Emission inventory in Asia (REAS) Version 1 (Ohara et al., 2007), showing an underestimation by the model (Uno et al., 2007).The identified underestimation was then attributed mainly to the limited accuracy of basic energy statistical data, emission factors, and socio-economic data used to construct the emission inventory.Han et al. (2009) also studied GOME vs. CMAQ comparisons using the 2001 ACE-ASIA (Asia Pacific Regional Aerosol Characterization Experiment) emission inventory and found a large discrepancy between GOME and CMAQ tropospheric NO 2 VCDs in fall and winter.In particular, the CMAQ tropospheric NO 2 VCDs were low by ∼ 57.3 % over North China.In these ways, absolute values of tropospheric NO 2 VCDs, derived from a single set of satellite data, have been mainly discussed so far, limiting discussion of other aspects, such as the diurnal variation.

Conclusions References
Tables Figures

Back Close
Full satellite data sets and it was concluded that biases between satellite and MAX-DOAS VCDs are insignificant for SCIAMACHY, OMI, and GOME-2.
Here the present work considers the validation results from Irie et al. (2012) and evaluates the reproducibility of satellite tropospheric NO 2 VCD data by the chemical transport model CMAQ, from the viewpoint of the diurnal variation of tropospheric NO 2 VCDs.For CMAQ simulations, we use the REAS Version 2 emission inventory (Kurokawa et al., 2013).Various sensitivity simulations, for example at different spatial resolutions (80, 40, 20, and 10 km), are examined for a detailed evaluation.

CMAQ
The present study evaluates CMAQ Version 4.7.1 (Byun and Schere, 2006), which is driven by meteorological fields generated by the WRF (Weather Research and Forecasting) Version 3.3 model (Skamarock and Klemp, 2008) with 2007 NCEP (National Centers for Environmental Prediction) Final Analysis data (ds083.2).We use the SAPRC-99 (Statewide Air Pollution Research Center-99) gas-phase atmospheric chemical mechanisms with 72 chemical species and 214 chemical reactions and 30 photochemical reactions for the gas-phase chemistry.The aerosol module AERO5 (Carlton et al., 2010) is used for aerosols.Cloud and aqueous chemistry and dry deposition are represented by the Asymmetric Convective Model (ACM) cloud processor using the ACM methodology to compute convective mixing.For the advection, the piecewise parabolic method (Colella and Woodward, 1984) is employed.ACM version 2 (ACM2) vertical diffusion instrumented for in-line calculation of emissions is used for the turbulent diffusion.The lateral boundary conditions are taken from monthly mean values calculated by the CHemical AGCM for Study of atmospheric Environment and Radiative forcing (CHASER) (Sudo et al., 2002).We use default profiles as initial conditions and the first 30 days are treated as a spin up period.
As anthropogenic emissions over East Asia, we adopt the REAS Version 2 data (Kurokawa et al., 2013).Biomass burning and biogenic emission data are taken from Introduction

Conclusions References
Tables Figures

Back Close
Full For the sensitivity simulations, we test 4 different horizontal resolutions of 80 (as the base case), 40, 20, and 10 km (Table 1), which result in grid cell numbers of 95 × 75, 110 × 88, 184 × 132, and 292 × 182, respectively.A one-way nesting method is adopted for each inner domain simulation.The corresponding domains are shown in Fig. 1.We also test the sensitivity of soil NO x emission by running models with or without it (Table 1).As other sensitivity tests, we change the emission strength by ±20 % over the whole of the model domain.This number is chosen as an increase or decrease by 20 % can roughly compensate differences found between wintertime tropospheric NO 2 VCDs from CMAQ base case simulations and satellite observations over CEC, as shown later.Additional sensitivity tests are performed in vertical layers that are represented by 37 or 14 sigma-pressure coordinated layers from the surface to 50 hPa with the first layer height being around 20 m.Also, the impact of reducing the emission injection height by half and the impact of the diurnal variation of emissions are investigated assuming a diurnal variation pattern that was best estimated by Lin et al. (2010).In this diurnal variation pattern, the ratio of hourly emissions to the daily mean has a minimum at ∼ 0.6 (at 00:00-04:00 LT) and maximum at ∼ 1.2 (at 09:00-19:00 LT).
We set the model to output data every 1 h for each simulation.For all grids, the model values are interpolated over time to estimate tropospheric NO 2 VCDs at local times of satellite observations.Introduction

Conclusions References
Tables Figures

Back Close
Full  products for SCIAMACHY and GOME-2) (Boersma et al., 2004(Boersma et al., , 2007(Boersma et al., , 2011)).For polluted situations, the error in the satellite tropospheric NO 2 VCD data was estimated to be ∼ 1 × 10 15 molecules cm −2 + 30 %, including uncertainties in the slant column, the stratospheric column, and the tropospheric air mass factor (AMF) (Boersma et al., 2004).In the present study, comparisons with CMAQ are made for the Version 2 retrievals under cloud-free conditions, i.

Comparisons
Here, comparisons are performed focusing on June and December 2007.As a case study, only these two months are examined in the present study, because the computational cost for CMAQ simulations under various conditions, particularly with fine horizontal resolution, is huge.The year 2007 was chosen considering that (1) the period covered by GOME-2, SCIAMACHY, and OMI at the same time was 2007-2012 and (2) an earlier time would be better, in general, for a lesser chance of satellite instrument degradation and the availability of more mature emission inventories.Concentrating on these two months, model evaluations were performed for 12 selected diagnostic regions, which are defined as in Fig. 1.In the figure, the 12 diagnostic regions are drawn with rectangles on CMAQ-simulated NO 2 fields with different spatial resolutions of 80, 40, 20, and 10 km.Latitude and longitude ranges for each region are given in Table 2.
While various horizontal resolutions are mixed among data sets from not only the CMAQ simulations but also the satellite observations, the 12 diagnostic regions have been selected to be wider than any horizontal resolutions, in order to discuss satellite vs. CMAQ comparisons under conditions with a similar spatial representativeness over the area of interest.Also, to increase the representativeness over time as well, monthlymean tropospheric NO 2 VCDs from satellite observations and CMAQ simulations are compared for each of the 12 diagnostic regions.Satellite-based monthly-mean tropospheric NO 2 VCDs were estimated from swath data with a cloud fraction less than 20 % to ensure moderate quality and a sufficient number of data points.The estimated monthly-mean VCDs can be different from the true monthly-mean values due to the lack of NO 2 VCDs under cloudy conditions.To investigate its impact, monthly-mean NO 2 VCDs were calculated using various cloud fraction thresholds.In Fig. 2, satellitebased tropospheric NO 2 VCDs are plotted against the cloud fraction threshold over the CEC region.The largest difference of NO 2 VCDs with respect to the value at a cloud fraction of 20 % is found to be ∼ 30 %, which is much smaller than the quoted uncertainty in the satellite retrievals, as discussed later.It is interesting to note that the de-Introduction

Conclusions References
Tables Figures

Back Close
Full pendence on the cloud fraction threshold for June is not as significant as for December.Also, a similar diurnal variation pattern, particularly as the difference between morning and afternoon values, can be seen even using different cloud fraction thresholds.The same characteristics were seen for the other diagnostic regions (not shown).Thus, below we assume that diurnal variations (differences between morning and afternoon values) seen from satellite observations are reliable, particularly in June.

Results for June 2007
In Fig. 3, monthly-mean tropospheric NO 2 VCDs from satellite observations are plotted in pink as a function of local time for the CEC, NCP, BEI, and SCN regions in June 2007.The associated error bars represent simple averages of quoted uncertainties in the satellite swath data used for monthly-mean calculations.Their magnitudes are much larger than the range (with respect to the value at a cloud fraction of 20 %, as mentioned above) of NO 2 VCDs at different cloud fraction thresholds (Fig. 2).As seen from Fig. 3, the NO 2 VCD tends to increase from 09:30 (GOME-2) to 10:00 LT (SCIA-MACHY).This tendency is consistent with that seen from MAX-DOAS observations at Beijing in summers of 2008-2011 (Ma et al., 2013) but different from MAX-DOAS observations showing a decrease of NO 2 VCD from 09:30 to 10:00 LT around the center of CEC (Tai'an) in June 2006 (Irie et al., 2008) and around Beijing in summers of 2008-2012 (Hendrick et al., 2013).Detailed validation comparisons using MAX-DOAS observations at several locations in Japan and China were conducted by Irie et al. (2012) and it was concluded that biases for GOME-2 and SCIAMACHY data are likely insignificant, with the most probable estimates of biases of +1 ± 14 % and −5 ± 14 %, respectively.
Assuming their results that GOME-2 data have a positive bias and SCIAMACHY data have a negative bias enlarges the tendency found in Fig. 3 and therefore enlarges the difference from the tendency reported by Irie et al. (2008) and Hendrick et al. (2013).
We should note, however, that uncertainties in both SCIAMACHY and GOME-2 data are much larger than the difference between GOME-2 and SCIAMACHY values, as seen in Fig. 3. Considering that the difference is likely to be insignificant, the present 14045 Introduction

Conclusions References
Tables Figures

Back Close
Full Several lines without error bars in Fig. 3 are tropospheric NO 2 VCDs simulated by CMAQ.Thick, solid black lines are for CMAQ base case simulations with a horizontal resolution of 80 km.Other black lines show results from simulations at horizontal resolutions of 40, 20, and 10 km.The ratio of CMAQ-simulated tropospheric NO 2 VCDs at horizontal resolutions of 10 and 80 km (R) is given for each simulation in Table 3.
It can be seen from Fig. 3 and Table 3 that the CMAQ NO 2 VCD increases, when the horizontal resolution is improved from 80 to 10 km.This can be interpreted as the pronounced effects of the nonlinearity of chemistry (Valin et al., 2011) over given model grids, where NO x emissions have to be artificially distributed.It is interesting to note that for CEC, NCP, and SCN the magnitude of such effects brought by improving horizontal resolutions can be as large as that caused by increasing emissions by 20 %, as shown by black lines with triangles (Fig. 3).A 20 % decrease (increase) in emissions leads to a 21-22 % decrease (a 23-24 % increase) in NO 2 VCDs for CEC.The impact on NO 2 VCDs by the same amount of emission changes is larger for December, as mentioned later.For BEI, the R value is as large as 1.48-2.06(Table 3) and the impact of the change in horizontal resolution is larger than that of an emission change by 20 %.As discussed later, this also occurs in December and for the PRD, JPN, and KOR regions in both June and December.These regions, including BEI, are characterized by strong emissions occurring in a limited space.The R values over these regions are larger than those over the other regions.The R value is usually larger at 13:45 than at 09:30-10:00, probably reflecting effects of the nonlinearity of chemistry, which should be OH-dependent.
Results for the other sensitivity simulations are also shown in Fig. 3.When the number of layers is reduced from 37 (black; base case simulations) to 14 (sky blue), the Introduction

Conclusions References
Tables Figures

Back Close
Full changes in NO 2 VCDs are negligibly small.When the injection height is reduced by half (purple), NO 2 VCDs tend to increase but only small changes take place.The omission of soil NO x emissions (orange crosses) leads to a significant reduction of NO 2 VCD in June.Considering diurnal variation of emissions, the morning and afternoon NO 2 VCD values decrease and increase, respectively.This does not always produce better agreement with diurnal variation patterns seen from satellite measurements.In Fig. 4, comparisons for the YTD, PRD, JPN, and KOR regions are shown.General characteristics of the comparisons for the YTD region are similar to those for CEC, NCP, BEI, and SCN (Fig. 3).For the PRD region, the CMAQ base run values show significant underestimation, where the values are out of the error ranges for both GOME-2 and SCIAMACHY NO 2 VCD data.An increase in emissions by 20 % or an improvement of the spatial resolution (from 80 to 20 km) leads to greater NO 2 VCDs in the model, but either of the two alone is insufficient in explaining the satellite NO 2 VCDs.It can be seen that NO 2 VCDs at a 20 km resolution exceed the values for the 20 % increase in the emissions.This is because the source regions influencing NO 2 VCD over PRD are limited, and we expect greater NO 2 VCDs at improved spatial resolutions over 20 km.
Of all the 12 selected diagnostic regions, comparisons for JPN show the best agreement.This may support that the REAS Version 2 emission inventory around JPN estimated based on the Japan Auto-Oil Program (JATOP; Kurokawa et al., 2013) is reliable.For JPN and KOR, the impact of spatial resolution is large, similar to those for PRD and BEI regions as mentioned above.
Figure 5 shows comparisons over 4 marine regions.All CMAQ simulations conducted for the marine regions show agreement with satellite observations to within their error range.It is worthwhile to mention here that tropospheric NO 2 VCDs can vary with different horizontal resolutions, probably due to effects of the nonlinearity of chemistry (Valin et al., 2011), where ship emissions might be significant enough to drive the effect.Or, a contamination might occur due to an intensified penetration of high-NO 2 air masses from land at finer horizontal resolutions, as discussed in detail later on.Introduction

Conclusions References
Tables Figures

Back Close
Full In summary, for June 2007, diurnal variation patterns (differences between morning and afternoon values) are reproduced well by CMAQ for all 12 diagnostic regions.Quantitative agreement can also be given, when potential influences of CMAQ horizontal resolution and uncertainty in the satellite data are taken into account.

Results for December 2007
For December 2007, comparisons for all 12 diagnostic regions are shown in Figs. 6, 7, and 8, in manners similar to Figs. 3, 4, and 5 for June 2007.For the CEC, NCP, BEI, and SCN regions, it can be seen that diurnal variation patterns differ between satellite and CMAQ NO 2 VCD values (Fig. 6).As given in Table 4, over CEC, for example, the satellite data indicate that the afternoon-to-morning ratios of VCDs are about 0.6 in both June and December, whereas CMAQ indicates ratios of 0.7 and 1.0 in June and December, respectively (Table 5).In general, since the photochemical activity in winter is weaker than in summer, it is reasonable to expect a larger afternoon-to-morning ratio in winter than in summer, with a smaller amplitude of the NO 2 diurnal variation or even a larger NO 2 VCD in the afternoon than in the morning.Similar features were reported by Cede et al. (2006) for NO 2 column measurements in Maryland, USA and by Ma et al. (2013) and Hendrick et al. (2013) for MAX-DOAS measurements around Beijing, China.According to simulations by Uno et al. (2007), the NO 2 chemical lifetime estimated from the chemical reaction term reveals a longer lifetime in December (about 12-72 h) than in June (about 6 h) in the lower troposphere.This expected variation is calculated by CMAQ but not seen from satellite data (Fig. 6 and Tables 4 and 5).CMAQ data better agree with or overestimate satellite data in the afternoon, whereas CMAQ tends to underestimate satellite data (GOME-2 and SCIAMACHY) in the morning.The same tendency is seen for the other years (Itahashi et al., 2013).Also, this is consistent with the results of Han et al. (2011), who have also shown a significant disagreement between diurnal variation patterns from model and satellite data.
To explain the difference seen in Fig. 6 from the viewpoint of the uncertainty in emissions, we may need for emissions to increase in the morning and to decrease in the 14048 Introduction

Conclusions References
Tables Figures

Back Close
Full afternoon.However, simulations assuming the diurnal variation pattern derived by Lin et al. (2010) (green crosses in Fig. 6) indicate that its sensitivity is too small to explain.None of the sensitivity simulations in the present study show a significant improvement in the slope of NO 2 VCDs between morning and afternoon.
From the sensitivity simulations, we quantify impacts by a simple change in the emission strength on NO 2 VCD values.A decrease (increase) in emissions by 20 % leads to a 35-36 % decrease (a 39-41 % increase) in NO 2 VCDs for CEC.These magnitudes are much larger than that found in June (Fig. 3).
For YTD, PRD, JPN, and KOR, both morning and afternoon CMAQ data agree with satellite values to within the error ranges (Fig. 7).Comparisons for marine regions (ECS, SOJ, SCS, and SJP) show a tendency that the CMAQ data are larger than satellite values (Fig. 8).This is evident in SOJ, where the differences cannot be explained by any sensitivity simulations or satellite data uncertainty.In SOJ, there are some negative values in the satellite swath data.These may suggest that the subtraction of the stratospheric NO 2 column from the total NO 2 in the satellite retrieval procedure had a difficulty in winter, when the stratospheric NO 2 is usually more abundant compared to the summer.
It is noted here that even over the ocean, NO 2 VCD increases as the spatial resolution is improved (Fig. 8 and Table 3).Compared to SOJ, this effect is large in ECS, SCS, and SJP, where ship emissions could be significant.It is thought that the increase occurs partly because of the nonlinear chemistry over these regions, where the spatial scale for NO x emission from ships is small enough.Also, at finer spatial resolutions, vertical transport of NO x by convection is more suppressed (Wild et al., 2006), potentially enhancing NO 2 VCDs.
The R values in December are usually similar to or smaller than those in June over all 12 diagnostic regions, except for ECS, SCS, and SJP (Table 3).The three regions are close to the eastern coast of China, where a penetration of high-NO 2 air masses from China is more visible at finer horizontal resolutions in Fig. 1.Thus, it is likely Introduction

Conclusions References
Tables Figures

Back Close
Full that a contamination by such high-NO 2 air masses occurs, enhancing R values at fine resolutions over ECS, SCS, and SJP in December.
As discussed above, we have faced a difficulty in interpreting a significant disagreement between satellite and CMAQ values in December.The present study has assumed so far that the satellite data can be used to derive diurnal variation patterns based on validation comparison results (Irie et al., 2012).In the validation comparisons done by Irie et al. (2012), however, the results were mostly based on comparisons with ground-based MAX-DOAS measurements performed in months other than December.Taking this fact into consideration, efforts for satellite validation seem insufficient and therefore we suggest further precise validation study to identify the cause of differences, especially for the wintertime.

Conclusions
To evaluate the CMAQ reproducibility of satellite-retrieved tropospheric NO 2 VCD data over East Asia, we utilized SCIAMACHY, OMI, and GOME-2 data to realize comparisons from the viewpoint of the diurnal variation of NO 2 .Various sensitivity simulations were conducted using CMAQ for detailed comparisons.The comparisons have been made focusing on 12 diagnostic regions (Fig. 1) and the two months of June and December 2007.In June 2007, diurnal variation patterns (differences between morning and afternoon values) were well reproduced by CMAQ for all 12 diagnostic regions.Quantitative agreement was also found, when potential influences of CMAQ horizontal resolutions and uncertainty in satellite data were taken into account.Over Central Eastern China (CEC), for example, the afternoon-to-morning ratios of VCDs derived from satellite and CMAQ data are about 0.6 and 0.7, respectively.For the Beijing (BEI), Pearl River Delta (PRD), Japan (JPN), and Korean (KOR) regions, where strong emissions occur in a limited space, large impacts of spatial resolution on NO 2 simulations were seen compared to the other regions.Comparisons for JPN show the best agreement, supporting the accuracy of the REAS Version 2 emission inventory, at Introduction

Conclusions References
Tables Figures

Back Close
Full least around JPN.In contrast, a difficulty arises in interpreting comparisons between satellite and CMAQ values over most of the diagnostic regions in December.For CEC, the afternoon-to-morning ratios of VCDs from satellite and CMAQ are about 0.6 and 1.0, respectively.The disagreement cannot be explained by any sensitivity simulations performed in this study.To address this, more investigations, including further efforts for satellite validation, particularly in wintertime, are needed.Introduction

Conclusions References
Tables Figures

Back Close
Full  Full  Full  Full  Full Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | van Noije et al. ( Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |the Reanalysis of the Tropospheric chemical composition over the past 40 yr (RETRO, http://retro.enes.org)and the Model of emissions of gases and aerosols from nature Version 2.0 (MEGAN Version 2.0, http://acd.ucar.edu/~guenther),respectively.
Discussion Paper | Discussion Paper | Discussion Paper | e. cloud fraction (CF) less than 20 %.To estimate the biases in the SCIAMACHY, OMI, and GOME-2 tropospheric NO 2 VCD data in a consistent manner, Irie et al. (2012) used a single data set from ground-based MAX-DOAS observations performed at three sites in Japan and three sites in China in 2006-2011.From their regression analysis between satellite and MAX-DOAS tropospheric NO 2 VCDs, it was concluded that the biases with respect to MAX-DOAS values are less than about 10 % and insignificant for all three data sets.It should be noted that their bias estimates are based mainly on the comparisons made around summer (May, June, and September) over China.This is discussed later to better interpret the differences seen in comparisons between NO 2 VCDs from satellite observations and CMAQ calculations in the present study.Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |study focuses only on larger differences that occur between afternoon and morning values.The afternoon values observed by OMI are smaller than the morning values derived from GOME-2 and SCIAMACHY.This tendency is consistent with all the abovementioned MAX-DOAS observations performed byIrie et al. (2008),Ma et al. (2013), andHendrick et al. (2013).

Table 1 .
Conditions for sensitivity simulations by CMAQ.

Table 3 .
Ratios of CMAQ-simulated tropospheric NO 2 VCD values at horizontal resolutions of 10 and 80 km.
a Ratios of tropospheric NO 2 VCDs at horizontal resolutions of 20 and 80 km.Introduction