Detectability assessment of a satellite sensor for lower tropospheric ozone responses to its precursors emission changes in East Asian summer

Satellite sensors are powerful tools to monitor the spatiotemporal variations of air pollutants in large scales, but it has been challenging to detect surface O3 due to the presence of abundant stratospheric and upper tropospheric O3. East Asia is one of the most polluted regions in the world, but anthropogenic emissions such as NOx and SO2 began to decrease in 2010s. This trend was well observed by satellites, but the spatiotemporal impacts of these emission trends on O3 have not been well understood. Recent advancement in a retrieval method for the Ozone Monitoring Instrument (OMI) sensor enabled detection of lower tropospheric O3 and its legitimacy has been validated. In this study, we investigated the statistical significance for the OMI sensor to detect the lower tropospheric O3 responses to the future emission reduction of the O3 precursor gases over East Asia in summer, by utilizing a regional chemistry model. The emission reduction of 10, 25, 50, and 90% resulted in 4.4, 11, 23, and 53% decrease of the areal and monthly mean daytime simulated satellite-detectable O3 (ΔO3), respectively. The fractions of significant areas are 55, 84, 93, and 96% at a one-sided 95% confidence interval. Because of the recent advancement of satellite sensor technologies (e.g., TROPOMI), study on tropospheric photochemistry will be rapidly advanced in the near future. The current study proved the usefulness of such satellite analyses on the lower tropospheric O3 and its perturbations due to the precursor gas emission controls.

In order to monitor the spatiotemporal variations of air pollutants in large scales and to detect their long-term trends, satellite sensors are powerful tools. However, in terms of the surface and lower tropospheric O 3 , it has been challenging to measure from the space, because of the presence of abundant stratospheric O 3 . Recently, a retrieval algorithm has been developed by Liu et al. 12 to detect the lower-tropospheric (i.e., approximately 0-3 km) O 3 and its legitimacy has been validated especially focusing on the East Asian region [13][14][15][16] . In fact, based on the climatological zonal and monthly mean O 3 profiles derived from 15 years of ozonesonde measurements 17 , the 0-3 km column amounts in the northern mid-latitudes range from 7.1 to 13.7 DU, which are only 2.1-4.0% of the total column amounts, approximately 300-380 DU. In this study, we investigated the statistical significance for the OMI sensor to detect the tropospheric O 3 responses to the current/future emission reductions of the O 3 precursor gases (NO x and NMVOCs) over East Asia by using the retrieval algorithm and numerical simulations. We selected June 2006 in the study, because both observed and simulated lower tropospheric O 3 concentrations were largest. Also, the summer was most favorable for the current analysis, evaluating sensitivity of O 3 to local precursor emission changes. In colder seasons, the contribution of long-range transport is larger due to monsoon and lower photochemical production. More than half of surface O 3 in East Asia were attributed to those transported from distant sources, whereas most of them were the domestic origin in the summer 11 . In addition, the retrieval sensitivity to lower tropospheric O 3 in this region maximizes in the summer due to smaller solar zenith angle 12 .

Results and Discussion
Spatial distribution of lower tropospheric O3. Figure 1 shows the monthly mean simulated and observed O 3 over East Asia in June 2006. The simulation was conducted by using a regional meteorology -chemistry model (NHM-Chem) 18 . The panels are to show how the simulation results were compared with the retrieved OMI O 3 12,19-21 (denoted as OMI-O 3 hereinafter) and how the horizontal distributions were changed accordingly by the averaging procedures. Figure 1a shows the simulated surface O 3 concentration (ppb; Δx = 30 km), the most important factor for the terrestrial ecosystems, and thus to be evaluated by the observation. In the summer, the atmospheric conditions are most favorable for the photochemical production of O 3 over the land. Because the surface wind was relatively weaker compared to the cold seasons, the surface concentration showed maxima over the high emission areas, such as North China Plain, Yangtze River Delta, and Sichuan Basin in China and densely populated regions in Korea and Japan. In fact, the cluster analysis, provided by Hayashida et al. 15 , proved that the locations of a high OMI-O 3 cluster matched with the high NO x emission areas. In the month, the Pacific High pressure system was dominant and the clean maritime air was transported from the south to the Northwestern Pacific. During the summer, the Pacific High blocks the long-range transport due to the mid-latitude westerlies is significantly larger than the a priori (climatological) O 3 at a one-sided 95% confidence interval. The spatial resolution Δx of (a,b) is 30 km, whereas Δx of (c,d) is 1 degree.
in the region. Figure 1b shows the simulated 0-3 km mean O 3 concentration (ppb; Δx = 30 km). Due to the prevailing mid-latitude westerlies at upper altitudes, the high concentration areas extend farther eastward compared to the surface concentrations. Figure 1c shows the simulated OMI-time 0-3 km mean O 3 column amount (DU; Δx = 1°), convolved with the retrieval averaging kernels (AKs) (denoted as simulated O 3 with AK, hereinafter) (see Eq. (1)), which can be quantitatively compared with the OMI-O 3 as shown in Fig. 1d. The "simulated OMItime" indicates simulation results at 6 UTC (13:40 local time at the center of model domain (115 °E)), while the OMI observation time is 13:45 local time. The spatial distributions of Fig. 1b,c are drastically different due to the following reasons: (1) influences of O 3 from the above layers and (2) exclusion of cloudy/rainy days and days under the influence of stratospheric O 3 intrusion due to the tropopause perturbations. In order to clearly show the effects, two more panels (OMI-time O 3 before and after screening for clouds and stratospheric inclusion) are added to Fig. 1 as illustrated in Fig. S1 in the supplement. The "simulated OMI-time 0-3 km O 3 " means that after the screening in this paper.
The Student's t-test was applied for the monthly mean OMI-O 3 and the a priori O 3 using the daily values in June 2006. Degrees of freedom at each grid point are the number of available data in the month minus one, i.e., up to 29. The hatched area of Fig. 1d indicates the grids where the OMI-O 3 was significantly larger than the a priori O 3 at a one-sided 95% confidence interval. Most of the O 3 rich regions (i.e., >12 DU) are statistically significant. For the emission reduction test as presented later, data in the hatched area inside the dashed box (110-146°E, 30-45°N) were used for the analysis.
The same horizontal distribution with a different confidence interval, two-sided 99%, is presented in Fig. S2, in the supplement. Over the relatively low concentration areas such as Sichuan Basin, Korea, and Japan, the hatched areas become small or disappeared. In contrast, over the high concentration areas such as North China Plain and Yangtze River Delta, the difference between the OMI-O 3 and the a priori O 3 remained significant at this confidence interval, due to substantially large near-surface photochemical O 3 production. comparison between observation and simulation. In order to compare the spatial distributions of the simulation (Fig. 1c) and the observation (Fig. 1d), as shown in Fig. 2, the cross sections of the simulated and observed O 3 at 117.5°E and 35.5°N are compared, in order to cover the areas, where O 3 enhancement was observed by the satellite retrieval, namely, the North China Plain and populated and industrial regions in Korea and Japan.
The upper panels of Fig. 2 show the monthly mean values of simulated hourly surface O 3 (blue, solid), O 3 at the OMI observed time (blue, dashed), and 0-3 km mean O 3 at the OMI time (red, solid). The surface mean OMI-time O 3 was significantly larger by more than 30 ppb than the mean of hourly O 3 over the area with the abundant presence of emission of precursor gases (30-40°N, 110-120°E). This daytime enhancement was smaller for 0-3 km mean O 3 but still significant: the 0-3 km mean OMI-time O 3 was larger by 10-20 ppb over the high emission area. On the other hand, over the low emission area or over the ocean, the 0-3 km mean OMI-time O 3 was even larger than the surface OMI-time O 3 due to the absence of photochemical production near the surface. In such areas, the daytime O 3 enhancement was not detected by the satellite and thus no statistical significance was found between the a priori and retrieved O 3 . The lower panels of Fig. 2 show the monthly mean values of simulated 0-3 km column amount of hourly O 3 (blue, solid), O 3 at the OMI time (blue, dashed), and O 3 at the OMI time with AK (red, solid). The monthly mean OMI-time O 3 was 1-4 DU larger than the monthly mean of hourly O 3 over the large emission areas, but the absolute values of monthly simulated OMI-time O 3 with AK were even lower than the monthly mean of hourly O 3 , due to the reasons presented later with Eq. (1). Still, the important things here are that the simulated OMI-time O 3 with AK agreed well with the OMI-O 3 (black, solid) (generally smaller than 0.5 DU over the high emission areas) and that the both simulated and observed O 3 were significantly larger than the a priori O 3 (black, dashed) over the high emission areas. The spatial correlation coefficient between the simulated and observed monthly values over the hatched area in the dashed box shown in Fig. 1d was 0.94. Figure 3 shows the monthly mean simulated (top to bottom) surface O 3 , 0-3 km mean O 3 , OMI-time 0-3 km O 3 after the screening, and that with AK at the reduction rates of anthropogenic precursor gases emissions of (left to right) 10%, 25%, 50%, and 90%. The contrasts between the results with 10% and 90% reductions become gradually smaller from surface O 3 , 0-3 km O 3 , to OMI-time 0-3 km O 3 with AK. For the simulated 0-3 km O 3 with AK, even though the differences looked small, the differences between the emission reduction simulations and the control run (i.e. 0% emission reduction simulation; Fig. 1c) were statistically significant at a one-sided 95% confidence interval (the Student's t-test), as indicated by the hatched areas in Fig. 3m-p. The significance areas became larger as the emission reduction rates were larger. (The same figures without the hatched areas are presented in Fig. S3 to make the results more visible.)

Sensitivity to emission changes and statistical significance.
The same horizontal distribution with a different confidence interval, two-sided 99%, is presented in Fig. S4, in the supplement. Similar to the differences between the two confidence intervals as shown in Fig. S2, the hatched areas become smaller with the two-sided 99%, but still cover the high concentration areas such as North China Plain and Yangtze River Delta. The most of the high concentration areas are covered with the hatched areas for the emission reduction rates greater than 25% in Fig. S4f -S4h. Figure 4 shows areal statistics of the horizontal distributions over the hatched regions in the dashed box of Fig. 1d: the monthly mean OMI-O 3 was significantly larger than the a priori O 3 over 110-146°E and 30-45°N.  Table S1. Here we define ΔO 3 . The difference between the OMI-O 3 and the a priori O 3 is denoted as OMI-ΔO 3 , which can be regarded as a satellite-detectable O 3 . Also derived and discussed is the simulated ΔO 3 (or simply referred to as ΔO 3 ), which is the simulated O 3 with AK minus the a priori O 3 used for the retrieval. ΔO 3 can be regarded as a simulated satellite-detectable O 3. Figure 4a,b show the areal and monthly mean simulated O 3 and their ratios to those at 0% emission reduction, respectively. The black and blue symbols indicate the means of hourly and OMI-time surface O 3 , respectively. The daytime enhancement (black minus blue) was 15 ppb at 0% emission reduction due to the local photochemical productions, which was almost 0 ppb at 90% due to the absence of anthropogenic emissions of precursor gases. O 3 level at 90% reduction rates can be almost regarded as the background level (or the contributions from the hemispheric transport 11 ), approximately 35 ppb for surface O 3 and 10 DU for 0-3 km column of O 3 . As shown in Fig. 4b, the decreasing rate of the OMI-time surface O 3 (approximately 50% at 90% reduction) is larger than that of hourly surface O 3 (approximately 70% at 90% reduction). The 0-3 km OMI-time column amount (red) decreased by 2.3, 6.1, 14, and 33% (Fig. 4b, Table S1) due to the emission reductions of 10, 25, 50, and 90%, respectively. This decreasing trend of the 0-3 km OMI-time column amount (red) was smaller than that of surface OMI-time concentration (black) and as small as that of the surface hourly concentration (blue). The decreasing trend of OMI-time O 3 with AK (orange in Fig. 4a) is very small, but that of simulated ΔO 3 (orange in Fig. 4b), i.e., OMI with AK (orange cross) minus the a priori OMI (orange dashed; 11.83 DU), is as large as surface OMI-time O 3 (Fig. 4b): The ΔO 3 was reduced to 47% at 90% reduction (Table S1). Therefore, even though the decreasing trend of O 3 with AK was small, the simulated O 3 with AK of various emission reduction simulations were significantly smaller than that at 0% reduction simulation, in substantial fractions of the domain (Fig. 4c). The fractions of significant areas at the emission reduction rates of 10%, 25%, 50%, and 90% to the hatched area in Fig. 1d are 55, 84, 93, and 96% at a one-sided 95% confidence interval (red) and 23%, 51%, 63%, and 68% at a two-sided 99% confidence interval (blue), respectively (Fig. 4c). www.nature.com/scientificreports www.nature.com/scientificreports/ By using a recently developed satellite product 12,13 and a regional meteorology -chemistry model 18 , we concluded that the Ozone Monitoring Instrument (OMI) sensor can detect summer-time lower tropospheric O 3 responses due to reductions of emissions of anthropogenic precursor gases, NO x and non-methane volatile organic compounds (NMVOCs), in East Asia, despite the abundant presence of stratospheric and upper tropospheric O 3 . A socio-economic future emission scenario study 22 estimated approximately 50% reduction in the global NO x emission in 2100 with compared to the year 2000, and even higher reduction rates up to 80% for additional climate mitigation cases. For such reduction cases greater than 50%, the satellite sensor could detect lower tropospheric O 3 changes over substantially large areas (Fig. 4c). The current study showed usefulness and importance of monitoring future O 3 trend by satellite sensors. The same reduction rates between NOx and NMVOC emissions are unlikely in reality, as the emission sources of the two components are very different. The realistic emission scenario needs to be applied in the future beyond the current study. Recently, TROPOspheric Monitoring Instrument (TROPOMI 23 ) has been onboard to the Copernicus Sentinel-5 Precursor (S5p) satellite. Because of the recent advancement of satellite sensor technologies, study on tropospheric photochemistry will be rapidly advanced in the near future, together with other technologies such as in-situ/remote measurements and numerical simulations.

Methods
OMI-retrieved lower tropospheric O 3 product. The retrieval methodology and the usage of lower tropospheric O 3 product are described in detail in Hayashida et al. 14,15 . We used the data obtained from the OMI sensor, a Dutch-Finnish-built nadir-viewing UV/visible instrument, carried by the Aura spacecraft of the National Aeronautics and Space Administration (NASA) Earth Observing System (EOS) in a sun-synchronous orbit with an equatorial crossing time of ~13:45 local time. The O 3 profiles are retrieved by Liu et al. 12 with several modifications described in Kim et al. 19 , from the ground upward to approximately 60 km in 24 layers, of which 3-7 layers are in the troposphere. To constrain the retrievals, they used climatological zonal mean O 3 profiles and standard deviations derived from 15 years of ozonesonde measurements and the Stratospheric Aerosol and Gas Experiment (SAGE) as a priori data 17 , which vary with altitude, month, and latitude. The retrieval was performed at a nadir resolution of 52 km × 48 km by adding 4/8 UV1 (270-310 nm) /UV2 (310-365 nm) pixels. In the current study, we use the Level 3 product gridded to 1° × 1° (latitude × longitude) spatial resolution on a daily basis. The gridded data were screened using the effective cloud fraction (ECF) < 0.2 and RMS (root mean square of the ratio of the fitting residual to the assumed measurement error of the UV2 channel) < 2.4 criteria 13 . The retrieved O 3 in the lower tropospheric layers was found to be affected by outstanding O 3 enhancement along with the sub-tropical jet due to the intrusion of the stratospheric O 3 13 . This artifact was successfully screened out by the method developed by Hayashida et al. 13,14 . In order to validate their product, Hayashida et al. 13 Table 1.
A regional-scale meteorology -chemistry model. A regional-scale meteorology -chemistry model, NHM-Chem 18 , was used to simulate tropospheric O 3 over East Asia. Among the three aerosol representation options employed in NHM-Chem, the bulk equilibrium method was selected in this study. The method is computationally efficient but found to be accurate enough for the prediction of mass concentrations 18 .
The simulation settings such as model domain, simulation period, and boundary conditions are the same as Kajino et al. 18 18 only provided model evaluation for surface concentrations, the comparison of the simulated and observed 0-3 km mean concentrations were conducted as shown in Table 1. Because the IAGOS measurement at Beijing ended November of 2005 and no data are available for 2006, we compared the simulation results against the other East Asian airports such as Hong Kong, Shanghai, Osaka, and Tokyo for 2006. In order to increase the number of data, we also drove NHM-Chem for 2005 and compared the results against the observation. We also used the ozonesonde data, conducted by the Global Atmospheric Watch (GAW) program of World Meteorological Organization (WMO) for the model evaluation. The ozonesonde data is available at https:// woudc.org/data/explore.php (last access: 2 November 2019). As shown in Table 1, the simulation results showed good agreements with the both in-situ measurements.
In addition to Kajino et al. 18 , we conducted sensitivity simulations of anthropogenic emission of O 3 precursors such as NO x and non-methane volatile organic compounds (NMVOCs). We have totally five sets of simulations, with 0%, 10%, 25%, 50%, and 90% reductions from the emission flux of the year 2006. The same reduction rates were applied for both NO x and NMVOCs. Note that the same reduction rates were unlikely in reality as the emission sources of the both species are quite different.
In order to compare against the OMI data, the simulation results were spatially allocated to the grids of the OMI products, which were horizontally 1° × 1° (latitude × longitude) with vertically approximately 3 km intervals. Then, we applied the OMI retrieval averaging kernels (AKs) (rows) to the simulation results for the consistent comparison as follows: where ′ X j is the simulated O 3 column amount (DU) at j-th OMI vertical grid, convolved with the retrieval AKs (A(i, j)), denoted as O 3 "with AK". X t,i is the simulated O 3 column amount at i-th OMI vertical grid and X a is the a priori profile used in the OMI retrievals. We applied Eq. (1) to the simulation results only for i = 22, 23, and 24, (i.e., lower than approximately 9 km above ground level, within the troposphere) and we used the OMI-retrieved data for X t,i above the 21st layer because NHM-Chem is a tropospheric chemistry model.

code availability
The NHM-Chem source code is available subject to a license agreement with the Japan Meteorological Agency. Further information is available at http://www.mri-jma.go.jp/Dep/ap/nhmchem_model/application.html. Unfortunately, this website is only in Japanese. Thus additionally, the source code, user's manual, analysis tools, and sets of boundary conditions can be provided upon request to the corresponding author (M.K.).