Atmospheric Trace Gas (NO 2 and O 3 ) Variability in South Korean Coastal Waters, and Implications for Remote Sensing of Coastal Ocean Color Dynamics

: Coastal environments are highly dynamic, and are characterized by short-term, local-scale variability in atmospheric and oceanic processes. Yet, high-frequency measurements of atmospheric composition, and particularly nitrogen dioxide (NO 2 ) and ozone (O 3 ) dynamics, are scarce over the ocean, introducing uncertainties in satellite retrievals of coastal ocean biogeochemistry and ecology. Combining measurements from different platforms, the Korea-US Ocean Color and Air Quality ﬁeld campaign provided a unique opportunity to capture, for the ﬁrst time, the strong spatial dynamics and diurnal variability in total column (TC) NO 2 and O 3 over the coastal waters of South Korea. Measurements were conducted using a shipboard Pandora Spectrometer Instrument speciﬁcally designed to collect accurate, high-frequency observations from a research vessel, and were combined with ground-based observations at coastal land sites, synoptic satellite imagery, and air-mass trajectory simulations to assess source contributions to atmospheric pollution over the coastal ocean. TCO 3 showed only small (<20%) variability that was driven primarily by larger-scale meteorological processes captured successfully in the relatively coarse satellite imagery from Aura-OMI. In contrast, TCNO 2 over the ocean varied by more than an order of magnitude (0.07–0.92 DU), mostly affected by urban emissions and highly dynamic air mass transport pathways. Diurnal patterns varied widely across the ocean domain, with TCNO 2 in the coastal area of Geoje and offshore Seoul varying by more than 0.6 DU and 0.4 DU, respectively, over a period of less than 3 h. On a polar orbit, Aura-OMI is not capable of detecting these short-term changes in TCNO 2 . If unaccounted for in atmospheric correction retrievals of ocean color, the observed variability in TCNO 2 would be misinterpreted as a change in ocean remote sensing reﬂectance, R rs , by more than 80% and 40% at 412 and 443 nm, respectively, introducing a signiﬁcant false variability in retrievals of coastal ocean ecological processes from space.


Introduction
At the interface between the land, oceans, and atmosphere, coastal regions are highly dynamic environments, characterized by strong variability in both water and air quality. Variability in atmospheric composition is associated with short-term (e.g., hourly) changes and periodic (e.g., weekly) patterns in anthropogenic emissions, as well as complex meteorological processes that influence the circulation and accumulation of atmospheric pollutants at land-water boundaries [1][2][3]. Previous studies, using high spatial resolution meteorological and air quality simulations covering the Chesapeake Bay and adjacent Atlantic Ocean coastline, have demonstrated that air pollution from coastal urban areas often accumulates over adjacent coastal waters due to low dry deposition rates over the water and a shallow marine boundary layer that traps marine emissions [3,4]. Local-scale meteorological processes, such as sea-and bay-breeze circulations, further affect the advection, recirculation, and accumulation of atmospheric pollutants in coastal areas, often leading to a build up of emissions and aggravation of air pollution along the shoreline caused by stagnation that develops as wind directions change [4][5][6][7][8]. Strong, prolonged sea or bay breeze events can transport a large amount of urban air pollution out of the planetary boundary layer and into the free troposphere, where pollutants have longer lifetimes and are susceptible to long range transport offshore and over adjacent ocean environments, with potentially significant ecological impacts for aquatic ecosystems [9].
Assessing the spatial and temporal variability of atmospheric pollutants, aerosols, and absorbing trace gases in coastal areas is critical for improving modeling and prediction of coastal tropospheric air quality, determining impacts on human health, and assessing the ecological implications of atmospheric pollutant deposition in coastal ecosystems [4]. It is also essential not only for improved satellite trace gas retrievals, but also for accurate atmospheric correction of satellite coastal ocean color observations, especially in coastal waters that are close to heavily polluted urban areas [2,10,11]. If not adequately corrected, this atmospheric variability can lead to a false estimate of time-dependent underwater processes from a satellite ocean color sensors. The errors can be aliased as day-to-day or weekly variability in retrievals from polar orbiting sensors, such as MODIS, VIIRS, PACE, and OLCI, impacting their ability to monitor the evolution of marine biological processes such as algal blooms. They can also introduce a false diurnal variability in ocean retrievals from geostationary sensors such as GOCI [12,13].
Ozone (O 3 ) and nitrogen dioxide (NO 2 ) in the atmosphere have absorption bands in ultraviolet, visible, and near-infrared wavelengths relevant to ocean color remote sensing [14]. NO 2 absorption peaks at 400-445 nm, affecting the top of atmosphere (TOA) signal measured by an ocean color satellite sensor at wavelengths traditionally used for retrievals of phytoplankton pigments and dissolved organic carbon dynamics [15][16][17][18]. Radiative transfer calculations by Tzortziou et al. [19] showed that 0.7 DU unaccounted variability in NO 2 , which is consistent with the NO 2 variability previously measured over coastal areas often covered by just a single early afternoon OMI pixel [7], results in a level of uncertainty in coastal water remote sensing reflectance, R rs , which may be as large as 40% at 412 nm, for low solar zenith angles (SZA < 30 • ), and NO 2 distributed within the first 2 km from the ground. This error increases with decreasing wavelength, increasing solar and satellite viewing angles, and as the NO 2 is distributed at higher altitudes, reaching up to 80% or more for SZAs expected with a geostationary sensor [10,19]. The resulting errors in R rs absolute magnitude and spectral shape subsequently introduce significant and time-dependent uncertainties in retrievals of ocean biogeochemical variables from space [10,13,19,20].
Satellite observations provide a powerful tool for assessing O 3 and NO 2 emissions, distribution, and transport over land and the ocean. To enhance the scientific return of satellite remote sensing, obtain a more complete picture of the atmosphere, and increase our understanding of the effects of air pollution on coastal ecosystem dynamics, satellite observations must be used in combination with model simulations and detailed measurements from surface platforms. This becomes extremely challenging in coastal regions, since ground-based networks that monitor tropospheric air quality abruptly end at the coastlines. Previous studies integrating high frequency observations to assess the impact of atmospheric trace gas variability on ocean color retrievals were based on measurements at coastal sites over the land [10,19]. This paper specifically addresses this challenge by quantifying, for the first time, atmospheric variability of NO 2 and O 3 in the coastal waters of South Korea, measured from a shipboard Pandora Spectrometer Instrument (PSI) specifically designed to collect accurate, high-frequency, direct-sun observations from a moving platform. Measurements were collected from the research vessel (RV) Onnuri during the 2016 Korean and US Ocean Color and Air Quality (KORUS OC/AQ) field campaign.
South Korea is embedded in a region of increasing economic change with strong gradients in air quality in time and space [21]. Rapid industrialization, increasing coastal urbanization, and weak regulation of local anthropogenic emissions have caused serious air pollution problems in the region, which are often severely aggravated by long range transport of pollutants. Dramatic increases in NO x emissions have been observed over eastern Asia since 2000 [22][23][24], posing major threats to both human health and downwind terrestrial and ocean ecosystems. Our high frequency measurements from RV Onnuri during KORUS OC/AQ captured, for the first time, the strong spatial dynamics and diurnal variability in total column atmospheric NO 2 and O 3 over the coastal waters of South Korea. In this study, shipboard observations over the ocean are integrated with ground-based observations at coastal land sites to examine differences across the land-ocean interface, and are combined with air-parcel back-trajectory simulations to assess source contributions to atmospheric pollution over these coastal waters. Comparisons of surface measurements with satellite Aura-OMI retrievals of O 3 and NO 2 reveal the capabilities and limitations of coarse resolution, polar-orbiting satellite observations in capturing variability in atmospheric pollution gradients over terrestrial-aquatic interfaces, and the need for concurrent measurements of NO 2 and O 3 to obtain accurate corrections of coastal ocean color satellite imagery in highly polluted coastal urban regions.

Study Region
The Korea-United States Ocean Color (KORUS-OC) expedition was an 18-day oceanographic field campaign conducted from 20 May to 6 June 2016 in the East Sea and Yellow Sea, under the leadership of the Korean Institute of Ocean Science and Technology (KIOST) and the United States National Aeronautics and Space Administration (NASA) [25]. KORUS OC was coordinated with the Korea-United States Air Quality (KORUS-AQ) study, led by Korea's National Institute of Environmental Research (NIER) and NASA [21]. The overarching objectives of these coordinated campaigns were to improve our understanding of the factors contributing to poor air quality in Korea, characterize highly dynamic biological and biogeochemical processes in coastal waters, and capture diurnal variability in biology, biogeochemistry, atmospheric properties and physics to enable improved capabilities for satellite remote sensing and model simulation of atmospheric composition, ocean biology, and biogeochemistry. One of the main goals of these studies was to perform ocean and atmospheric measurements for risk reduction of future satellite observations from geostationary platforms, in an area that is within the field-of-view of the only available geostationary Ocean Color instrument: the Korean GOCI (Geostationary Ocean Color Imager) sensor.
The coastal waters of the Yellow Sea and East Sea/Sea of Japan are strongly impacted by both local atmospheric pollution and the long-range transport of air pollutants from other highly polluted urban centers in East Asia, including Shanghai and Beijing [26,27]. Yet, the impacts of atmospheric pollutants on sea-air exchanges and coastal ocean color retrievals remain largely unknown. Using observations from different platforms, including aircraft, ground sites, and research vessels, the KORUS-OC/AQ study provided unprecedented comprehensive measurements of pollutants (both trace gases and aerosol particle properties) with extensive spatial and vertical coverage. Here, we focus on the spatial and temporal variability of total column NO 2 and O 3 measured by the shipboard NASA PSI-24, operating in direct-sun mode. The spectrometer was mounted on the 1422-ton RV Onnuri, one of the Korean Ocean Research and Development Institute (KORDI)'s research vessels.
During the KORUS OC campaign, the RV Onnuri covered an area along the Eastern, Southern, and Western South Korean coasts ( Figure 1). Measurements in South Korean territorial waters are excluded by international agreement from this study. The RV Onnuri departed from the KIOST South Sea Research Institute dock at Geoje, near the Busan metropolitan area, on 20 May, sailing North-East into the Sea of Japan, and looped around Ullung-do ("do" means island). Its maximum northerly extent was 37.87 • latitude, and its maximum easterly extent was 131.  (Figure 1). These shipboard measurements were compared to measurements from ground-based PSIs to assess differences in total column amounts of NO 2 and O 3 over the land and over the coastal ocean (Table 1; Figure 1). Ground sites included both heavily-polluted urban areas (e.g., Seoul and Busan) affected by anthropogenic emissions, as well as relatively less polluted, rural areas (e.g., Gwangju and Anmyeon) [28].

Measurements of Total Column Trace Gases
Pandora Spectrometer Instrument (PSI) retrievals: A direct-sun PSI [1,11], modified for shipboard deployment, was used to measure total vertical column density of NO 2 (TCNO 2 ) and ozone (TCO 3 ). The Pandora system consists of a small Avantes low stray light spectrometer (spectral range: 280-525 nm; spectral resolution: 0.6 nm with 4 times oversampling) connected to an optical head by a 400 micron single strand fiber optic cable. The spectrometer is temperature stabilized at 20 • C inside two weather resistant containers [28]. The head sensor is mounted on two computer-controlled motors to control the azimuth and zenith viewing angles to ±0.01 • . For the shipboard PSI, the Blick Software suite operating the PSI uses feedback from an internal digital camera to a sun-tracker to adjust the motors and maintain a centered direct-sun view. During KORUS-OC, Pandora was mounted on the bow/forecastle deck of the RV Onnuri, away from the ship engine's exhaust, to avoid any obstructions and contamination. Results from an instrument inter-comparison, using the 11 different PSIs that were deployed during the DISCOVER-AQ (Deriving Information on Surface Conditions from COlumn and VERtically Resolved Observations Relevant to Air Quality) field campaign in the Chesapeake Bay, showed excellent agreement among the instruments: differences were within ±4.8 DU (or ±1.5%) for TCO 3 and ±0.07 DU for TCNO 2 , with no air-mass-factor dependence [2]. A similar result was shown for NO 2 from two PSIs deployed during the KORUS-AQ campaign [28].
Pandora TCO 3 and TCNO 2 amounts are determined using the Differential Optical Absorption Spectroscopy (DOAS) method. The spectral fitting algorithm uses laboratory-measured absorption cross sections for each atmospheric absorber, a 4th order polynomial in wavelength to remove aerosols and Rayleigh scattering effects, and wavelength shift and squeeze functions to remove wavelength errors at the 1 picometer level. The shift and squeeze functions provide the best match to the measured spectrum compared to a solar reference spectrum containing the solar Fraunhofer line structure [1,11]. PSI data were filtered following Tzortziou et al. [2] for a normalized root-mean-square of the weighted spectral fittings of <0.05, SZAs < 70 • , and for uncertainties less than 0.05 DU for TCNO 2 and less than 2 DU for TCO 3 .
OMI retrievals: The Ozone Monitoring Instrument (OMI), launched on NASA's Earth Observing System Aura satellite on July 15, 2004, has been collecting data since 9 August 2004 [29]. Aura has a sun-synchronous polar orbit at approximately 705 km altitude with a period of 100 min and a local equator crossing time between 13:40 and 13:50 local time. OMI is a UV/Visible grating imaging spectrometer that provides near-global coverage in one day. The size of an OMI pixel varies with cross-track viewing zenith angle from 24 km in the nadir to almost 128 km for the extreme viewing angles of 57 • at the edges of the swath [30]. OMI measurements cover a spectral region of 264-504 nm with a spectral resolution between 0.42 nm and 0.63 nm, and provide retrievals for a number of trace gases including O 3 , NO 2 , SO 2 , HCHO, BrO, and OClO.
PSI results are compared here with most recent versions of OMI TCNO 2 , and TCO 3 station overpass data (OMNO2 product, Collection 3.0, Version 3.1; and OMTO3, Collection 3.0, Version 8.5 overpass products; http://avdc.gsfc.nasa.gov/index.php?site=1593048672&id=28) for the four coastal ground-based PSI sites Busan, Gwangju, Anmyeon and Seoul (Table 1). OMI station overpass data files provide the nearest OMI measurement in an OMI track (orbit), if it is closer than 100 km, to the ground-station. For comparison with the shipboard PSI measurements on RV Onnuri, match ups were created based on collocation such that the location of the ship was within the OMI pixel [31]. The maximum collocation distance between the center of the OMI FOV and the PSI location was 26 km for the NO 2 retrievals and 42 km for the O 3 retrievals. OMI data were not available during our shipboard measurements from 30 May to 2 June 2016 as the instrument went into survival mode on 29 May and resumed taking measurements on 10 June 2016. We used OMI measurements during clear-sky conditions (cloud fraction < 0.3) and excluded the data affected by Row Anomaly [32].

HYSPLIT Trajectory Generation
The NOAA Air Resources Laboratory's (ARL's) HYbrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT4) modeling system [33,34] was used to compute air parcel trajectories and determine the origin of air masses sampled by the shipboard PSI-24 on RV Onnuri during KORUS OC. HYSPLIT has been used extensively in the literature in a variety of applications to simulate the atmospheric transport, dispersion, and deposition of pollutants and hazardous materials [35][36][37][38]. HYSPLIT simulations were run using meteorological data from the Global Data Assimilation System (GDAS1.0), because the complex terrain of the Korean Peninsula necessitated favoring a data-set which allowed for the most representative modeling of vertical motion [39]. A previous comparison of HYSPLIT backward trajectories generated from both GDAS1.0 and GDAS0P5 datasets in the nearby Pearl River Delta reported that trajectories produced using GDAS1.0 showed better agreement with observations due to the absence of vertical velocity in GDAS0P5 dataset [40]. Thus, GDAS1.0 datasets were used in modeling trajectories in our study, while GDAS0.5 datasets were used to confirm the general direction of the trajectory. Results are shown here for the RV Onnuri measurements at Geoje and offshore the coastal sites of Gwangju, Anmyeon and Seoul. Backward trajectories were initialized at 1500 m AGL (above ground level) to examine air mass transport and origin at an altitude over the research vessel where NO 2 changes have a considerable impact on atmospheric correction of ocean color retrievals [1,19]. To check for errors in modeling, forward and back trajectories were created for each observation [39]. The termination point for each backward trajectory was used to produce a forward trajectory. All points along the two trajectories were then compared, with differences in excess of 10 km resulting in trajectory disposal and flagging of the observation for error-checking. Due to the need for automation, modification of the control file, downloading of met files, and generation of trajectories was handled using a modified and updated version of the R::SplitR package.

NO 2 and O 3 Dynamics over South Korean Coastal Land Sites
High frequency measurements from ground-based PSIs at the coastal land sites of Busan, Gwangju, Anmyeon, and Seoul ( Figure 1, Table 1) provided detailed information on the spatial and temporal variability in atmospheric total column NO 2 and O 3 across areas characterized by different levels of anthropogenic influence in the South-east, Central-west, and North-west shorelines of South Korea.
The range and temporal dynamics in TCO 3 measured by the ground-based PSI network during May 2016, when the KORUS-OC campaign was held, were very similar across these four coastal sites ( Figure 2, Table 2). Monthly average TCO 3 values were almost identical across sites, 314 DU (stdev = 19 DU), 311 DU (stdev = 19 DU), 315 DU (stdev = 17 DU), and 320 DU (stdev = 15 DU) at Busan, Gwangju, Anmyeon, and Seoul, respectively. Overall, TCO 3 varied by approximately 25% at each station, with most of this change captured by a quasi bi-weekly oscillation (13-15 days) that was clearly evident at all sites, and particularly, at Busan and Gwangju ( Figure 2). Maxima in TCO 3 were measured on 4 and 17 May 2016, while minima were measured on 8-9 and 23-24 May 2016 ( Figure 2). The quasi-biweekly oscillation is one of the major systems that affect tropical and subtropical weather and seasonal mean climate [41,42]. This quasi-biweekly mode in TCO 3 variation was also captured in satellite imagery from Aura-OMI (Figures 2 and 3).  Satellite Aura-OMI retrievals of TCO 3 were in excellent agreement with the ground-based observations (Table 2), also showing 26% variability in TCO 3 (281-367 DU) over the study region during May 2016. The absolute percent difference (APD) in average TCO 3 retrieved by OMI and PSI was 0.38%, 1.17%., 2.15%, and 1.15% at Busan, Gwangju, Anmyeon, and Seoul, respectively, with OMI estimating slightly higher ozone values. OMI synoptic observations over South Korea highlighted the influence of large scale, eastward moving air masses on TCO 3 distributions (Figure 3). These results suggest that column O 3 dynamics over the South Korean peninsula were driven primarily by larger scale meteorological processes that can successfully be captured in the coarser satellite OMI imagery, as opposed to small scale heterogeneities and local emissions.
Contrary to TCO 3 , total column NO 2 showed clear differences among coastal land sites and a strong influence of smaller scale processes and local emission sources. During May 2016, considerably higher NO 2 pollution levels were measured at the coastal megacities of Seoul and Busan, with average TCNO 2 values of 1.01 DU (stdev = 0.64 DU) and 0.71 DU (stdev = 0.39 DU), respectively. TCNO 2 reached as high as 5.78 DU at Seoul and 3.05 DU at Busan (Table 2). Changes as large as 2.0 or 3 DU within 3 to 4 h were often recorded at these two stations (Figure 4), consistent with previous studies over longer time periods covering different seasons [19]. In the most rural areas of Gwangju and Anmyeon, TCNO 2 was significantly lower with average values of 0.31 DU (stdev = 0.13 DU) and 0.27 DU (stdev = 0.14 DU), but still reached as high as 1.05 DU and 1.27 DU, respectively, on certain days. Diurnal variability in NO 2 was not consistent across days or across stations, with maximum values occurring some days in the morning and some days in the early or late afternoon. A clear weekly pattern was observed at the coastal urban sites of Seoul and Busan (Figure 4). Minimum TCNO 2 values were consistently observed on Sundays and maximum values occurred mostly in the middle of the week, consistent with weekly cycles in anthropogenic NO x emissions. This weekly pattern was much less pronounced at the less polluted Gwangju PSI site and absent in the more rural area of Anmyeon (Figure 4). These results agree with previous measurements in the Chesapeake Bay region, where a well-defined weekly behavior in TCNO 2 was captured with Pandora ground-based spectrometers at urban coastal locations [2]. Despite its coarse spatial resolution, satellite imagery from OMI highlighted the role of the two coastal megacities of Seoul and Busan as the major sources of anthropogenic NO x emissions and hot spots of NO 2 over this coastal environment ( Figure 5). Although not as well defined as in the high-frequency ground-based measurements, a weekly NO 2 pattern also emerged from the OMI satellite imagery with TCNO 2 considerably less pronounced over the Seoul and Busan metropolitan areas on Sundays (22 May and 29 May, Figure 5). Still, due to its overpass at 13:30 local time, OMI cannot provide the full range in NO 2 pollution over South Korea, missing the early morning and/or late afternoon rush hour peaks in NO x emissions often observed in urban regions, including Seoul and Busan. Tzortziou et al. [11] previously reported that during March 2012 to March 2013, OMI retrievals showed an average TCNO 2 of 0.5 DU (stdev = 0.21 DU) over Busan, and 0.79 DU (stdev = 0.45 DU) over Seoul, significantly underestimating NO 2 pollution levels compared to continuous measurements from ground-based instruments [11] (Figure 6). Similarly, during May 2016, the average TCNO 2 pollution observed over Busan and Seoul by OMI was 0.32 DU (stdev = 0.17 DU) and 0.75 DU (stdev = 0.48 DU), respectively. These NO 2 levels are 55% and 25% lower compared to average TCNO 2 amounts of 0.71 DU (stdev = 0.39 DU) and 1.01 DU (stdev = 0.64 DU) measured over the same period at Seoul and Busan, respectively, by the ground-based Pandora instruments. On May 18, the PSI at Geoje measured TCNO 2 in the range 0.3-0.9 DU, compared to 0.27 DU observed by OMI over the same location. In the less polluted coastal areas of Gwangju and Anmyeon, the absolute difference in average NO 2 was smaller, with OMI still underestimating TCNO 2 by 37% and 11%, respectively ( Table 2). These differences between OMI and ground-based observations of TCNO 2 are to a large extent due to the coarser resolution of the satellite imagery and the time of the OMI overpass. Uncertainties in the OMI retrievals (e.g., NO 2 profile shape and surface reflectivity) would also be expected to contribute to the discrepancies between OMI and Pandora. Current research is focusing on reducing these uncertainties over land and over the ocean [43,44].

NO 2 and O 3 Dynamics Over the South Korean Coastal Waters
Shipboard measurements from the RV Onnuri for the first time allowed us to capture the strong spatial and temporal variability in air quality over South Korean coastal waters. In agreement with measurements over the land, TCO 3 over the ocean varied by approximately 20% during the KORUS-OC campaign from a maximum value of 340 DU on 20 May to a minimum value of 278 DU on 28 May ( Figure 6). Consistent with the quasi bi-weekly oscillation observed by the ground-based PSI network, relatively higher (>320 DU) ozone values were measured by the RV Onnuri during May 18-20, while TCO 3 remained relatively low (<300 DU) in the southern Korean Peninsula from 26 May to 30 May ( Figure 6). Diurnal variability was small, with TCO 3 over the ocean changing by less than 7% (or, 22 DU) during any single day ( Figure 6). These results are consistent with previous studies on TCO 3 temporal variability measured by Pandora instruments at various mid-to high-latitude sites in Europe and the US [2,11]. The average daily range in TCO 3 observed at urban and rural coastal sites along the Chesapeake Bay shorelines during the 2011 DISCOVER-AQ campaign ranged between 10 and 25 DU [2].
The range in TCO 3 observed over the ocean by the shipboard PSI was in remarkably good agreement with the coarser satellite observations from Aura-OMI ( Figure 6). The mean APD between OMI and PSI observations (averaged over a 1 h window around the OMI overpass time) was 3%, while maximum APD was 7%. As discussed in the previous section, the larger scale synoptic observations from OMI revealed that these dynamics in TCO 3 over the ocean were to a large extent due to the passage of a front transporting air masses with total column ozone amount of 344-400 DU eastward and over the South Korean peninsula during 18 May to 21 May, while relatively low TCO 3 values, <320 DU, persisted over the study region later in May (Figure 3). The range in TCO 3 observed over the coastal ocean by the shipboard PSI was very consistent with ground-based observations at the coastal sites of Busan, Seoul, Gwangju, and Anmyeon, despite the >25 km distance of the RV Onnuri from the shoreline ( Table 2).
Similar to measurements over the land, TCNO 2 over the South Korean coastal waters showed significantly stronger variability and heterogeneity compared to TCO 3 . TCNO 2 varied by more than an order of magnitude, from 0.07 DU to 0.92 DU, both spatially and temporally during the campaign (Figure 7). The highest TCNO 2 values, as high as 0.92 DU, were measured at the beginning of the campaign (18)(19) June 2016) at the KIOST dock at Geoje located at a distance of more than 35 km from the city of Busan. The amount and diurnal pattern in TCNO 2 measured by the shipboard PSI at the coastal site of Geoje on the afternoon of 18 May was in very good agreement with ground-based measurements in Busan (Figure 8), with a mean absolute difference of less than 0.22 DU (Figure 8). At both sites, TCNO 2 changed by more than 0.6 DU within a period of less than 3 h, with values as high as 0.9 DU at Geoje and 1.1 DU at Busan at 14:00 local time, decreasing to 0.3 DU at 17:00 local time. Satellite imagery from OMI, however, showed TCNO 2 of just 0.29 DU over Geoje and 0.55 DU over Busan, 0.5 and 0.6 DU lower, respectively, than coincident (±1 h from OMI overpass) PSI observations (Figure 8).   (Figure 8). OMI was in good agreement with the shipboard PSI under these relatively low pollution levels, slightly overestimating TCNO 2 by 0.07 DU on average (Figure 7).
Significantly higher NO 2 levels were measured by the shipboard PSI on 1 June 2016, in the coastal waters offshore Seoul at a distance of more than 50 km from the shoreline (Figures 1 and 7). Column NO 2 showed strong diurnal variability, changing by 0.37 DU during a period of less than 3 h. Although between 8:00 and 10:00 local time TCNO 2 was relatively low (less than 0.2 DU), it suddenly exceeded 0.5 DU by local noon, decreasing again to <0.25 DU in the afternoon while RV Onnuri was almost stationary (Figure 8). Comparison of the shipboard PSI measurements with the ground-based PSI located near the city of Seoul showed some differences in TCNO 2 during the morning (up to 0.46 DU), very good agreement at around 12 noon when NO 2 peaked over the ocean, and differences as high as 1.0 DU (mean absolute difference of 0.4 DU) in the afternoon (Figure 8). These results suggest influence of different air masses and pollution sources at these two locations over the land and over the ocean. HYSPLIT backward trajectories, discussed in the next section, provided more insights into the impact of different air masses on NO 2 variability over the ocean.

Back-Trajectories of Atmospheric Pollution Plumes
Backward air parcel trajectories were simulated using HYSPLIT4 to determine the origin of air masses sampled by the shipboard PSI on RV Onnuri at Geoje and offshore the coastal sites of Gwangju, Anmyeon and Seoul (Figures 9 and 10). Results for 18 May 2016 (9:00 to 13:00 local time) showed that the air at 1500 m altitude over the coastal area of Geoje had gone through the city of Busan 2-3 h earlier, suggesting transport of pollution from this coastal megacity over the shipboard Pandora (Figure 9a). Similar results were obtained for back trajectories initiated hourly from 14:00 to 18:00 local time. This is consistent with the very good agreement in absolute TCNO 2 values and diurnal pattern measured at Geoje and the city of Busan that day (Figure 8a). Backward trajectories initiated over the location of RV Onnuri on 29 and 30 May also showed little diurnal variability, but in this case, results suggested southward and eastward transport, respectively, of marine air masses (Figure 9b,c). This is in agreement with the low and relatively constant TCNO 2 values measured over the ship on both days (Figure 8b,c).
Backward trajectories run for the location of RV Onnuri on 1 June 2016 highlighted the impact of transport of different air masses on diurnal NO 2 variability over the ocean. Trajectories initiated over the boat (1500 m AGL) before 10:00 local time, when TCNO 2 over the ocean was <0.25 DU, showed transport of air masses originating from a relatively rural area at the border of North and South Korea where TCNO 2 is typically relatively low (Figure 10). At 12:00 local time, however, the air mass reaching the RV Onnuri was coming mostly from the city of Seoul, coinciding with the peak in TCNO 2 observed over the ocean. After 12:00 local time, back trajectory simulations showed a gradual shift of the air mass pathway and origin from a more rural, lower TCNO 2 area south of Seoul (Figure 10), concurrent with a gradual decrease in the TCNO 2 measured by the shipboard PSI. At the same time, TCNO 2 amounts over the city of Seoul continued to rise, reaching maximum values later in the afternoon, associated with rush hour NO x emissions in this urban area (Figure 8).  Simulations were initiated at the ship location (indicated by white square), at 1500 m AGL and 10:00, 12:00, 13:00 and 16:00 local times (KST). Satellite OMI TCNO 2 imagery (from, 27 May 2016, the nearest weekday when OMI data was available) is superimposed on the results for reference.

Implications of TCNO 2 Variability for Remote Sensing of Ocean Color Dynamics
The large spatial and temporal variability in TCNO 2 observed over this coastal environment has strong implications for remote sensing retrievals of ocean biogeochemical variability and ecological processes using ocean color observations. If this variability in TCNO 2 is not properly accounted for in satellite retrievals of ocean color, the decrease in the top of the atmosphere signal caused by NO 2 absorption will be incorrectly attributed to a lower reflectance from the ocean surface resulting in false variability in ocean properties [1,19]. Here, we used results from radiative transfer model calculations previously reported in the literature [19], and our measurements in South Korean coastal waters to examine the impact of observed variability in TCNO 2 on estimates of water remote sensing reflectance, R rs , from space based platforms.
Previous studies using detailed radiative transfer calculations over optically thick coastal waters [1,19] showed that there is significant error in R rs from unaccounted (temporal or spatial) NO 2 variability in atmospheric correction of ocean color observations. For 1 DU of unaccounted NO 2 the error is as large as 57-100% at 412 nm and 28-50% at 443 nm, for SZAs in the 30-60 • range and for NO 2 homogeneously distributed within the first 2 km from the ground [19] (Figure 10). This error increases with increasing solar zenith and viewing angles due to (i) the increasing optical path, and (ii) the decreasing relative contribution of ocean water-leaving radiance to the TOA signal [19].
In South Korean coastal waters, TCNO 2 over the coastal area of Geoje ranged from 0.3 to 0.92 DU on 18 May 2016, compared to the OMI retrieval of 0.29 DU. This corresponds to a TCNO 2 uncertainty of 0.6 DU during the day, with some of the largest differences occurring at 14:00 and 16:00 local time corresponding to 25 • and 50 • SZA, respectively (Figure 11a). If unaccounted in atmospheric correction of coastal ocean color retrievals, the impact of such a change in TCNO 2 would be a false change in R rs (412) by 37% during the day (Figure 11b). The error will be smaller for 443 nm, and almost negligible at 555 nm (Figure 11b), introducing an uncertainty not only in the absolute amount of R rs but also in its spectral shape. Although a number of previous studies demonstrated the advantages of using green-red wavelengths for improved chlorophyll a retrievals in coastal waters [45][46][47], current operational retrievals for standard chlorophyll a products are still based on empirical relationships between chlorophyll a and R rs at blue wavelengths (e.g., 443 and 490 nm) that are considerably affected by variability in atmospheric NO 2 [19]. Moreover, the spectral shape of R rs in the spectral region affected by NO 2 absorption (i.e., <490 nm) is critical for discrimination of phytoplankton pigment assemblages in the ocean and for retrievals of the amount and quality of colored dissolved organic matter, both from current and future multi-and hyper-spectral satellite ocean color sensors [15][16][17][18]48]. In the absence of information on the diurnal variability of TCNO 2 over the ocean (from shipboard or satellite platforms), the diurnal variability in TCNO 2 measured by ground-based instruments at nearby coastal land sites could be used as a reasonable alternative. However, our measurements in this coastal environment reveal that large differences often occur in the diurnal patterns of TCNO 2 over the land and over the adjacent ocean depending on meteorological processes and air-mass transport. In the case of measurements offshore Seoul, the amount and temporal variability in TCNO 2 over the ocean was in relatively good agreement (mean absolute difference of 0.09 DU) with measurements over the coastal city of Seoul at around 12:00 local time when the dominant source of air masses at 1500 m over the RV Onnuri was from the Seoul metropolitan area ( Figure 10). However, considerably larger differences were observed in TCNO 2 amounts between the two locations earlier in the morning and later in the afternoon, as air masses from areas characterized by considerably lower NO 2 pollution were transported over the R/V Onnuri (Figure 10). Differences in TCNO 2 over the two locations in the morning were as high as 0.46 DU at 8:30 local time (SZA of 56 • ). Differences in the afternoon had an average value of 0.4 DU and were as large as 1 DU and 0.96 DU at around 14:00 and 16:30 local time or, 25 • and 53 • SZA, respectively (Figure 11c). Assuming homogeneous distribution of NO 2 within the first 1.5-2 km from the ground, such neglected changes in TCNO 2 would result in a false change in R rs (412) of 43% in the morning and 82% in the afternoon, while changes in R rs (443) would be 21% and 40%, respectively (Figure 11d). Such large errors in the amount and spectral shape of coastal ocean R rs will, in turn, introduce large uncertainties in satellite retrievals of spatiotemporal changes in coastal water quality and coastal ocean biogeochemistry [19].

Conclusions
Variability in atmospheric aerosols and absorbing trace gases remains one of the largest sources of uncertainty for remote sensing of short-term (e.g., hourly to weekly) ocean processes, especially in coastal waters that are close to heavily polluted urban areas. A key objective of the Korea-US Ocean Color (KORUS OC) field campaign was to characterize atmospheric variability in the South Korean coastal waters in the context of the underlying optically complex waters to better understand future retrieval algorithm development needs for both polar orbiting and geostationary ocean color sensors [25].
Our measurements from a shipboard Pandora Spectrometer Instrument (PSI), specifically designed for direct-sun measurements from a moving platform, revealed for the first time the spatial and temporal dynamics that characterize total column (TC) O 3 and NO 2 amounts over the coastal waters of South Korea. In contrast to the relatively low (less than 20%) variability in column ozone, TCNO 2 over the ocean varied by more than an order of magnitude reaching values as high as 0.92 DU during the campaign. Temporal changes in TCO 3 mostly exhibited a quasi bi-weekly oscillation and spatial dynamics were driven primarily by larger scale meteorological processes and synoptic weather fronts that were captured successfully in the relatively coarse satellite imagery from the Aura Ozone Monitoring Instrument (OMI). Variability in TCNO 2 , in contrast, was mostly affected by local anthropogenic emissions and highly dynamic air mass transport pathways. NO 2 emissions over the land considerably affected TCNO 2 dynamics over the ocean, but in complex ways, often resulting in very different temporal patterns between coastal land sites and offshore coastal waters. Although ground-based measurements showed a clear weekly pattern in NO 2 at coastal urban sites such as Seoul and Busan, with minima consistently observed on Sundays, this weekly cycle was not observed over the adjacent ocean. A combination of observations from high-frequency ground-based sensors, synoptic satellite imagery, and air-mass trajectory simulations explained the observed diurnal variability, which varied widely across the coastal ocean domain. A greater than 0.6 DU change in TCNO 2 over a period of less than 3 h in the coastal area of Geoje was mostly due to constant transport of air masses through the city of Busan, resulting in remarkably good agreement in the NO 2 diurnal variability observed at the two locations, despite their distance of more than 35 km. In the coastal waters offshore Seoul, transport of different air masses over the ocean caused TCNO 2 over the RV Onnuri to vary by almost 0.4 DU within a period of less than 3 h while the ship held its position at more than 50 km distance from the shoreline. Backward trajectory simulations initiated at the time of maximum NO 2 over the RV Onnuri, indicated transport from the Seoul metropolitan area, while a gradual decrease in NO 2 over the ocean 2 to 3 h later coincided with a gradual shift in air mass origin from a rural area south of Seoul characterized by considerably lower NO 2 levels. With coarse spatial resolution and an overpass at around 13:30 local time, OMI is not capable of detecting these small-scale, short-term changes in NO 2 observed by high-frequency shipboard remote-sensing sensors over these coastal waters. Observations from the higher spatial resolution (7 × 7 km) polar orbiting TROPOspheric Monitoring Instrument (TROPOMI) are expected to provide more information on spatial variability at a global scale, but will still be limited to one measurement per day temporal resolution. Future geostationary missions, such as NASA's TEMPO (Tropospheric Emissions: Monitoring Pollution), the Korean GEMS (Geostationary Environment Monitoring Spectrometer), and the European Sentinel-4, will provide a unique satellite capability to monitor the spatially and temporally varying TCNO 2 over coastal oceans.
If the observed variability in TCNO 2 is not properly accounted for in atmospheric correction retrievals of coastal ocean color, it will result in a false spatial or temporal variability in ocean biogeochemical properties and ecological processes. We found that uncertainties in TCNO 2 of 0.48 DU, similar to the difference between OMI and shipboard PSI retrievals in the coastal area of Geoje, would result in 37% uncertainty in R rs (412) and almost no change in R rs (555), introducing significant uncertainty in the spectral shape of ocean remote sensing reflectance. Uncertainties in NO 2 diurnal variability over the ocean, on the order of the difference in TCNO 2 measurements between Seoul and its offshore coastal waters on June 1 2016, would result in a false change in R rs (412) and R rs (443) by more than 80% and 40%, respectively, during the day. The resulting errors in R rs subsequently introduce a false variability in retrievals of ocean biogeochemical variables from space, which would be practically impossible to distinguish from real spatiotemporal patterns in coastal ocean dynamics without nearly coincident measurements of NO 2 over the area of interest [12,13,19]. These results are consistent with recent studies by Pahlevan et al. [20] showing that even 10% uncertainty in TCNO 2 would introduce significant, and time-dependent, uncertainties in retrievals of R rs at 412-440 nm in South Korean coastal waters, inhibiting the ability of satellite ocean color sensors to retrieve information on ocean dynamics and biological responses to short-term biogeochemical and physical forcing.
Our measurements demonstrated that variability in TCO 3 over the South Korean coastal waters is mostly driven by large-scale, synoptic features that can accurately be captured by coarser polar orbiting satellite sensors. Short-term (i.e., diurnal to weekly) and small-scale variability in TCNO 2 needs to be taken into consideration when attempting to retrieve short-term coastal ocean processes from satellite ocean color imagery. This is necessary to avoid misinterpreting atmospheric variability as diurnal changes, weekly cycles, or spatial gradients in ocean composition. Since current model predictions of trace gas variability are not adequate for deriving atmospheric corrections, more measurements from shipboard platforms are needed across inland and coastal aquatic environments. Improved modeling is essential for the prediction of atmospheric trace gas variability and its impacts on satellite retrievals of highly dynamic biogeochemical and ecological processes in nearshore ecosystems.