The measurements reported here from 20:25 KST 21 May through 09:00 KST 4 June 2016 (Korean standard time, KST = UTC+9) were made outside of South Korea's territorial seas (> 12 nautical miles, 22.2 km, from the coast, Fig. 1a). The instrument suite (Fig. S2) was deployed above the bridge strapped to the starboard rail in a custom-built box designed to keep the instruments dry yet ventilated to prevent overheating (Fig. S1). The pumps were located in a separate box a few meters away tied to the stern rail (Fig. S1) to limit thermal and mechanical interference with the measurements. Measurements were made at ambient temperature (T), pressure (P), and relative humidity (RH), (i.e., the aerosols were not dried prior to measurement) with the exception of the TAP instrument which is internally heated and kept at a constant 35 ∘C. T, P, and RH were measured both aboard ship and within the sampling system. Although the sampling box was ventilated, it was not climate controlled. On the top deck it was in direct sunlight during the day, and even at night various components in the system kept the interior warmer and drier than ambient air. These conditions led to diurnal variability in the difference between sampling and ambient conditions, such that the sampling T was ∼ 6 ∘C warmer at night, increasing to as much as ∼ 15 ∘C warmer in mid-afternoon. Similarly, ambient RH ranged from 55 %–98 % RH throughout the cruise, while the sampling RH ranged from ∼ 30 %–70 %. For the purposes of this work (both Parts 1 and 2), the comparisons made are all within this sampling system with T and RH consistent across all measurements, except for TAP. Since non-volatile black carbon (BC) is expected to dominate the absorption measurement, this difference is expected to have a limited impact on the results. The primary objective in this work is to evaluate the performance of SpEx by direct comparisons to the data from the two commercial instruments in the measurement suite. Hence, we did not perform any corrections to either ambient T and RH or to standard temperature and pressure (STP) prior to comparing these data. There are no comparisons in this pair of papers to other data sets, so such corrections are not necessary for this study.

The available berths aboard ship constrained the number of personnel available to operate the sampling system, so it was necessary to configure the system to operate nearly autonomously. To ensure that water did not enter the sampling line, a tall stainless-steel sampling mast (19.05 mm (3/4 in.) inner diameter tubing) was used to minimize the chance of sea spray entering the line from below. Further, the inlet was attached to the vertical sampling mast with a curved section of stainless-steel tubing such that the inlet was downward facing to prevent any potential precipitation from entering the line from above as well (Fig. S1). This configuration worked as intended with the inlet approximately 10 m above the sea surface. However, the fast flow rate required for the SpEx measurement (approximately 70 Lmin-1) through the curved inlet resulted in the removal of coarse aerosol with a 50 % size cut of ∼ 1.3 µm diameter (Fig. S3). Hence, all of the measurements here (sampling from the same inlet, Fig. S2) reflect only the fine-fraction aerosol in the marine boundary layer (MBL).

While the SpEx and commercial instruments were run mostly uninterrupted, colleagues from other groups aboard ship carried out the routine filter changes needed throughout the cruise. They also ensured that the system ran properly and were there to handle problems. The methodology and results of the filter sampling will be presented in the companion paper (Part 2, Jordan et al., 2021).

The IN101 and TAP instruments provide data at a higher temporal resolution than SpEx and were deployed with two objectives: (1) to identify and flag incidents of ship exhaust (“plume”) contamination of the data set and (2) to evaluate the new spectral measurements (both from SpEx and the filters). The TAP (model 2901, Brechtel, Hayward, CA) measures absorption coefficients (σabs) at 467, 528, and 652 nm with 1 s resolution, and the IN101 (AirPhoton, Baltimore, MD) measures scattering coefficients (σscat) at 450, 532, and 632 nm with ∼ 10 s resolution.

To identify ship plume interceptions, an initial examination of the IN101 σscat and TAP σabs data was performed. One-minute averages were calculated in order to calculate the single-scattering albedo (ω=σscat/(σscat+σabs)). The σscat data were corrected using the submicron factors in Anderson and Ogren (1998). These corrections are needed to resolve truncation errors in the forward scattering (the 7 to 0∘ range missing from the measurement) and non-Lambertian errors that arise from the distribution of light by the opal glass diffusor. Due to limited personnel, calibrations were not performed during the cruise. Pre- and post-cruise calibrations using pure CO2 were performed in the laboratory to correct the final archived data and indicate the instrument performance was stable throughout the measurement period. The σabs data were corrected for scattering as recommended in Ogren et al. (2017).

Interception of the ship plume was first identified using 1 min averages of the gas-phase NO2 and O3 measurements (Thompson et al., 2019). The inlets for the gas-phase instruments were near but not co-located with the aerosol inlet, so the 1 min averages of ω, σabs, and σscat were used to further assess evident ship plume interceptions in the aerosol data set. The σabs data were particularly sensitive to such interceptions exhibiting dramatic brief spikes in the otherwise smoothly varying time series (the interested reader can compare σabs in Fig. S4, which shows all of the data in the time series, including ship plume interceptions to σabs in Fig. 2a, where the plume interceptions have been removed). These were easily identifiable and removed manually. The inlets for the in situ instruments were towards the starboard bow (gas phase on a lower deck, a few meters aft of the aerosol inlet) opposite of the ship stack pointed toward the port side stern. While underway, ship plumes did not affect the measurements, but they were encountered occasionally when on station or, as was more frequently the case, when getting underway from a stationary sampling site. The archived plume flags (= 0 for ambient air, 1 for ship plume interceptions) may be used to either remove data points affected by ship contamination or to assess ship plume contamination of filter samples (see Part 2).

In addition to the 1 min averaged σscat and σabs data reported in the KORUS-AQ data archive, averages were also calculated for each 30 s SpEx sampling interval and corrected in the same manner as the 1 min data set. The IN101 and TAP instruments measure at different wavelengths, so in order to calculate σext (=σscat+σabs), the TAP data were adjusted to the IN101 wavelengths using Eq. (). For clarity, these extinction coefficients are denoted NT (for nephelometer + TAP) σext in comparison to the measured SpEx σext at those wavelengths (450, 532, and 632 nm). Similarly, a second set of plume flags was created for the SpEx sampling interval (i.e., based on plume interceptions for 30 s intervals approximately every 4 min; see Sect. 2.3).

Developed from a prototype described in Chartier and Greenslade (2012) SpEx is described in detail in Jordan et al. (2015). SpEx is a White-type optical cell (White, 1942) with a 39.4 m path length and a 17 L internal volume. A flush time of 90 s completely exchanges the air in the cell (3 times the volume) for flow rates > 34 Lmin-1; here, the flow rate was ∼ 70 Lmin-1. The optical cell is coupled via fiber optics to a UV5000 system (Cerex Monitoring Solutions, LLC, Atlanta, GA) with a 150 W xenon lamp source (Cerex P/N CRX-X150W), integrated with an Ocean Optics, Inc. (Dunedin, FL) QE65Pro 16 bit spectrometer. Sampling for 30 s using an integration time of ∼ 20–50 ms optimizes the signal-to-noise ratio for each measured intensity spectrum (Jordan et al., 2015). An automated valve system controls the 4 min sampling cycle by switching the flow between the filtered line (ambient air without aerosols) and an unfiltered line (ambient air with aerosols): 90 s flush, 30 s sample without aerosols, 90 s flush, 30 s sample with aerosols, and repeat.

The intensity spectrum is measured for the sample without aerosols for a reference (I0(λ)) and for the sample with aerosols (I(λ)) from which the extinction spectra (σext(λ)) are calculated using the extinction law (Eq. ), 4σext(λ)=-LN(I(λ)/I0(λ))L, where λ is the wavelength of light and L is the optical path length (here, in units of Mm). Hence, extinction spectra are acquired every 4 min. Reference spectra before and after each sample spectrum are averaged together to account for drift in the intensity between measurements in the calculation of σext. A particular strength of this measurement is it explicitly accounts for extinction arising from gas-phase constituents via the reference spectra. No calibrations are needed to obtain aerosol σext. Laboratory tests have shown the measurement error over the full spectral range of the measurements (300–700 nm) to be about ± 5 Mm-1 (Jordan et al., 2015).

Several modifications have been made to the instrument since the laboratory studies reported in Jordan et al. (2015). These include an automated switching system between valves that control flow between filtered and unfiltered air and new custom control software that allows for continuous unattended operation. Further, the metal-jacketed optical fibers used previously were replaced with bare optical fibers securely packed in foam to reduce noise when deployed on mobile platforms. This last change resulted in greatly improved data quality with far fewer spectra disrupted due to mechanical disturbance under field conditions. Nonetheless, some spectra had to be rejected from the final data set due to features that provided clear evidence of disturbance. Typically, this occurred around the times of the filter changes when the lid of the box housing the sampling system needed to be raised to change filters (flagged using the plume flag field, value set to 2; see Fig. S4). Sometimes the filters were changed during flush times, leaving measured spectra unaffected, while at other times, disturbed spectra persisted for some period of time around the filter change. This suggests other vibrations arising from work involving nearby instruments or elsewhere on the ship may have contributed to the noisy rejected spectra. More work is required to further reduce the susceptibility of SpEx to sources of vibrational noise; nonetheless <8 % of the 4255 measured spectra were rejected from the data set overall. Note that ship plume interception (affecting <6 % of the measured spectra) often resulted in distorted spectra likely due to rapid changes in both the aerosol population and the gas-phase concentrations used to establish the reference spectrum for the calculation of aerosol extinction as in Eq. (). Hence, only spectra obtained when the plume flag = 0 should be used for analysis of ambient conditions.

One consequence of unattended operation is the intensity drift sometimes resulted in saturated pixels around 332 and 467 nm due to peaks at those wavelengths in the lamp spectrum (Fig. S5). Saturation can be prevented either by adjusting the integration time of the spectrometer or by realigning the optics. However, unattended operation necessitated filtering the spectra at those two wavelengths when saturation occurred. The 16 bit spectrometer has a maximum intensity count of 65 536 (216); however, saturation effects became apparent before reaching this maximum. Tests showed that a threshold of 63 000 counts was a suitable limit for removing saturation effects from the spectrum. Filtering the 467 nm channel and the two adjacent pixels on either side of that channel removed the saturation distortions completely. The saturation effects at 332 nm were filtered in the same way; however, there appeared to be a minor shoulder effect that extended over a broader wavelength range in the UV. The effect was minimal (a few percent of the measured value) and well within the measurement error. However, caution is recommended against overinterpreting the shape of a spectrum in the vicinity of 332 nm when the saturation flag (Sat332Flag) for this channel is set to 1. Saturation at 332 nm affected 16 % of the total 4255 measured spectra.

Four distinct synoptic meteorological regimes have been described in detail for the KORUS-AQ period (Peterson et al., 2019). The KORUS-OC cruise departed the peninsula (Fig. 1a) sailing to the East Sea (Sea of Japan) near the end of the second of these periods, the Stagnant period, characterized by limited transport and enhanced photochemical production of secondary organic aerosols (SOAs; Kim et al., 2018; Nault et al., 2018; Choi et al., 2019; Jordan et al., 2020). This regime started breaking down on 23 May, followed by a precipitation event on 24 May that reduced ambient aerosols to low concentrations as reflected by low (tens of Mm-1) σscat (Fig. 2a).

From 25 through 31 May the Transport/Haze period was characterized by air mass transport (from the west/northwest carrying pollutants from China), overcast hazy conditions, and rapid local South Korean secondary production of inorganic aerosols resulting in the largest concentrations of the PM2.5 fraction of aerosols (i.e., particulate matter with diameters ≤ 2.5 µm) observed during the KORUS-AQ campaign (Peterson et al., 2019; Eck et al., 2020; Jordan et al., 2020). The greatest σscat values (hundreds of Mm-1) observed aboard the R/V Onnuri were found during the first half of the Transport/Haze period while the ship was downwind of the Korean peninsula in the East Sea (Figs. 1a and b, and 2a). Lower σscat values were observed during this period when the ship was upwind of the peninsula following its transit to the West Sea (Yellow Sea).

The final of the four synoptic periods, the Blocking period, followed a frontal passage that ended the previous Transport/Haze period (Peterson et al., 2019). This final frontal passage swept in cleaner air from the north, leading to a rapid decrease in aerosol concentrations reflected in the reduced σscat observed aboard ship (Fig. 2). The Blocking period was then characterized by limited transport and occasional brief stagnant periods due to adjacent high- and low-pressure systems with the high poleward of the low (called a Rex block). Under these conditions local sources dominated pollutants, but aerosols did not accumulate to large concentrations (Peterson et al., 2019; Jordan et al., 2020). This led to strikingly low σscat values (tens to < 10 Mm-1, Fig. 2), even lower than those observed during the precipitation event on 24 May.

Throughout the cruise σabs values were typically an order of magnitude smaller than σscat (Fig. 2a) with peak values generally found at the same locations and times as peak σscat (Fig. 1). Hence, while the temporal variability of σabs largely followed σscat, the range in the magnitude of σabs (0.36–18.07 Mm-1) at 532 nm was far less than that of σscat (2.8–332.2 Mm-1). This led to the finding that single-scattering albedo values (ω(λ)=σscat(λ)/σext(λ)) were driven by the change in scattering not absorption. Typically, ω was > 0.9 (Figs. 1d and 2c). The excursions below this value observed on 23 and 24 May occurred when the reduction in σscat exceeded that observed in σabs.

The most extreme low values of ω (∼ 0.7 for 532 nm, Figs. 1 and 2), however, occurred during the Blocking period, when the temporal evolution of σabs did not closely follow σscat. At that time, the scattering Ångström exponents (αscat) increased to values > 3, while absorption Ångström exponents (αabs) were ∼ 1 (Fig. 2b). These data indicate that the aerosol population was dominated by small particles, likely black carbon (BC). The temporal evolution of carbon monoxide (CO, Fig. 2c) and σscat (Fig. 2a) was strikingly similar from 04:00 1 June through 20:00 2 June, exhibiting an r2 = 0.841 for a linear regression over that time. This is in contrast to the Blocking period as a whole (r2 = 0.514) or the rest of the campaign, excluding the Blocking period (r2 = 0.623). The good correlation with CO, the BC signature in αabs, the small particle population indicated by αscat, and the limited transport of the Blocking period together suggest that the low ω values arose from local ship emissions from either commercial or fishing vessels or both. Note that the inference that aerosols during the Blocking period were from local ship emissions refers to the regional ambient environment and should not be confused with ship plume contamination from the R/V Onnuri (see Sect. 2.2 for the criteria used to remove ship stack contamination from the data set).

Aside from the Blocking period, αscat typically ranged from ∼ 1.5–2, with evident wavelength dependence in the αscat wavelength pairs (Fig. 2b). Little difference was found among the wavelength pairs in the αabs values that typically ranged from ∼ 0.5–1 throughout the cruise.

σext values calculated from the σscat+σabs data (denoted NT σext to distinguish it from SpEx σext) averaged over the SpEx sampling intervals (see Sect. 2.2) were compared to the 450, 532, and 632 nm channels of SpEx (Fig. 3). Excellent agreement was found with slopes of 1.020 ± 0.002, 0.998 ± 0.003, and 1.057 ± 0.004 for all of the data in each of the three channels (gray plus markers in Fig. 3), respectively. Three intervals were used to look at mean comparisons: 15 min (light colored circles), 30 min (dark colored triangles), and 60 min (black diamonds). The slopes, intercepts, and r2 values for all of the fits are shown in Fig. 3. These fits were performed using data limited to the valid measurement range (i.e., above the lower limit of detection, LLOD). Tests indicated that a limit of twice the measurement uncertainty provided a suitable LLOD (10 Mm-1).

The SpEx data were more variable than the NT σext as is evident in time series comparisons throughout the cruise (Fig. 4a and Figs. S6 and S7, top panel). This is partly attributable to the differing noise characteristics of the measurement techniques, but it also arises from differences in sampling intervals where the standard error of the means reduces the variability by the square root of the number of samples in the mean. The NT σext represent 30 s means calculated from ∼ 10 s σscat and 1 s σabs measurements for each SpEx σext spectrum. Averaging SpEx σext reduces the variability (Fig. 4b and Figs. S6 and S7, middle panel) according to the standard error of the means. However, not all of the variability evident in Fig. 4 is noise. Limiting the data range in the time series to 2 d (Fig. 5) illustrates that the native resolution of SpEx captured rapid changes that were also evident in the NT data. For example, consider the double peak feature that occurred around 03:00 on 25 May (Fig. 5a). This temporal variation is lost even in the 15 min average (Fig. 5b).

For clarity, the SDs for the means are shown separately (Fig. 4c and Figs. S6 and S7, bottom panel). Most of the SDs of the means in these channels were consistent with the measurement error of about ± 5 Mm-1 with larger values arising from changes in the ambient conditions. Hence, the SDs of the 60 min means describe an upper envelope compared to the 15 and 30 min means as changing ambient conditions led to greater variability in the means. Together, the results shown in Figs. 3–5 show that quantitative data were obtained under field conditions with SpEx at its native sampling resolution.

With the spectral data available from SpEx, it was possible to test the power law behavior of the individual and mean spectra by fitting a line to each spectrum in logarithmic space per Eq. (), where the Ångström exponent, αext, is the negative value of the slope. As discussed in the introduction, if a power law described the relationship well, αext should be invariant with wavelength. Yet clear separation was found in αext from linear fits to all of the individual spectra (all data, Fig. 6a) depending on the wavelength range used for the fit. The αext from fits to the full range (300–700 nm) and two partial ranges (450–532 nm) and (532–632 nm) deviate from the 1:1 line expected when plotted vs. αext from fits to the 450–632 nm range. This result was also found for αext determined from all of the mean spectra sets as well (e.g., those from the 30 min mean spectra, Fig. 6b).

Most of the SpEx spectra measured during the cruise exhibited curvature over the 300–700 nm range in logarithmic space. Note that in order to compare results here to previously published work (Sect. 4), for the remainder of this section fits will be shown using wavelength in units of micrometers (µm). This change in units does not alter the values of αext (which are invariant with wavelength, and hence the choice of units does not matter), but it does change the other coefficients from the mathematical fits to the spectra as will be discussed further in Sect. 4. An example of the observed curvature is provided in Fig. 7. The measured spectrum shown (red curve, Fig. 7a–f) is curved such that residuals (the difference between the measured spectrum and the mathematical fit, blue curves) from the linear fits (black lines, Fig. 7a, b and c) are also curved. If a particular mathematical function (here, a power law) provides a good fit to the data, the residuals should be randomly distributed around zero. If there is a trend (the curvature evident here), then other functions should be considered to see if a better fit may be obtained. The curvature in the residuals is evident not only across the full wavelength range (Fig. 7c), but for subsets of that range as well (e.g., the 0.45–0.632 µm fit, Fig. 7a). Note that the log scale used to plot the spectrum (red curves) in Fig. 7 along with the relatively small extinctions at long wavelengths exaggerates the appearance of noise at those wavelengths (i.e., ± 5 Mm-1 at the red end where σext ∼ 20 Mm-1, LN(20) ∼ 3, is more obvious than in the UV where σext ∼ 150 Mm-1, LN(150) ∼ 5). Nonetheless, the intensity of the xenon lamp decreases from 600 to 700 nm (Fig. S5; in Fig. 7, LN(0.6 µm) ∼ -0.51, LN(0.7 µm) ∼ -0.36) such that smaller values of I0 combined with the small differences between I and I0 in this wavelength range lead to slightly greater uncertainty in σext (Eq. ).

The wavelength-dependent curvature can lead to large errors if used to extrapolate to wavelengths beyond the measured range of values. For example, using αext found from the 0.45–0.632 fit to extrapolate to 0.3 µm (Fig. 7b) leads to a value of LN(σext)=5.7099, while the measured value was 5.0839, i.e., an extrapolated extinction of 302 Mm-1, 87 % larger than the measured extinction of 161 Mm-1. This wavelength is the most extreme, but the positive artifact is present throughout the UV range: with the extrapolated extinction 73 % and 29 % greater at 0.315 and 0.365 µm, respectively. At 0.45 µm (LN(0.45 µm) = -0.8, Fig. 7), the lower end of the fit range of values, the fit extinction was 6 % greater than the measured value, while at 0.532 µm (the middle of the fit range) it was 3 % less.

Previous work has shown that second-order polynomials in logarithmic space can provide a better fit to ambient aerosol optical depth (AOD, aerosol extinction integrated over an atmospheric column) spectra than power law fits (e.g., Eck et al., 1999, 2001a, b, 2003a, b; Schuster et al., 2006; Kaskaoutis et al., 2010, 2011). In the example shown in Fig. 7, it is clear that a second-order polynomial fit (black curves, Fig. 7d–f) reduces the trends in the residuals both over the full wavelength range (Fig. 7f) and over the 0.45–0.632 µm subset of that range (Fig. 7d) from that obtained from the linear fits. The improved fit provided by a second-order polynomial for all of the measured spectra is shown using histograms of the residuals at six wavelengths across the measured wavelength range (Fig. 8). For both the linear fits (black bars) and the second-order polynomial fits (red bars), the range of values for residuals at each wavelength is divided into 20 bins. The narrower bins for the second-order polynomial fits reflect the smaller range in residual values compared to those from the linear fits. The best agreement between the two sets of residuals shown in Fig. 8 is found at 0.532 µm, where the linear fit residuals are distributed around zero. At longer and shorter wavelengths, however, the linear residuals tend to be either positive or negative at any given wavelength, while the second-order polynomial fit residuals are centered around zero across all wavelengths. These results confirm that second-order polynomials provide a better fit to the data than linear fits.

Unfortunately, just as extrapolating linear fits beyond the measurement range is problematic, the same is also true for the second-order polynomial fits (Fig. 7e). In this case, at 0.45 and 0.532 µm (within the fit range) the fit agrees with the measured extinctions within 1 % (∼ 85 and 56 Mm-1, respectively). However, at increasingly shorter wavelengths the fit diverges from the measured spectrum with the fit values 15 %, 28 %, and 34 % too small at 0.365, 0.315, and 0.3 µm, respectively. The divergence of either fit from the measured spectrum when extrapolating beyond the fit wavelength range (Fig. 7b and e) highlights the need for measurements across a broad spectral range in order to minimize the need for extrapolation.