Digital Commons @ Michigan Tech Digital Commons @ Michigan Tech Opportunistic validation of sulfur dioxide in the Sarychev Peak Opportunistic validation of sulfur dioxide in the Sarychev Peak volcanic eruption cloud volcanic eruption cloud

. We report attempted validation of Ozone Monitoring Instrument (OMI) sulfur dioxide (SO 2 ) retrievals in the stratospheric volcanic cloud from Sarychev Peak (Kurile Is-lands) in June 2009, through opportunistic deployment of a ground-based ultraviolet (UV) spectrometer (FLYSPEC) as the volcanic cloud drifted over central Alaska. The volcanic cloud altitude ( ∼ 12–14 km) was constrained using coincident CALIPSO lidar observations. By invoking some assumptions about the spatial distribution of SO 2 , we derive averages of FLYSPEC vertical SO 2 columns for comparison with OMI SO 2 measurements. Despite limited data, we ﬁnd minimum OMI-FLYSPEC differences within measurement uncertainties, which support the validity of the operational OMI SO 2 algorithm. However, our analysis also highlights the challenges involved in comparing datasets representing markedly different spatial and temporal scales. This effort represents the ﬁrst attempt to validate SO 2 in a stratospheric volcanic cloud using a mobile ground-based instrument, and demonstrates the need for a network of rapidly deployable instruments for validation of space-based volcanic SO 2 measurements.


Introduction
Validation of satellite retrievals of trace gases is a crucial part of any mission, but the approach is highly dependent on the species in question. Some molecules (e.g., NO 2 ) have well-characterized sources and somewhat predictable distri-Correspondence to: S. A. Carn (scarn@mtu.edu) butions, making it easier to plan validation campaigns and extended deployments of ground-based or airborne instrumentation (e.g., Brinksma et al., 2008). However, this is not usually the case for volcanic SO 2 , emitted by largely unpredictable volcanic activity, often in remote locations. Validation of satellite SO 2 measurements in such situations is mostly opportunistic (e.g., Carn et al., 2011), and may require rapid mobilization or deployment of ground-based assets (e.g., Spinei et al., 2010). Efforts to validate volcanic SO 2 retrievals have become increasingly important in the light of recent volcanic ash crises (e.g., the 2010 Eyjafjallajökull eruption), during which satellite measurements of SO 2 and ash were used to track the drifting volcanic clouds (e.g., Thomas and Prata, 2011).
We show here that useful SO 2 validation data can be collected by rapid deployment of a simple, mobile ultraviolet (UV) spectrometer system similar to those used widely for volcano monitoring. The opportunity arose when the volcanic cloud produced by the June 2009 explosive eruption of Sarychev Peak (Matua Island, Kuril Is; 48.1 • N, 153.2 • E) drifted over central Alaska. Satellite measurements of SO 2 by the Ozone Monitoring Instrument (OMI) on NASA's Aura satellite were the object of validation. Our measurements represent the first attempt to validate SO 2 in a drifting, stratospheric volcanic cloud using road-based vehicular traverses beneath the plume.
The 2009 eruption of Sarychev Peak began on 11 June. Haywood et al. (2010) provide an overview of the eruption and present SO 2 data from the Infrared Atmospheric Sounding Interferometer (IASI) on the MetOp-A satellite, along with climate model simulations of volcanic cloud dispersion. IASI measured a total SO 2 burden of 1.2 ± 0.2 Tg in the upper troposphere and lower stratosphere (UTLS) following the eruption sequence, which is commensurate with total SO 2 burdens measured by OMI. Dispersion of the volcanic SO 2 after eruption was complex but is not the focus of this paper. An animation of OMI SO 2 data showing dispersion of the Sarychev Peak SO 2 cloud over the Northern Hemisphere from 10 June-31 July 2009 is available as Supplementary Online Material. Beginning on 15 June, the SO 2 cloud began to drift across Alaska, providing the opportunity for validation described herein. In contrast to the SO 2 clouds released by the eruptions of Okmok and Kasatochi (Aleutian Islands) in July-August 2008 (e.g., Spinei et al., 2010), the bulk of the Sarychev eruption cloud resided at high latitudes, presenting few opportunities for ground-based measurements.

Data
The Level 2 OMI dataset used here is derived from version 003 of the operational SO 2 algorithm (hereafter referred to as OMSO2), the theoretical basis of which is described by Yang et al. (2007). We also use the Level 2 OMPIXCOR OMI Ground Pixel Corner Coordinate product. OMSO2 and OMPIXCOR data are publicly available from the NASA Goddard Earth Sciences (GES) Data and Information Services Center (DISC; http://daac.gsfc.nasa.gov/). Operational OMI SO 2 retrievals require an a-priori assumption of the vertical SO 2 distribution (Yang et al., 2007), which is currently addressed by providing retrieved SO 2 columns corresponding to four different a-priori SO 2 profiles. For this analysis we use the mid-tropospheric (TRM) and lower stratospheric (STL) SO 2 products, which correspond to SO 2 layer center of mass altitudes (CMAs) of ∼7.5 . This plot and other coincident CALIOP data can be seen at: http: //www-calipso.larc.nasa.gov/products/lidar/browse images/show date.php?s=production\&v=V3-01\&browse date=2009-06-18. Features showing elevated backscatter at altitudes above ∼11 km are probably all due to aerosol in the Sarychev Peak volcanic cloud, based on collocation with SO 2 measured by OMI (Fig. 1). Low values of the lidar depolarization ratio in these features (not shown) suggest that the aerosol was dominated by liquid sulfate aerosol and/or liquid-coated solid particles. The volcanic cloud filament discussed in this paper is the northernmost volcanic aerosol feature at ∼64-65 • N. and ∼17.5 km, respectively. Overall uncertainty on the OMI SO 2 retrievals (including CMA errors) for SO 2 clouds above 5 km altitude is ∼20 % (Yang et al., 2007). We also employ other NASA A-Train satellite datasets, including Moderate Resolution Imaging Spectroradiometer (MODIS) data from Aqua, and Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) profiles from CALIPSO (Vaughan et al., 2004), to characterize meteorological clouds and assess volcanic cloud altitude (Figs. 1, 2).
Ground-based SO 2 data were collected using a FLYSPEC, a simple but flexible and rapidly deployable UV spectrometer that is calibrated using integral SO 2 gas cells . Conditions were favorable on 18 June 2009 as the Sarychev SO 2 cloud drifted over southern and central Alaska ( Fig. 1; see Supplement). The FLYSPEC was mounted on a vehicle and pointed to zenith, and a high-quality instrument calibration was obtained under clear-sky conditions in Fairbanks (64.84 • N, 147.72 • W) at 21:06 UTC. Measurements of overhead SO 2 column density were then performed south of Fairbanks along Route 3 towards Healy (Fig. 3), at a roughly constant speed of ∼100 km h −1 (∼27 m s −1 ; Fig. 4). While this speed exceeds that typically recommended for road-based traverses of volcanic SO 2 plumes (10-60 km h −1 )  (2003) (1 ppmm = 2.663 × 10 −6 kg m −2 ). Further mobile and stationary FLYSPEC measurements were also made in Fairbanks on 20 June, when the Sarychev SO 2 cloud was again overhead. However, in this case the spatially extensive SO 2 cloud precluded FLYSPEC calibration under clear-sky conditions, and these measurements are not discussed further here. Noise in FLYSPEC SO 2 data has been reported to be ∼0.4-1.1 DU, with errors of up to 6 % on retrieved SO 2 columns based on testing with standard calibration cells .   Table 1. Note visible haze in the volcanic cloud across image center. (b) OMI SO 2 columns (22:28 UTC; colored pixels) and ground-based FLYSPEC SO 2 traverse southbound from Fairbanks (periodically time stamped in UTC). OMI SO 2 color bar is shown in (c); FLYSPEC data color bar is shown in (d); (c) OMI SO 2 columns (22:28 UTC) and ground-based FLYSPEC SO 2 traverse northbound towards Fairbanks; (d) lateral extrapolation of southbound traverse FLYSPEC data to assess effects of spatial averaging. OMI pixel boundaries are overlain in black. a relatively smooth, symmetrical plume cross-section in the southbound traverse, are shown in Fig. 5. The FLYSPEC traverses intersected five near-nadir OMI pixels in orbit 26207 (Fig. 3), and the average FLYSPEC SO 2 columns for each OMI pixel are reported in Table 1. We compute the OMI SO 2 column for SO 2 at 12-14 km altitude (we use 13 km) by linear interpolation between the OMSO2 TRM and STL columns.

Results and discussion
One factor that complicates the analysis is that the volcanic cloud was moving northwards during FLYSPEC data collection. Radiosonde soundings from Fairbanks (http://weather. uwyo.edu/upperair/sounding.html) at 12 Z on 18 June and 00 Z on 19 June indicate southerly to southeasterly winds at ∼2.6-5.1 m s −1 (9.4-18.5 km h −1 ) at 13 km altitude, but with significant wind shear at 10-12 km. This is consistent with the observation that SO 2 was not detected over June show similar wind patterns, although we note that the measurement location is situated in the lee of mountainous topography (the Alaska Range) and locally complex wind patterns could have affected the SO 2 cloud. A consequence of the observed wind speed and direction relative to the size (∼13 × 24 km) and orientation of the OMI pixels (Fig. 3) is that significant advection of SO 2 would have occurred on timescales of ∼40-80 min (i.e., the time taken for winds at ∼10-20 km h −1 to travel 13 km). For the spatially heterogeneous Sarychev SO 2 cloud (Fig. 5), this complicates the interpretation of FLYSPEC measurements collected more Collection 3 OMSO2 data for a SO 2 CMA of 17.5 km (STL product). 5 TRM and STL OMSO2 columns linearly interpolated to a SO 2 cloud altitude of 13 km. 6 Average FLYSPEC SO 2 column for southbound traverse from Fairbanks. Data in parentheses account for unweighted and weighted spatial averaging across the OMI pixel, respectively. 7 Average FLYSPEC SO 2 column for northbound traverse towards Fairbanks. Data in parentheses account for unweighted and weighted spatial averaging across the OMI pixel, respectively. 8 OMI-derived cloud fraction in pixel. 9 Rate of change of SO 2 column in each pixel based on southbound and northbound FLYSPEC surveys. Data in parentheses correspond to unweighted and weighted spatial averages within the OMI pixel, respectively. 10 Estimated average FLYSPEC SO 2 column at time of OMI overpass (22:28 UTC). Data in parentheses correspond to unweighted and weighted spatial averages within the OMI pixel, respectively. than ∼40-80 min before or after the OMI overpass. This applies to pixel P1 in the southbound FLYSPEC survey, and all pixels with the exception of P5 in the northbound survey (Fig. 5).
Although MODIS data indicate possible meteorological cloud interference, OMI cloud fractions were quite low in most pixels (with the exception of pixel P1; Table 1) and OMI cloud pressures (not shown) imply predominantly lowaltitude clouds which would not significantly impact OMI retrievals of stratospheric SO 2 . The effect of clouds on the FLYSPEC measurements is unclear, but the general smoothness of the SO 2 cross-sections (Fig. 5a) suggests no significant cloud interference in most pixels.
Comparison between the OMI SO 2 columns and the average FLYSPEC SO 2 columns for each OMI pixel reveals some significant differences (Table 1), but the relative timing of the measurements and spatial averaging of SO 2 over the OMI pixels must be accounted for. In the absence of any ground-based constraints on horizontal variations in SO 2 column, we assessed the effects of spatial averaging using a novel approach facilitated by A-Train satellite synergy. This technique involved extending the FLYSPEC SO 2 columns laterally to simulate the 2-D distribution of SO 2 in the volcanic cloud (Fig. 3d). The azimuth for this extrapolation is constrained using the location of maximum SO 2 in the FLYSPEC profile (64.27 • N, 149.03 • W; Fig. 5) and the location of the matching volcanic aerosol feature in CALIOP data (64.49 • N, 145.96 • W; Fig. 2), giving an azimuth of ∼81 • . This is qualitatively consistent with the SO 2 distribution mapped by OMI (Fig. 3), but the technique clearly fails to account for any heterogeneity in the volcanic cloud. We then computed the mean of the simulated 2-D SO 2 columns over the OMI pixels (Table 1), resulting in adjustments of only a few percent relative to FLYSPEC traverse averages for some pixels, but more substantial (∼10-130 %; Table 1) for others. Note that for some pixels (e.g., P1 in the southbound survey) the spatial average of the extrapolated FLYSPEC  Table 1 are shown at top), extrapolated FLYSPEC SO 2 columns averaged (unweighted) over corresponding OMI pixels (horizontal purple lines) and interpolated OMI SO 2 columns in the same pixels (horizontal green lines). The abscissa shows FLYSPEC time; time relative to the Aura overpass is indicated in parentheses. (a) Southbound traverse from Fairbanks; (b) northbound traverse towards Fairbanks. The absence of a peak in (b) is due to northward drift of the volcanic cloud during measurement. SO 2 columns exceeds the mean of the raw FLYSPEC data for that pixel (Fig. 5a). In the case of pixel P1, this arises because the FLYSPEC traverse only characterized the northern half of the pixel, whereas the extrapolated SO 2 amounts are higher in the southern half (Fig. 3d).
To account for temporal differences between the OMI and FLYSPEC measurements, we first calculate the temporal gradient in SO 2 column amount for each pixel, based on the FLYSPEC measurements preceding and following the OMI overpass (Table 1). The resulting gradients are ∼ ±4-6 DU h −1 for all pixels except P3. We then interpolate the Fig. 6. Histograms of SO 2 column amount for each OMI pixel surveyed by FLYSPEC during the southbound traverse from Fairbanks (Fig. 3). Each histogram pair shows the distribution for raw FLY-SPEC measurements on the left, and unweighted spatially-averaged FLYSPEC data on the right. The ordinate shows the number of data points in each bin normalized to the mode. SO 2 column amounts for each pixel to the OMI overpass time (22:28 UTC) using these gradients (Table 1). These calculations also allow us to assess the relative magnitude of the spatial and temporal SO 2 gradient for each pixel. In pixels P2, P4 and P5, which were surveyed in ∼20 min or less (Fig. 5), the large spatial gradients in SO 2 column (Figs. 4,5) in either one or both of the FLYSPEC traverses indicate that adequate spatial characterization was required to validate these pixels. In pixels P1 and P3, the temporal gradient dominates, although as noted above P1 may have been impacted by a large meteorological cloud fraction (Table 1), and P3 is very poorly characterized in the spatial domain (Figs. 3, 4, 5).
Two important results arise from this analysis. Firstly, we find the best agreement between the OMI and spatially averaged FLYSPEC measurements for pixel P2 (13 % difference; Table 1). We attribute this to the fact that pixel P2 has the best FLYSPEC data coverage of all the pixels, both along-track and across-track with respect to the OMI pixel geometry (Fig. 3), providing the best constraints on the subpixel SO 2 distribution, despite a ∼30-45 min time difference between the OMI and FLYSPEC measurements. This is corroborated further by inspection of histograms of SO 2 column amount for each pixel (Fig. 6). These histograms show that the FLYSPEC traverse data for pixel P2 best match the mode and distribution of SO 2 column amounts predicted for the entire pixel (Fig. 3d), which is a result of the good spatial characterization of this pixel.
Secondly, we also find good agreement between the OMI measurements (3.1 DU) and the interpolated FLYSPEC data for pixel P5 (∼11-19 % difference; Table 1). We attribute this to the fact that the FLYSPEC data were collected closest in time to the OMI overpass (Fig. 5). For the other pixels, the differences between the FLYSPEC and OMI measurements are significant and generally increase with temporal offset from the OMI overpass time (Table 1). This may indicate that the SO 2 columns in each pixel were changing in a non-linear manner during the period of data acquisition, and hence the assumption of a linear SO 2 column gradient, coupled with uncertainties on the spatial SO 2 distribution, was inadequate for most pixels.
Differences between OMI and spatially averaged FLY-SPEC SO 2 columns are larger for the other analyzed pixels, particularly around the location of maximum SO 2 (Table 1; Fig. 5a). This is perhaps best explained by spatial heterogeneity in the volcanic cloud that is not captured in our simple 2-D model of the plume (Fig. 3d). Furthermore, pixels P3 and P4 were not well-characterized by the FLY-SPEC data and the traverses were close to the pixel boundaries, where the precise definition of the OMI spatial resolution becomes critical due to the variable OMI pixel spatial response function (SRF). All these factors may have contributed to the lower SO 2 column measured by OMI in these pixels. It is likely that the larger cloud fraction in pixel P1 caused the relatively poor agreement for the southbound traverse ( Fig. 3; Table 1). Increasing discrepancies are seen for the northbound FLYSPEC traverse data (Fig. 5b) as by this time the SO 2 cloud had moved north over Fairbanks.
It is important to note here that the spatial resolution of OMI (or any CCD-array based spectrometer) cannot be simply represented by the mapped pixel edges (Figs. 3, 5). The SRF in both the flight-and across-track direction is approximately Gaussian in shape (Dobber et al., 2006). More precisely, the spatial resolution in the flight direction (∼13 km at nadir) is defined as the full-width at half-maximum (FWHM) of the telescope instantaneous field-of-view (IFOV; ∼1 • ) convolved with a 2-second integration time, while in the across-track direction it is the FWHM of the sum of the Gaussian SRFs of 8 binned OMI CCD pixels (∼24 km at nadir). Therefore, each OMI pixel is also influenced by photons received from beyond the mapped pixel boundaries. A detailed analysis of the OMI spatial resolution is beyond the scope of this paper, but we assessed the effect of non-uniform spatial response by weighting the FLYSPEC SO 2 columns with a flat-topped Gaussian function (of the form f(x) = exp(c(x − x 0 ) 4 ), where x − x 0 is absolute distance from the pixel center (x 0 ) in the along-track direction, and c is a constant) to simulate an along-track SRF with FWHM corresponding to the mapped pixel boundaries. No attempt was made to account for contributions from beyond the FWHM pixel boundary, or to simulate the across-track SRF. The resulting weighted, spatially averaged FLYSPEC SO 2 columns are ∼13-24 % lower than the unweighted averages (Table 1) and improve some of the comparisons. For pixel P2, the OMI-FLYSPEC difference for the southbound traverse reduces to ∼6 %, but for pixel P5, the OMI-FLYSPEC difference for the interpolated SO 2 column increases to ∼30 %. Significant differences remain for the other pixels, which we conclude is probably due to unconstrained sub-pixel spatial heterogeneity in the volcanic SO 2 cloud.

Conclusions
As this study demonstrates, validation of volcanic SO 2 measurements is challenging, since acquisition of high-quality ground-based data may be precluded by time constraints. Nevertheless, we find good agreement between selected OMI and spatially averaged FLYSPEC SO 2 data when some necessary assumptions about the 2-D SO 2 distribution are invoked, providing additional support for the validity of the operational OMSO2 dataset (see also Spinei et al., 2010;Carn et al., 2011). Unresolved spatial heterogeneity in the volcanic cloud on a sub-OMI pixel scale coupled with non-linear temporal changes in SO 2 column amount appear to be the best explanations for other observed OMI-FLYSPEC differences.
On the basis of this analysis, validation of satellite SO 2 measurements in drifting volcanic clouds using sensors mounted on mobile platforms would only be recommended when the SO 2 cloud is moving at a velocity much lower than that of the ground-based (or airborne) sensor. For clouds drifting at higher velocities, stationary ground-based measurements would provide better and more easily interpretable validation data (e.g., Spinei et al., 2010). The advantage of using mobile platforms is that multiple contiguous satellite sensor pixels can be surveyed. Furthermore, our analysis confirms that spatial coverage of pixel areas in both alongand across-track directions should be maximized to increase the likelihood of successful validation.
We conclude by stressing the uniqueness of our FLY-SPEC SO 2 dataset for the Sarychev Peak eruption, one of the last decade's largest explosive eruptions. As discussed above, the Sarychev volcanic cloud mostly resided at high Arctic latitudes, precluding widespread ground-based measurements. Perro et al. (2010) report ground-based Brewer spectrophotometer detection of SO 2 from Sarychev over Eureka, northern Canada on 1 July 2009. To our knowledge, the FLYSPEC SO 2 data presented here are the closest measurements to the eruption in time and location. The successful ground-based measurement of the volcanic SO 2 cloud underlines the need for a widespread network of rapidly deployable instruments in order to successfully validate volcanic SO 2 retrievals from satellite sensors.