Liquid Containing Clouds at the North Slope of Alaska Demonstrate Sensitivity to Local Industrial Aerosol Emissions

Cloud condensation nucleus control alter cloud solar albedo through cloud droplet size.

. Simultaneously, liquid droplet characteristics affect ice crystal properties through precipitation process such as riming (e.g., Borys et al., 2003;Hallett & Mossop, 1974), meaning CCN concentrations also influence precipitation, cloud lifetime and radiative effects (Norgren et al., 2018).The presence of liquid water may even be a prerequisite for ice formation at moderate temperatures (de Boer et al., 2011), and increased concentrations of large drop sizes have been associated with ice production (Lance et al., 2011;Rangno & Hobbs, 2001).Recent modeling work to untangle these competing effects concluded that INP perturbations dominate proportional increases in CCN (Solomon et al., 2018).
Statistical analysis across the seasonal cycle of Arctic aerosol properties provides opportunities for understanding key relationships (Coopman et al., 2018;Garrett & Zhao, 2006;Zamora et al., 2017) but the quantification of aerosol influences on clouds is additionally muddied by the conflation of these effects with meteorological drivers (i.e., changes in moisture content and dynamics, Feingold et al., 2016;Gryspeerdt et al., 2016;Rosenfeld et al., 2019;Sena et al., 2016).To circumvent this issue, Arctic ship emissions are used as a natural laboratory (Gilgen et al., 2018;Possner et al., 2017), but are limited in extent and occurrence.
In this study, the natural laboratory concept is expanded by investigating aerosol influences on clouds using surface-and satellite-based observations from Northern Alaska.Petroleum extraction in and around the Prudhoe Bay Oilfield has resulted in continuous release of anthropogenic emissions over the past several decades with anthropogenic sulfur dioxide emissions of 17.37 kt/year (Klimont et al., 2017).These rates are similar to those observed at mid-latitudes, even though only a small area is impacted at the North Slope of Alaska (Figure 1).Localized aerosol gradients from these emissions have been observed by research aircraft (Creamean et al., 2018) and linked to increased l N (Hobbs & Rangno, 1998) and reduced e r (Maahn et al., 2017).In combination with decreased aerosol transport from lower latitudes (Quinn et al., 2009), the role of local emissions in controlling cloud properties could be increased due to higher susceptibility of cleaner clouds (Platnick & Twomey, 1994).In addition to localized aerosol sources, the North Slope of Alaska features flat terrain and limited spatial variability in meteorology: surface measurements of temperature, humidity, and surface pressure at Oliktok Point and Utqiaġvik (formerly Barrow) are correlating at 0.96, 0.95, and 0.97, respectively, according to data recorded at the Department of Energy Atmospheric Radiation Measurement (DOE ARM) sites (Holdridge & Kyrouac, 1993).For Utqiaġvik, Sedlar et al. (2021) found that cloud formation and dissipation is mostly controlled by synoptic events.This results in a favorable setting with limited confounding factors (Grandey & Wang, 2019;Sena et al., 2016) in which to evaluate aerosol-cloud interactions.Leveraging two decades of observations, we demonstrate statistically significant aerosol-based modification of cloud properties over the Prudhoe Bay Oilfield, similar to the case shown in Figure 2 where-in comparison to adjacent areas-more overcast clouds, increased cloud brightness, and decreased cloud e r are seen.

Data and Methods
We analyzed MODIS (Moderate Resolution Imaging Spectroradiometer) swaths collected between 04/2006 and 12/2019 from the Terra and Aqua satellite cloud products (MOD06L2, MYD06L2, collection 6, Platnick & Ackerman, 2015a;Platnick & Ackerman, 2015b).The MODIS products were nearest-neighbor gridded to a 2 km resolution, which greatly expands the data set in comparison to the published MODIS cloud products with daily resolution because multiple daily overpasses are available at high latitudes.To quantify the impact of localized emissions, we selected a Prudhoe Bay region for further analysis that includes almost all oil wells and is slightly extended downwinds (insert Figure 1, orange box) in accordance with the predominant wind direction being from the East (see wind rose in Figure 1).To allow comparisons, we selected a reference region between Utqiaġvik and Prudhoe Bay with similar properties with respect to the number of land pixels, proximity to the ocean, and elevation (insert Figure 1, green box).
To minimize contamination in connection with known deficiencies in the retrieval algorithm, only those 14,755 overpasses were analyzed that feature satellite viewing zenith angles (VZA) less than 56° (Maddux et al., 2010) and solar zenith angles (SZA) less than 65° (Grosvenor & Wood, 2014;Khanal et al., 2020) with the latter limiting the investigated time period to April-September.Also, cloud observations with small optical thickness ( ) are know to be unreliable (Goren et al., 2018;Sourdeval et al., 2015) and only clouds with  greater than five were included in the analysis.Additionally, the MODIS single layer flag is used to remove multi layer clouds to attempt to limit analyzed clouds to those impacted by localized pollution.This analysis focuses on the liquid-containing clouds as identified by the MODIS cloud phase flag.This removes only 4%-7% of the clouds in our data set (see Text S1), because most ice clouds are already removed by the optical thickness threshold.Due to deficiencies of the standard MODIS retrieval for identifying clouds over bright surfaces, the standard retrieval is only applied to pixels without surface snow (based on the National Ice Center's Interactive Multisensor Snow and Ice Mapping System (IMS) Daily Northern Hemisphere Snow and Ice Analysis at 4 km Resolution, Ramsay, 1998;Helfrich et al., 2007;National Ice Center, 2008, updated daily) and a retrieval developed for use over bright surfaces using the 1.6 and 2.1 μm channels (Platnick et al., 2001) is applied otherwise.Our analysis is limited to clouds occurring in pixels over land surfaces only due to the potential impacts of inhomogeneous sea ice surfaces on cloud property retrievals.Pixels containing ocean or lakes were removed to ensure homogeneous surface properties.Also data with surface elevation higher than 100 m above sea level has been removed to minimize interference from orographic lifting processes.
From the MODIS products, cloud liquid droplet effective radius ( e r ) and liquid water path (LWP) were used for further analysis.Cloud frequency of occurrence (CFO) is estimated by the ratio of pixels with identified clouds (i.e.,   5) to the number of total observations after removal of data with too large SZA or VZA as MAAHN ET AL.  ) and the appropriate CFO (Charlson et al., 1992;Meskhidze & Nenes, 2006): here,


SW is derived from monthly clouds and the Earth's Radiant Energy System (CERES) observations on Terra.Monthly values for liquid cloud frequency CFO are estimated using the criteria outlined above except the single layer filter, because the goal is to asses the mean impact on radiation for all cloud conditions.We assume that clouds are adiabatic with vertically constant droplet number concentrations and estimate the where we use monthly pixel mean values for  following Lacis and Hansen (1974).The estimated mean  SW value is only applicable to the six months season covered by the observations (April-September).Assuming that  SW for winter months without observations is 0 W m −2 , the  SW value representative for the full year would be reduced by 50%.

Results
Figure 3a shows how the mean cloud liquid droplet effective radius ( e r ) is reduced in the Prudhoe Bay region in comparison to the reference area.The reduction of e r is not homogeneous in the Prudhoe Bay region and is strongest in the eastern part (up to 1.0 μm).The area with reduced e r expands about 100 km down-wind in accordance with the predominant Easterly wind direction.This is consistent with the location of the largest sulfur dioxide emissions according to the inventory (insert Figure 1).Because surface elevation in the area with reduced e r is below 50 m (insert Figure 1), we assume that the impact of lifting processes on these signatures can be neglected.Characterizing the spatial variability of e r change, the first and fifth percentiles of the 970 pixel in the Prudhoe Bay region are reduced 0.78 μm and 0.60 μm, respectively, but it should be noted that these values depend also on the boundaries used for the Prudhoe Bay region.The relatively small e r reduction is explained by the fact that the data is temporally averaged over all cases and a variety of cloud types and background aerosol concentrations.For individual cases, the e r reduction can be substantially larger as illustrated in Figure 2b.
To account for e r changes of individual cases, we also compared distributions of all MODIS cloud observations within the two regions, that is, temporal averaging is not applied (insert Figure 3a).For the Prudhoe Bay region, we find that the median e r is significantly (5% confidence interval, Mood's t-test) reduced by 0.28 μm in comparison to the reference region.The comparison of the distributions of individual cases reveals that the e r reduction in the Prudhoe Bay region is stronger for the 75th percentile (0.44 μm difference between both regions) than for the 25th percentile (0.21 μm difference).This shows the higher susceptibility of rather pristine clouds with larger e r values with respect to changes related to localized pollution (Platnick & Twomey, 1994).Our results for e r are robust with respect to the chosen filtering and channel combination as shown in Text S2.
When limiting the analysis to times for which winds aligned with the predominant wind direction (60°-110°, 36% of the total cases based on the fifth generation ECMWF atmospheric reanalyzes (ERA5) with 30 km resolution (Copernicus Climate Change Service (C3S) 2017) at the 925 hPa level), the detected temporal mean e r reduction becomes larger (up to 1.35 μm, Figure 3b) and the median of the distribution of all cloud observation is reduced by 0.41 μm.Spatially, the plume extends more than 200 km to 153° W Longitude which corresponds to approximately 8 h of airmass lateral advection considering the ERA5 median 925 hPa wind speed of 7 ms −1 -consistent to the time scales reported by Gryspeerdt et al. (2021) for ship tracks.The plume generally does not extend to Utqiaġvik, even though it has been found that aerosol properties at Utqiaġvik are impacted by localized pollution 8% of the time (Kolesar et al., 2017).
To further support our hypothesis that observed changes of e r are connected to emissions from the Prudhoe Bay region, we leverage MODIS' cloud top pressure (CTP) retrieval to subdivide the data set.This is done with the assumption that local emissions should more directly impact lower clouds than higher clouds, particularly given the stratified nature of the Arctic atmosphere.We assume that lower clouds, those with CTP  750 hPa (Figure 3c), have on average also lower cloud bases making them more susceptible for localized pollution effects.Indeed, low clouds show more pronounced reductions in e r around the Prudhoe Bay area ( e r reduction up to 1.13 μm) than observed for the complete data set (Figure 3a).
To assess whether these changes in cloud microphysics alter cloud lifetime, we evaluated spatial patterns of cloud frequency of occurrence (CFO) from MODIS (Figure 4a).In the eastern part of the Prudhoe Bay region with largest sulfur dioxide emissions, individual pixels shows CFO values increased by 1-2 percentage points in comparison to the remaining Prudhoe Bay region.But it is unclear whether these are random effects, because potential signals related to localized pollution are overlayed by an 8 percentage point meridional CFO gradient that is observed for the whole study region.We are confident that the meridional gradient is not caused by an instrument artifact related to a change of mean VZA within the study area, because we removed data at high VZA and the remaining spatial variability of mean VZA is only 1.5°.Also, mean liquid water path (LWP, Figure 4b) of the identified liquid containing clouds shows a pronounced MAAHN ET AL.LWP difference of 20 g m -2 between the pixels located close to the shore and in the southern parts of the study area.Contrasting to the meridional gradient for CFO, this gradient appears to be perpendicular to the coastline so that mean LWP values are almost identical for both regions.The coastal LWP gradient correlates with the orography (insert Figure 1) so that a connection to lifting processes seems a possible explanation.No LWP patterns are visible within the Prudhoe Bay region that are consistent with the location of the strongest emissions.The strong regional gradients imply that we cannot asses whether there is an impact of localized emissions on CFO or LWP.However, our data allows to define an upper threshold for such effects.If there was a change in LWP and CFO, it must be smaller than the large scale variability observed within the Prudhoe Bay region.This indicates an upper limit for CFO and LWP changes related to localized emissions of approximately 2 percentage points and 5 g , respectively −2 m.
Ultimately, any aerosol-induced changes to cloud micro-and macrophysical properties are important because of their combined impact on the Earth's energy budget.It is estimated that the decrease of e r leads to a mean enhancement of 0.03 W m −2 of  SW for the Prudhoe Bay region during the observation period (April-September).The enhancement is not distributed homogeneously and individual pixels reach 0.79 W m −2 ; the first and fifth percentiles are 0.53 W m −2 and 0.37 W m −2 , respectively (Figure 4c).The distributions of monthly averaged  SW values also indicate the spatial variability with the 75th percentile featuring a larger change than the medians (see insert Figure 4c).The southern parts of the study area show a mean  SW of 0.4 to 0.2 W m −2 which is related to increased e r values in this region (Figure 3).The  SW enhancement in the Prudhoe Bay region reduces radiation arriving at the surface, thereby resulting in a net cooling of the surface environment.However, this relatively small cooling could be further modulated by small changes in CFO below our detection threshold (e.g., 1 percentage points).Such changes would if positive, for all but peak summer months, result in a compensating effect to the aerosol cloud interaction influence on net radiative forcing, taking into account the typical longwave forcing of liquid-containing Arctic clouds, which can be up to 65 W m −2 (Shupe & Intrieri, 2004).The current analysis does not support the presence of any pollution-induced CFO change.This makes it challenging to come to firm conclusions about the ability of local sources of aerosol particles to drive CFO changes, and to determine if such eventual changes would compensate or enhance changes of  SW .

Discussion
To evaluate the impact of anthropogenic aerosol particles on Arctic clouds an extended (14 years) time series of MODIS observations consisting of 14,755 overpasses during the polar day was analyzed.To our knowledge, the current study represents the first attempt to constrain the impact of anthropogenic emissions on clouds existing in the clean Arctic background state using "natural laboratory" observations over a long time scale.While the mean observed signals are relatively small, the long observational period allowed for identification of the following patterns for the observational period (April-September): 1.For liquid and mixed-phase clouds, liquid effective radius e r is reduced in the Prudhoe Bay by up to 1.0 μm when averaging over the full data set (Figure 3).When limiting the data to easterly winds, the main wind direction, the reduction increases to up to 1.35 μm.For individual cases, the reduction in e r can be larger (Figure 2) 2. For cloud frequency of occurrence (CFO) and liquid water path (LWP), pronounced regional gradients in CFO and LWP overlay potential local effects centered on the Prudhoe Bay area (Figures 4a and 4b).
If there were such effects, they must be smaller than approximately 2 percentage points and 5 g −2 m for CFO and LWP, respectively 3. In combination, these anthropogenic aerosol impacts on cloud properties are estimated to result in a mean increase of upwelling shortwave radiation (  SW ) of up to 0.79 W m −2 for the April-September period (Figure 4c) Despite the challenging conditions for MODIS retrievals for land pixels at high latitudes due to low solar zenith angles and bright surfaces (Grosvenor & Wood, 2014;King et al., 2004;Platnick et al., 2001), we found our results to be robust with respect to the applied filters and retrieval methods.Our findings are consistent with the theory of aerosol cloud interaction that enhanced CCN concentrations modify cloud properties.Reduced cloud droplet sizes associated with elevated CCN concentrations are notable in the MODIS observations, but the question as to whether there is a small change in CFO or LWP related to localized pollution could not be answered with the available data and is left open for future studies.Answering this question is crucial to fully quantify the impact of localized pollution on the radiative budget.The found changes in e r and  SW are small in comparison to those often observed for ship tracks (e.g., up to 4.1 μm according to Christensen & Stephens, 2011) but localized emissions do-unlike ships-not move so that even small e r can be locally highly relevant over time.In general, the exact value of the cloud response depends strongly on the methodological approach.This includes the question whether the polluted plume boundaries are well identified and compared with its pristine surroundings (Christensen & Stephens, 2011;Goren & Rosenfeld, 2014), and whether a climatological mean is estimated (Christensen & Stephens, 2011;Diamond et al., 2020) rather than an instantaneous effects (Goren & Rosenfeld, 2012, 2014).For instance, a case study approach focusing on well defined ship tracks showed a negative cloud radiative effect (CRE) of 4 W m −2 for ship tracks embedded within closed cells (Goren & Rosenfeld, 2014).In contrast, Diamond et al. (2020) provides a climatological estimate of 2 W m −2 due to cloud brightening within the southeast Atlantic shipping corridor.
It is important to note that the current approach likely underestimates the total impact on cloud radiative forcing because (a) optically thin clouds are not considered due to instrument limitations and (b) we use a surface estimate for  SW from CERES which is slightly more attenuated than the relevant  SW at cloud top (McCoy & Hartmann, 2015).Our  SW estimate is based only on e r change and does not account for potential LWP adjustments.A full assessment of the radiative impact is important to understanding the net impacts of industrial activities on the surface energy budget, ice and snow formation and melt, the subsequent controls imparted on terrestrial ecosystems, and the associated feedback mechanisms.Our  SW estimate is only applicable to the period covered by observations (April-September) and needs to be reduced by approximately 50% to be representative for a full year.
This study gives a first estimate on the types of cloud perturbations that might be possible in the future in other industrialized regions of the Arctic.In a future warmer, more easily accessible Arctic, industrial activities are expected to increase which is potentially leading to rising local-source aerosol concentrations.Assuming that industrial emissions related to oil extraction produce higher emission than other anthropogenic activities in the Arctic, this study provides an upper boundary for microphysical and radiative effects related to localized pollution.Additional work is required to fully understand the impact of localized pollution on the ice phase of liquid containing clouds.This includes developing methods to reduce the uncertainties of space-based cloud retrievals for optically thin clouds and analyzing data of the ground-based observations of the Department of Energy Atmospheric Radiation Measurement (DOE ARM) sites in Alaska.Oceanic and Atmospheric Administration (NOAA) Physical Sciences Laboratory (PSL).We gratefully acknowledge the funding by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) for the "Arctic Amplification: Climate Relevant Atmospheric and Surface Processes, and Feedback Mechanisms" (AC)

Figure 1 .
Figure 1.Yearly sulfur dioxide emissions based on an inventory by Klimont et al. (2017) for the northern hemisphere and the study region (insert).The insert shows also isolines for height above sea level (gray) and a wind rose for the 925 hPa level based on ERA5.The green dots correspond to oil wells active in March 2017, the green and orange boxes indicate the the reference region and the Prudhoe Bay region, respectively.

er
and spatially averaged monthly mean values for ,reference e r based on the reference region. C is estimated from monthly averaged MODIS  with

Figure 2 .
Figure 2. Case study showing the impact of local emissions on cloud properties.(a) Terra Moderate Resolution Imaging Spectroradiometer (MODIS) true color image and (b) MODIS liquid cloud effective radius ( e r ) retrieval on 2016-08-29 at 22:15:10 UTC with clear areas indicating areas without liquid clouds.Black lines show shorelines and dark green dots mark oil wells.The case shows a change in cloud brightens and e r downwind of the eastern part of the Prudhoe Bay region where the emission inventory reports the highest sulfur dioxide emissions (Figure 1).For this case, ERA5 925 hPa wind direction at Oliktok Point was 95 o .

Figure 3 .
Figure 3. (a) Spatial deviation of Moderate Resolution Imaging Spectroradiometer (MODIS) mean liquid cloud effective radius e r in comparison to a reference region (green quadrangle).The MODIS standard retrieval is used during periods without snow on the ground while the 1.6/2.1 μm retrieval is used for periods with snow cover.The normalized e r distributions for all observations in the Prudhoe Bay region (orange quadrangle) and the reference region are shown in the insert with the lines representing the median as well as the 25th and 75th percentiles.The gray circles indicate (from left to right) Utqiaġvik, Oliktok Point, and Deadhorse; black lines show shorelines.(b) as (a), but limited to the main wind direction defined as 60°-110° from ERA5 at 925 hPa at Oliktok Point.(c) as (a), but limited to clouds with MODIS' cloud top pressure larger than 750 hPa.The medians of the distributions of the two regions are significantly different (5% confidence interval) for all three data sets.

Figure 4 .
Figure 4. Spatial patterns of Moderate Resolution Imaging Spectroradiometer (MODIS) products showing (a) liquid containing cloud frequency of occurrence (CFO), (b) mean liquid water path (LWP), and (c) change in shortwave upwelling radiation  SW with respect to the reference region (green box).The embedded violin plots show the underlying normalized distributions for the study (orange) and reference (green) region with the lines representing the median as well as the 25th and 75th percentiles.Note that for CFO and  SW , the distribution of monthly values is shown and the full (i.e., non averaged) data set is shown for LWP.Median values colored in black indicate a significant difference (5% confidence interval) between the medians of the two regions.The gray circles indicate (from left to right) Utqiaġvik, Oliktok Point, and Deadhorse; black lines show shorelines.
3 Project 268020496-TRR 172 within the Transregional Collaborative Research Center.T. G. acknowledges funding by the European Union via its Horizon 2020 project FORCeS (GA 821205) and by the DFG project CDNC4ACI (GZ QU 311/27-1).This study contains modified Copernicus Climate Change Service information (Copernicus Climate Change Service (C3S), 2017).Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus Information or Data it contains.Open access funding enabled and organized by Projekt DEAL.