An assessment of the spatial extent of polar dust using satellite thermal data

The effect of mineral dust aerosols and subsequent deposition in polar regions has historically been underestimated in climate models. Dust aerosols increase melt rates by reducing surface albedo and modifying atmospheric radiative properties. In this study 127,413 satellite images covering the Arctic and Antarctic from 2007 to 2019 were assessed for dust content using thermal infrared wavelengths. The results show a strong linear trend in which the relative spatial extent of dust (RSED) increased annually by 0.31% in the Arctic (8.5% to 12.1%) and 0.19% in the Antarctic (5.2% to 7.5%). Seasonally, the maximum aggregated average RSED occurred in the Arctic during boreal winter (11.2%), while the Antarctic peaked in austral spring (9.5%). Maximum RSED rates occurred in boreal winter/austral summer (Dec–Jan–Feb) for both polar regions. The data suggests that finer dust particles are more efficiently distributed by aeolian processes leading to higher RSED values that are not necessarily reflective of polar dust loading models.

The temporal distribution, physical extent, and environmental effect of mineral dust at high latitudes is an underrepresented facet of polar climate science and the impact of dust deposition on cryospheric processes is not well understood 10,28 . This research adds to the growing body of knowledge on polar dust loading by measuring the relative spatial extent of dust (RSED) in the Arctic and Antarctic from 2007 to 2019 using satellite thermal infrared (TIR) imagery. The 12-year record of RSED establishes annual, seasonal and monthly trends in areal dust distribution at the poles.

Methods
The detection of dust using satellite thermal data is well established in the scientific community [29][30][31][32][33][34] . Numerous studies have shown that dust may be detected with a spaceborne radiometer by comparing bands in the TIR regime. The selective emissivity of snow/ice, water, vegetation, and sand allows these materials to be identified by measuring the brightness temperature difference between 11 μm and 12 μm (BTD [11][12] ). BTD [11][12] is highly positive for snow/ice, slightly positive for water, slightly negative for vegetation and highly negative for sand or dust 29,35 . A problem with identifying dust with BTD [11][12] is discriminating it from the contribution of various surface types 30 . This issue is mitigated in the polar regions where ice/snow and water predominate. Arctic tundra has been noted to have a highly positive BTD [11][12] signature 30 , which means that both poles generally exhibit a positive surface BTD [11][12] value throughout the year. The predominantly positive BTD [11][12] signature at high latitudes allows the identification of dust with Advanced Very High Resolution Radiometry (AVHRR) satellite data by applying a threshold of BTD 11-12 < 0 K to each pixel in a polar scene 30 .
This study uses AVHRR imagery from the MetOp-A satellite with a nadir resolution of 1.1 km and swath width of approximately 2900 km. Data files were obtained on-line from the National Oceanic and Atmospheric Comprehensive Large Array-data Stewardship System. Data ranged from 01 June 2007 to 31 May 2019 and was constrained from 66° N to 90° N latitude for the Arctic and 66° S to 90° S latitude for the Antarctic. For the purposes of this research these latitude ranges are referred to as the Arctic and Antarctic regions. The polar orbit of the satellite allowed for 14 to 15 passes per day over the study areas, resulting in 127,996 images representing 2.24 × 10 11 pixels. Table 1 shows specifications of the MetOp-A AVHRR sensor, while Table 2 lists statistics of analyzed data for each year of the study. Figure 1 illustrates the area coverage of a typical day for both poles. www.nature.com/scientificreports/ For this research, an algorithm was developed to perform bulk thermal calibration of Channel 4 (11 μm) and Channel 5 (12 μm) for each AVHRR file. Calibration procedures were in accordance with EUMETSAT Metop-A AVHRR specifications and the results compared against existing commercial software (L3Harris ENVI) for validity. BTD [11][12] was determined for each pixel in a calibrated image and a threshold applied to determine the percentage of pixels with BTD 11-12 < 0. The processed files were sorted by polar region and date, then sanitized for anomalies such as 100% dust coverage or extreme BTD [11][12] values not supported by physics. A total of 583 files were removed (0.46%), resulting in a total of 127,413 images for the final analysis. The Python code for the algorithm used in the study is available on-line at https ://githu b.com/matt-bowen /PyFRA C. The MetOp-A AVHRR is a stable sensor with calibration errors in the order of 0.3% 36 . Pixels close to the threshold may potentially be misclassified as dust or non-dust because of calibration errors, but statistically these should even out for the dataset.
In the polar environment over areas of snow/ice, water, and tundra where BTD 11-12 > 0, suspended and deposited dust particulates return a distinctive negative value. Considering the underlying landscape, positive BTD [11][12] values are still possible in the presence of dust since the pixel value is a summation of everything within the instantaneous field of view of the sensor. Space-based TIR radiometers such as AVHRR measure the skin temperature of an object, so only a thin layer of dust is necessary to produce a negative BTD 11-12 signature. As such, the processed data reflects a dust concentration that is dense enough to dominate a pixel value. While it is possible that a substance other than mineral dust is causing large-scale negative BTD [11][12] signatures in the polar environment, there is nothing in the literature to support this conjecture 30 . Thus, the method used in this study allows the detection of airborne and deposited dust in the Arctic and Antarctic with high confidence since the quantity of dust within a pixel must be sufficient to overcome the positive BTD 11-12 contribution of the surface.

Results
The RSED is defined as the percentage of dust pixels (BTD 11-12 < 0) in relation to the total number of pixels in a thermally calibrated AVHRR image. The yearly aggregated RSED showed strong linear trends for both poles between 2007 and 2019 (Fig. 2). In the Arctic, RSED increased from 8.5% to 12.1% or 0.31% per year, while in the Antarctic there was a smaller increase of 0.19% per year, ranging from 5.2 to 7.5%. The lower value in the southern polar region is reflective of decreased dust sources in that hemisphere.
Seasonal aggregated values for RSED were calculated for the study period. The boreal winter/austral summer is Dec-Jan-Feb followed by Mar-Apr-May (boreal spring/austral fall), Jun-Jul-Aug (boreal summer/austral winter), and Sep-Oct-Nov (boreal fall/austral spring). The dataset for this study begins in June 2007, which was the first available MetOp-A AVHHR imagery. The final season is Mar-Apr-May 2019, giving 12 years of seasonal data. Figure 3 shows the seasonal trends for both polar regions. The largest RSED rate increase occurred in Dec-Jan-Feb for both the Arctic (0.65% per year) and Antarctic (0.18% per year). The Arctic showed no change during the study period in boreal summer (Jun-Jul-Aug), while the Antarctic experienced minimal increase during the austral spring (Sep-Oct-Nov) and fall (Mar-Apr-May). The RSED rate increase for both poles were similar for Mar-Apr-May and features the only season when the average RSED of the Antarctic exceeded that   www.nature.com/scientificreports/ of the Arctic. The average RSED for all seasons is markedly more stable for the southern pole region. Table 3 shows statistics for the seasonal RSED. The mean value of BTD 11−12 for all dust pixels was determined for each image. The overall average BTD 11−12 was − 0.16 K for the Arctic and − 0.22 K for the Antarctic. Both regions showed consistent yearly averages with a standard deviation of 0.01 K for both poles. A seasonal analysis of BTD [11][12] shows more variability (Fig. 4). There is a clear distinction between seasons in the Antarctic, with the most negative values occurring in Mar-Apr-May (− 0.28 K) and the least negative in Dec-Jan-Feb (− 0.19 K). The Arctic exhibits a smaller range of BTD [11][12] values for most of the year (− 0.13 K to − 0.16 K) except for Dec-Jan-Feb when it reaches − 0.21 K. This is the only season when the average Arctic BTD [11][12] is less than that of the Antarctic.
Finally, the results were aggregated monthly from June 2007 to May 2019. On a monthly basis the dataset becomes highly variable compared to the yearly and seasonal averages. When the two poles are compared, a pattern emerges with respect to RSED and BTD [11][12] where peaks and troughs for both parameters are generally opposite between the two regions (Fig. 5). This overall view of the data is not surprising given the antipodal nature of the study areas.

Discussion
Annual trends. The principle result in this study indicates a yearly increase in average RSED rates at both poles (Fig. 2). This trend is likely a result of several factors, such as increased global dust generation, enhanced dust transport mechanisms and accelerating local dust production at high latitudes. Overall, the global amount of dryland has increased between 1948 and 2008, with this expansion expected to continue to the end of the twenty-first century 37 . Large-scale aridification can be attributed to many anthropogenic and natural variables 38 . The average aerosol optical depth (AOD) is a proxy for dust aerosol concentration and can be an indicator of transport mechanism strength. From 1998 to 2010, the AOD over the global oceans increased slightly, while stronger positive trends were associated with seasonal cycles 39 . Glacial activity is a contributor to local high latitude dust, as several processes surrounding glaciers are highly efficient at dust generation and emission. For example, strong winds are produced by gravity and thermal gradients associated with glaciers and the retreat of ice masses expose fine sediments for entrainment 40 . Dust produced by glaciers is expected to increase in the coming decades as glaciers throughout the cryosphere retreat 40 . In the Canadian Arctic Archipelago, nearly every glacier has shrunk since 1958, with region-wide retreat rates accelerating by a factor of five between 2000 and 2015 41 . The West Antarctic glaciers terminating in Pine Island Bay are also retreating rapidly. Smith and Kohler glaciers retreated more than 30 km between 1992 and 2011, with other glaciers in the area retreating between 9 and 14 km during the same time frame 42 .
The RSED in this research does not quantify the amount of dust transported to the poles but shows more widespread distribution since 2007. Modeling of polar dust loading indicates that 16 times more dust is transported to the Arctic than the Antarctic 16 , but the overall average RSED for the Arctic (10.4%, σ = 1.4% ) is only 4.1% greater than that of the Antarctic (6.3%, σ = 0.9%). Assuming that polar dust loading models are accurate, this implies that dust is spread more efficiently in the Antarctic. Dust transport mechanisms from global sources work on time scales of five days towards the Arctic 43 and seven to ten days towards the Antarctic 44 . Finer particles are indicative of more distal sources 45 , which supports enhanced distribution of dust at the southern pole.

Seasonal variations.
Simulations indicate that dust loading in the Arctic peaks during boreal spring because of strengthening dust transport mechanisms from global sources 18 . In this study, Mar-Apr-May produced the lowest average Arctic RSED (9.2%, σ = 2.0%) with an increase of 0.41% per year. The maximum average Arctic RSED occurred during Dec-Jan-Feb (11.2%, σ = 2.5%) with the highest annual increase (0.62% per year) of any season. Total aerosol concentration in the Antarctic peaks in Dec-Jan-Feb 46,47 , which is concurrent with the highest increase in RSED rate (0.18% per year) but features the lowest observed average RSED (4.8%, σ = 0.8%). The results highlight the difference between total dust transport and spatial distribution. Finer parti- Table 3. Statistics for the seasonal RSED from 2007 to 2019 for each polar region, including average RSED and standard deviation (σ) as well as the linear trend and related R 2 , P-Value and Standard Error of the Mean (SEM).

Season 2007 to 2019
Average RSED (%) σ (%) www.nature.com/scientificreports/ cles spread more readily and may result in a higher RSED that is not a function of absolute dust quantity. This explains why RSED in this study does not necessarily align with established seasonal dust loading patterns.

BTD 11-12 variation.
There is a notable contrast in the average value of BTD 11−12 between the Arctic and Antarctic, with the southern polar region reporting a more negative signature except for Dec-Jan-Feb (Fig. 4). BTD 11-12 generally decreases with decreasing particle size 48 . Larger dust particles are preferentially removed from atmospheric transport mechanisms over large distances 49,50 , suggesting that Arctic dust particles are potentially larger than the southern counterpart due to the more proximal sources in the northern hemisphere. Another factor affecting BTD 11-12 is dust composition. Mineral dust produces a negative BTD 11−12 value because of the emissive properties of common silicate minerals present in desert sand 51 . Saharan and East Asian dust consist of 18.9% and 23.2% silicon respectively, while the primary Antarctic contributor of dust in Patagonia is 28.8% silicon with Australian dust at 18.5% silicon 52 . Since Patagonian and Australian dust affect different geographical areas of Antarctica, a detailed mapping of BTD [11][12] values could help evaluate the relative importance of silicate concentration on BTD 11-12 signatures.
A third influence on BTD [11][12] signatures is the contribution of the underlying surface. The Antarctic continent and surrounding Southern Ocean consist almost entirely of water and ice throughout the year whereas the Arctic has a more varied surface, particularly in the warmer months. While the Antarctic surface makeup is relatively constant throughout the year with approximately 70% snow/ice and 30% water, the Arctic experiences seasonal fluctuations in the amount of ice, water, tundra, and bare rock. As such, there is more variation in BTD [11][12] in the Arctic than the Antarctic. Interestingly, the most negative Arctic BTD [11][12] occurs in Dec-Jan-Feb when the surface is almost entirely covered with snow/ice that has a highly positive BTD [11][12] . This implies that finer dust particles from more distal sources are present in the Arctic during boreal winter.  www.nature.com/scientificreports/ Monthly variations. The monthly average RSED (Fig. 5a) shows a series of peaks and troughs for both poles, with the Arctic demonstrating greater variability (2.3% to 21.4%) than the Antarctic (2.1% to 12.5%). Long range transport for Arctic dust source regions such as the Taklamakan Desert vary on monthly scales 53 , while significant intermonth variations of dust deposition have been observed at Antarctic sites 23 . The RSED normally peaks around January in the Arctic and May in the Antarctic, while the troughs in each region generally correspond to the opposite's peak. The greatest RSED in the Arctic occurred in February 2016 which coincided with the highest recorded value of the Oceanic Niño Index (ONI), suggesting a teleconnection with the El Niño-Southern Oscillation. Overall correlation between monthly Arctic RSED and ONI was 0.14 and slightly stronger for the North Atlantic Oscillation (0.18). The Antarctic showed no correlation with either climatic phenomenon, however, it is notable that the lowest average BTD 11-12 occurred in March 2016 during the period of record high ONI. Monthly BTD [11][12] (Fig. 5b) show greater monthly variation in the Antarctic (− 0.38 K to − 0.12 K) than the Arctic (− 0.26 K to − 0.09 K). This may be attributable to finer dust in the Antarctic causing low BTD [11][12] values that is offset periodically by reduced dust concentration. A strong negative correlation (− 0.67) was revealed between BTD [11][12] and RSED in the Arctic with a moderate negative correlation (− 0.30) in the Antarctic. The negative correlation indicates that there is greater areal spreading of dust as BTD [11][12] decreases, implying that finer dust particles are spread more readily. This relationship is pronounced in the Arctic winter where the highest seasonal RSED and lowest seasonal BTD [11][12] both occur in Dec-Jan-Feb, which coincides with the strongest boreal surface winds 54 . Significance of study. This research offers unique insight on recent trends in polar dust extent using satellite thermal infrared imagery. Whereas the results show an increase in the spatial extent of dust at both poles, there are variations between the two areas that suggest different mechanisms of dispersal and distinct dust characteristics. Modelled seasonal dust loading peaks do not generally align with the maximum spreading of dust, which may have significant impact on climate models with respect to the darkening of polar surfaces and the influence on cloud radiative properties. The next step in this research is to map areas of Arctic and Antarctic dust influx.

Data availability
Satellite Advanced Very High Resolution (AVHRR) data is available on-line at the National Oceanic and Atmospheric Administration (NOAA) Comprehensive Large Array-data Stewardship System (CLASS) (https ://www. bou.class .noaa.gov/saa/produ cts/welco me).