To provide a background for characterizing the seasonality of AARs, we first illustrate the seasonal variability in the overall aerosol transport (i.e., regardless of AAR conditions) during 1997–2014 (Fig. 1). Here, we combine black carbon (BC) and organic carbon (OC) aerosols, denoted as CA, owing to their similar sources and seasonality (BC and OC are accounted for separately in subsequent analysis of AARs). Dust IAT (first column; Fig. 1) is higher during the MAM and JJA seasons. Global deserts, such as the Sahara, Gobi, and Taklamakan deserts, and the Middle East act as a significant source of dust aerosols year round (Fig. 2a), emitting more than 2×10-9 kg m-2 s-1 of dust, primarily in the northern spring. The sea salt IAT (second column; Fig. 1) increases during the winter seasons of the Northern and Southern hemispheres due to the increased mean westerly flow and storm activities during these seasons. Annual maps of SS emission (Fig. 2b) show that a large number of SS aerosols are emitted over the global oceans, especially over the tropical oceans because of the convective activities, and in the midlatitudes where synoptic storm activities are dominant. Sulfate IAT (SU; third column; Fig. 1) is high over a region between China and the North Pacific Ocean region, extending up to the western USA, during all seasons with the lowest SU IAT values in JJA. The large SU IAT values (Fig. 2c) are likely due to high emissions of SO2 over China (Dai et al., 2019; Wang et al., 2013) coupled with strongly varying synoptic flows. A secondary region of a higher amount of SU IAT is also detected between the eastern USA and the northwestern Atlantic Ocean, as a high amount of SO2 is emitted east of the Rocky Mountains, particularly over the Ohio valley (Fig. 2c). Figure 2c also shows that some SO2 is emitted due to global shipping activities.

Globally, boreal and rainforests are the most significant contributors to the CA aerosols (Fig. 2d). Accordingly, it appears that the Congo rainforest dominates the Amazon rainforest in terms of CA IAT (right column; Fig. 1). The Amazon rainforest region has higher IAT during its dry and transition seasons and lower IAT in the wet season. Similarly, the Congo rainforest releases a high amount of CA aerosols during the dry season (JJA) and another peak in the Sahel during DJF (Fig. 2d). An increase in dry period CA IAT might be due to the increased vegetative stress, forest fire, and agricultural burning over these two regions. However, it is interesting to note that the Congo rainforest also emits CA aerosols during the boreal autumn (SON) rainy season, which is the stronger of the two rainfall seasons in this region, but not during the MAM rainy season. In MAM, the tropical rainforests over eastern India, Myanmar, and southern China have higher CA IAT values. Emissions from the boreal forests appear to be less than that of the rainforests. Over most of the midlatitudes in the Northern Hemisphere, the emissions of CA aerosols (Fig. 2d; also IAT values in Fig. 1d) are less than that over the global rainforests and China in all four seasons (Fig. 1). As a result, although CA (OC and BC) AARs are more frequent over the midlatitude region, and the mean IAT by those rivers is less than those AARs that originate from the global rainforests and China (Chakraborty et al., 2021a). These results point out the existence of seasonality and variability in the overall aerosol emission and transport. As a result, we expect the frequency and intensity of AARs might also vary among different seasons in accordance with the distribution of surface emissions.

To illustrate the characteristics of individual AAR objects, Fig. 3 shows examples of AARs of each species, including the location, shape, and transport direction and magnitude. Also indicated in the figure is the length and width of each illustrated AAR. Figure 3a shows all the DU AARs detected on 25 June 2008 at 12:00 UTC. Many of those DU AARs are detected between the Sahara desert and the Caribbean, the Middle East region and Europe, over the central USA, and over the Patagonia region. Figure 3d shows details of one DU AAR (encircled in Fig. 3a) that extends from the western boundary of the Sahara desert to the southern USA and Caribbean and has IAT values greater than 12×10-3 kg m-1 s-1. The AAR is 8082 km long and 1158 km wide. Similarly, we show details of one SS AAR (Fig. 3e) with a length of 5295 km and a width of 984 km (IAT ∼1.0×10-3 kg m-1 s-1) over the Southern Ocean among other AARs detected on 19 January 2010 at 12:00 UTC (Fig. 2b). Unlike DU AARs, SS AARs are located mostly over the ocean (see also Fig. 4) and carry the extratropical AR signature. SS AARs over the tropical regions appear to be smaller than the extratropical AARs in Fig. 3b.

A number of SU AARs occurred in the polar region due to the Eyjafjallajökull volcanic eruption (Fig. 3c). The volcano, located in Iceland, erupted on 20 March 2010, causing disruption to the aviation industry (https://volcanoes.usgs.gov/volcanic_ash/ash_clouds_air_routes_eyjafjallajokull.html, last access: 1 January 2022). In Fig. 3f, we show details of one of these SU AARs on 29 March 2010. The SU AAR is 6203 km long and 294 km wide and transported a large amount of SU aerosols (IAT ∼0.1×10-3 kg m-1 s-1) into and across the polar region. It is important to notice that MERRA-2 did not include the ash aerosols that were co-emitted with the SO2 plumes that were eventually converted into sulfate aerosols by gaseous and aqueous processes. In April 2010, our algorithm detected ∼80 SU AARs (at every 6 h of interval, i.e., 20 SU AAR days) originating over that region and propagating to different directions.

Hereinafter, we separately show the CA rivers as OC and BC rivers, especially because BC AARs can significantly impact the radiative forcing compared to OC AARs and because the mean mass of aerosols being transported by the OC is about 5 times that of BC AARs (Fig. 4). Figure 3g shows examples of OC AARs detected on 1 July 2013. We show details of one OC AAR that stretches from Madagascar to New Zealand in Fig. 3i. The OC AAR is 13078 km long and 730 km wide and flows west over the Indian Ocean with an IAT of ∼0.1×10-3 kg m-1 s-1. Figure 3h shows a few BC AARs detected on 27 April 2009. BC AARs often originate near the China–northwestern Pacific Ocean region (see also Fig. 4). We show the details of one such BC AAR in Fig. 3j. It is to be noted that the IAT values of BC AARs are smaller than the OC AARs; however, they might play a significant role in the atmospheric radiation budget owing to the capability of BC aerosols to absorb solar energy.

To illustrate the overall climatology of AARs, Fig. 4 shows the annual mean frequency of occurrence (shading; days per year) and the mean IAT (arrows; kg m-1 s-1) associated with AARs. Figure 4a shows that a strong anticyclonic motion over the Sahara desert is associated with many DU AARs, consistent with the annual DU emission (Fig. 2a) and seasonal DU IAT (Fig. 1 – leftmost column) over that region. Around 30 AAR d yr-1 (shades) carry on average 3–15 ×10-3 kg m-1 s-1 (vectors) of dust from the Sahara desert over the North Atlantic Ocean and reach the southern USA and the Caribbean regions. A similar number of AARs also transport aerosols towards Europe and the middle east region. A relatively high number of DU AARs are detected over China, Mongolia, and Kazakhstan; however, the mean transport over these regions by AARs is lower (IAT ∼1×10-3 kg m-1 s-1) than those originating from the Sahara desert. The southern hemispheric deserts also emit numerous AARs but with a smaller amount of dust transport as the dust emissions (Fig. 2a) and IAT (Fig. 1) are low over there. Overall, the Sahara desert dominates any other desert in the world regarding the formation of the strongest and most intense DU AARs.

Figure 4b shows the climatological maps of SS AARs. SS AARs are mostly located over the global oceans, especially over the tropical and subtropical trade wind regions. The SS AARs in the midlatitudes carry the signature of the storm tracks and have distributions similar to ARs (Guan and Waliser, 2015). Every year, in the midlatitude region, 30–40 ARs are detected (Chakraborty et al., 2021a), whereas ∼20 AAR d yr-1 occur in the midlatitudes with mean a IAT of ∼2×10-3 kg m-1 s-1. The distributions between ARs and SS AARs are not quite the same; the SS AARs are biased equatorward toward the trade winds. In comparison, tropical SS AARs are more frequent but have less IAT than the midlatitude SS AARs. Around 30 SS AARs have mean IAT of ∼1×10-3 kg m-1 s-1 over the tropical region.

It is important to note SU AARs are more frequent (∼40 d yr-1; Fig. 4c) in the Northern Hemisphere than in the Southern Hemisphere, consistent with the predominance of emissions of SO2 over China, Europe, and the eastern USA (Fig. 2c), owing to anthropogenic emissions and biogenic activities in these regions. It is important to note that MERRA-2 aerosols data do not account for biogenic sources like carbonyl sulfide in the version of the data we have used for this study. The SU AAR hotspot regions over the Southern Hemisphere include pathways from the southern edges of the global rainforests to the south Indian, south Atlantic, and Southern Oceans. Mean IAT (∼0.2×10-3 kg m-1 s-1) by SU AARs is lower than that of the DU and SS AARs. Many BC and OC AARs originate from the global rainforests, such as the Congo and Amazon, and from the regions that are susceptible to biomass burning and CA aerosol emissions, such as Europe, the eastern USA, industrialized areas over eastern China, and northern India. Annually, 20–40 BC or OC AARs are generated from these regions. It appears that BC AARs are more numerous than the OC AARs; however, OC AARs have larger IAT than BC AARs.

To understand the anomalous atmospheric conditions that account for an AAR event, it is important to know the relative contribution of the two quantities that make up these IAT extremes, namely the aerosol mixing ratio and the wind speed. Also, as mentioned before, the altitude of aerosol particles within AARs can be of importance due to aerosol impacts on the radiation budget, convective anvil lifetime (Bister and Kulmala, 2011), cloud formation (Froyd et al., 2009; Khain et al., 2008), and air quality near the surface. Here, we characterize the vertical profiles of the AARs at a number of different locations with high AAR frequency (see the inset map in the middle column; based on Fig. 4). For example, the orange box over the Saharan–Caribbean pathway experiences a presence of 25–30 AARs yr-1 (Fig. 4a). Based on our analysis between 1997 and 2014, the profiles in Fig. 5a, b, and c show the mean and standard deviation of ∼ 400–500 AARs.

It is important to keep in mind that the aerosol vertical structure in MERRA-2 is not directly constrained by measurements and is chiefly determined by the injection height of the emissions and turbulent and convective transport processes parameterized in the model. Evaluation of the vertical structure of MERRA-2 aerosols appears in Buchard et al. (2017). Buchard et al. (2017) has used Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) to examine the vertical structure of aerosols in MERRA-2. The left column of Fig. 5 shows the composite vertical profiles of aerosol mass fluxes for each aerosol species; both those for AARs (solid) and for all time steps regardless of AAR conditions (dotted) are shown for comparison. Mean fluxes of the synoptic AAR events are larger than the overall annual mean over all the regions, which is not unexpected since AARs represent the extreme transport events. For DU AARs (Fig. 5a), the strongest transport is observed between the Sahara desert and North Atlantic/Caribbean (orange) and western Europe/Middle East (red) pathways. Flux values decrease with altitude within AARs that transport aerosols from the Sahara desert to the Europe/Middle East region (red). However, AARs that transport aerosols from the Sahara desert to the North Atlantic region (orange) have peak IAT values near 750 hPa. There is also a secondary peak at the 960 hPa height (Fig. 5a). The aerosol mixing ratio (Fig. 5b) slightly decreases from the surface up to a height of 960 hPa and then increases above it. This might suggest the influence of the gravitational force and settlement on the aerosol mixing ratio, as observed in many of the aerosol mixing ratio profiles of other species over other regions (for example, the red line showing the aerosol mixing ratio profile for the Sahara to European pathway). However, a smaller peak in the wind profile (Fig. 5c) around 950 hPa indicates the influences from the low-level jets and the seasonal variations in the transatlantic dust transport that have not been studied yet. A detailed seasonal analysis is required to understand the existence of the smaller peak at 960 hPa. However, the presence of the African easterly jet–north (AEJ-N) above, centering at 600–700 hPa, further lifts the aerosols and attains a peak around 750 hPa.

We also show the vertical aerosol mixing ratio profiles (middle column) and the wind profiles (right column) separately to help disentangle their influences on the aerosol flux profiles. Given the very few globally gridded datasets to use for this purpose, we have relied on the data provided in the MERRA-2 data, which in this case is provided as aerosol mixing ratio. Figure 5b shows that high aerosol levels inside AARs over the Sahara desert to Caribbean pathway extend vertically up to the 700 hPa height, with a peak around 750 hPa, unlike any other regions analyzed. The third column, which shows the mean wind speed profile U2+V22 multiplied by the sign of the zonal wind to show the east–west direction of the AARs, confirms the influence from the AEJ-N (Cook, 1999; Wu et al., 2009) on the AARs wind profile that peaks around 650 hPa and lifts aerosols over this region (Fig. 4c). Higher flux values near the surface are due to a high aerosol mixing ratio, but as altitude increases, higher flux values inside AARs are contributed by both the aerosol lifting and the high zonal wind speed of AEJ-N between the Sahara desert and the North Atlantic Ocean region. Other regions like the Southern Ocean (near South America), Australia, eastern China, and western USA have a smaller aerosol mixing ratio as compared to the Sahara to the North Atlantic Ocean and Caribbean and Sahara to Europe/Middle East pathways. Wind speed contributes to larger flux values as altitude increases over some of these locations near eastern China, the Southern Ocean, Australia, and the western USA; however, their flux values are smaller than those taking off the Sahara desert.

Aerosol flux and mass mixing ratio profiles for SS species (Fig. 5d and e) decrease with altitude. This is because the SS particles are typically found within the boundary layer (Gross and Baklanov, 2007), are larger in size, and often form due to high wind speed and surface evaporation along the storm tracks and may not travel long distances outside the storm tracks (Sofiev et al., 2011; May et al., 2016). The largest aerosol mixing ratio and IAT values are observed over the Southern Ocean consistent with the persistent and year-round AAR activities over there.

Consistent with the SO2 emission (Fig. 2c), eastern China (blue line) dominates in terms of SU aerosol mass mixing ratio by a factor of 5 as compared to the other regions (Fig. 5h) analyzed and shown here. The aerosol mixing ratio (Fig. 5h) and the flux values (Fig. 5g) increase with height, attaining a peak around 900 hPa over there, and probably exhibit the impact of the boundary layer inversion effect on the accumulation of the pollutants (Li et al., 2017). Above 900 hPa, both the aerosol mass mixing ratio and flux values gradually decrease. However, the SU aerosol mass mixing ratio decreases when those AARs reach the western USA five times (Fig. 5h). Owing to the fact that the eastern USA originates many SU AARs that can contribute to a transatlantic transport driven by strong westerly winds, we have also computed the IAT profiles of SU AARs generated over there. Figure 5g and h show that AARs over the eastern USA also have a high IAT (Fig. 5g) and aerosol mass mixing ratio (Fig. 5h). Flux profiles and aerosols mixing ratio over the Indian Ocean (red) and Southern Ocean (green) show different behavior. Aerosols appear to be continental in origin and are lifted to attain a peak concentration around 650 hPa. Contributions from wind speed (Fig. 5i) as compared to aerosol mixing ratio (Fig. 5h) appear to be less on the SU flux profiles (Fig. 5g) since the wind speed associated with AARs is the highest over the Indian and Southern oceans, but the largest flux values are associated with the AARs over the eastern China region which has the largest aerosol mass mixing ratio.

Next, we examine the vertical profiles of CA AARs over the Congo Basin, the western USA, eastern China, the Indian Ocean, and the Southern Ocean. As in the case of dust aerosols over the Sahara desert to North Atlantic transport pathway (Fig. 5a; orange line), both the OC and BC AARs off the Congo Basin show elevated flux values around 700 hPa (Fig. 5j and m). It is to be noted that the wind speed within AARs peaks around 650 hPa, suggesting the influence from the AEJ-S (south; Adebiyi and Zuidema, 2016; Chakraborty et al., 2021b; Das et al., 2017) on the AARs' wind profile (Fig. 7k and n) that might be responsible for the peaks in the aerosol mixing ratio profiles around 750 hPa (Fig. 5l and o) instead of the exponential profiles as observed in most of other regions. AARs over eastern China (blue) also carry a large number of near-surface aerosol particles that contribute to large BC and OC IAT values in this region. However, when those AARs reach the western USA, the number of the BC and OC aerosol mass mixing ratio significantly decreases by ∼8 times, presumably because of the settlement and removal of the aerosol particles during the transport. As compared to SU AARs, the IAT values and aerosol mass mixing ratio of OC and BC AARs over the eastern USA is significantly lower (3–4 times less) than those originating from China and the Congo rainforest. Wind speed appears to have less influence on the flux profiles since the Indian Ocean and the Southern Ocean have the largest (smallest) wind speed (flux values) associated with the AARs.

It is apparent from Fig. 5 that the mean IAT averaged over AAR time steps is greater than the mean IAT averaged from all time steps. This is not surprising, since AARs detected by our algorithm represent the extreme transport events (see Sect. 2 for the definition and methodology). Considering that the original analysis on water vapor ARs (Zhu and Newell, 1998) highlighted that 90 % of the poleward transport of water vapor in the midlatitudes occurred in a relatively small number of extreme transport events (i.e., ARs), a similar question can be raised here for AARs. Specifically, what fraction of the total annual aerosol transport is accounted for by the 20–40 AAR days that occur each year on average (Fig. 4)? Also, do AARs transport a higher fraction of the total global annual transport over the major aerosol transport pathways, and is this dependent on aerosol species? In Fig. 6, we show the fraction of total annual IAT accounted for by AARs (shades), counting only the component of AAR transport in the direction of the total annual transport (arrows). Figure 6a shows that, annually, ∼ 40 000–100 000 kg m-1 of DU IAT is transported by All events (i.e., considering all data points in the record) that originate from the Sahara desert. Near the Equator, the fractional transport by AARs (FAAR) is around ∼20 % over the Atlantic Ocean in the direction of the Amazon rainforest (Fig. 6a), which is realized within ∼30 AAR d yr-1 (see Fig. 4a). In this same longitude sector, FAAR gradually increases with latitude and reaches up to 80 % around 35–40∘ N, where AARs act to transport dust from the Sahara desert to Europe and the Middle East region. Over Mongolia and China, the total annual IAT is ∼ 20 000–40 000 kg m-1 and about 30–40 AAR d yr-1 (Fig. 4a) contribute up to ∼ 40 %–50 % of that transport. In the Southern Hemisphere, Patagonia and South American drylands give rise to ∼4000 kg m-1 of annual dust transport and AARs contribute to a maximum of 20 %–40 % of that transport in about 20–30 AAR events per year (Fig. 4a). Over some regions far from the dust source region, such as over the maritime continent, the annual IAT value is very small by all events (Fig. 6a). About 5 AARs (shading, Fig. 4a) with very small object mean IAT values (arrows, Fig. 4a) are observed over there each year. It appears that, although AARs are not frequent (∼5 AAR d yr-1) over there, they are responsible for 80 %–100 % of the total annual transport.

SS AARs are far more frequent over the oceans than over the land (Fig. 4b), with peak frequencies of about 30 AAR d yr-1 occurring over the subtropical trade wind regions, the Southern Ocean and northern Atlantic Ocean. In these regions, the total annual IAT is about 10 000–20 000 kg m-1, and FAAR is about ∼ 20 %–30 % in the subtropical regions, reaching a maximum of ∼50 % over the Southern Ocean and the northern Atlantic Ocean (Fig. 6b).

AARs for other species of aerosols can contribute to a maximum of 40 % of the total annual IAT over their major transport pathways. For example, the ∼30 SU AAR d yr-1 that originate over China (Fig. 4c) transport ∼ 30 %–40 % of the total annual aerosol transport (∼2500 kg m-1) over the northern Pacific Ocean. FAAR associated with OC (BC) AARs reaches a maximum of 40 % (30 %) between the Congo Basin and the tropical Atlantic Ocean. Over China and the North Pacific Ocean, South Africa and the Southern/Indian oceans, south China and the western Pacific Ocean, and Amazon and the southern Atlantic/Southern oceans FAAR can reach up to a maximum of 40 %. Higher FAAR is also observed, despite low total annual IAT, over the northern part of the Sahel region for both BC and OC AARs.

Figure 7 shows AAR frequency along with the direction and magnitude of the mean AAR IAT during four different seasons. DU AARs (shading) originate during all the seasons over the Sahara desert. DU AAR IAT (arrows) is higher during DJF and MAM over the Sahara desert owing to a stronger anticyclonic motion in the boreal winter and spring. The largest number of DU AARs are generated during the MAM season when the DU IAT is the highest (Fig. 1) and widely spread across the Northern Hemisphere. In JJA and SON, DU AARs appear to have lower IAT (shorter arrows) but are more frequent than in the DJF season. The frequency of SS AARs also depends on the season. A higher number of SS AARs is detected over the eastern Pacific Ocean and the west coast of the USA, northern Atlantic Ocean, and Europe in DJF and MAM. In JJA, no SS AARs are detected over there. Over the Southern Ocean, SS AARs with large IAT are more frequent during austral winter or MAM and JJA. In comparison, tropical SS AARs are present year round.

SU AARs in the Northern Hemisphere are more (less) frequent in MAM (JJA) and can be related to the seasonal variations in IAT there (Fig. 1). The frequency of occurrences of SU AARs is higher during SON compared to DJF, while their IAT values are larger in DJF. This might be because of the readiness of sulfate aerosols to form cloud condensation nuclei (CCN) and their hygroscopic nature. The occurrences of intense AR-related precipitation in DJF over the extratropical region in the Northern Hemisphere might cause scavenging and wet removal of SU aerosols when the AARs and ARs coexist. A detailed investigation considering AR-AAR interactions is needed in future studies to better understand the coexistence and influence of AARs and ARs.

Over the Congo rainforest, several OC AARs occur during its dry season (also when CA IAT is high; Fig. 1). A similar frequency of occurrences of OC AARs is also observed during the dry season over the Amazon rainforest. Many OC AARs occur east of China during the DJF and MAM season when the IAT values are large over there (Fig. 1). Many BC and OC AARs are generated over the global rainforests during their dry seasons.

In this section, we characterize basic features of AARs (Fig. 8), including those related to geometry and IAT intensity and directions, and discuss and show the sensitivity of various AAR features to algorithm specifications and thresholds (Fig. 9). Figure 8a shows that the frequency of AARs decreases monotonically as AAR length increases for all aerosol species. The mean lengths of DU, SS, SU, OC, and BC AARs are 4264, 3722, 4121, 4528, and 4378 km, respectively. Figure 8b shows that AAR widths exhibit a skewed distribution, unlike AAR lengths, implying an optimum or common value for AAR widths around 400 km. The mean width of DU, SS, SU, OC, and BC AARs is 586, 642, 542, 625, and 589 km, respectively. When considered together in the form of the aspect ratio of AARs, i.e., length / width ratio, the distribution is also a skewed distribution (Fig. 8c). The lengths are typically 7–8 times their widths due to the skewness of the distribution. DU and SS AARs are the two species having the largest object mean IAT values (Fig. 8d). While, on average, DU AARs have IAT ∼1.65 and 1.3×10-3 kg m-1 s-1, respectively, SU, OC, and BC AARs have smaller IAT (Fig. 8d). Average object mean IAT values for SU, OC, and BC AARs are 0.1, 0.1, and 0.016×10-3 kg m-1 s-1, respectively. The frequency of object mean IAT for BC AARs decreases sharply and seldom reaches beyond 0.1×10-3 kg m-1 s-1. SU (OC) AARs attain a maximum object mean IAT values around 0.1×10-3 kg m-1 s-1, with a maximum frequency of ∼0.6 (0.4).

The coherence of IAT directions (Fig. 8e) is computed as a fraction of the number of grid cells with IAT directed within 45∘ of the direction of the object mean IAT to all the grid cells within that AAR. A large value implies that a larger fraction of the grid cells has IAT directed in the same overall direction as the AAR. AARs of all the species have a mean coherence of IAT directions between 0.91 (DU) and 0.94 (SU). The distribution of the direction of the object mean IAT implies that the AARs are mostly directed in the zonal direction (Fig. 8f). Relative to north, peak frequencies near 90∘ and between 250 and 310∘ imply that the AARs are either westerly or easterly in nature. A few westerly AARs also transport aerosols in the northeasterly (45–90∘) and southeasterly (90–135∘) directions. A wider peak between 250 and 310∘ implies that most of the easterly AARs also have meridional components in the southwesterly and northwesterly directions. The average values of the object mean IAT direction for all the species of AARs are ∼90∘ (westerly) and between 260 and 270∘ (easterly). These results show that the AARs mostly transport aerosols in the zonal direction as compared to ARs that have notable transport of water vapor in the meridional direction (Guan and Waliser, 2015).

Finally, Fig. 9 shows the sensitivity of AAR detection in our algorithm to three primary thresholds that define the geometry (length and aspect ratio) and grid-wise IAT limits (to identify the extreme transport) of AARs. The results presented above in this study are based on a length limit of 2000 km, aspect ratio limit of 2, and a pixel-wise IAT limit of the 85th percentile. Keeping the length and aspect ratio limits fixed, while increasing the IAT limit to the 90th and 95th percentile values, yields a lower number of AARs detected (top row). On the other hand, relaxing the length limit to 1750 and 1500 km greatly increases the number of AARs detected (middle row). Relaxing the aspect ratio limit to 1.5 and 1 does not significantly alter the number of AARs detected, suggesting that the other limits applied (such as length greater than 2000 km) already effectively remove IAT objects that are not elongated. It appears, from Fig. 9, that grid-wise IAT thresholds pose a large sensitivity to AARs detection as the number of AARs detected is reduced roughly by half as we increase that threshold to the 95th percentile from the 85th percentile used in our main analysis.

Figure 10 shows the sensitivity of the average global area coverage of all AARs to AAR algorithm parameters. Each bar represents the area-weighted global mean AAR frequency or a spatiotemporal measure of the amount of time/space that there is an AAR. Annually, AARs cover 3.42 % of the global area, respectively. Relaxing the length limits to 1750(1500) km increases the number of detections of AARs; thus, the areal coverage of AARs increases to 3.56 % (3.76 %). On the other hand, increasing the pixel-wise IAT limit to 90th and 95th percentile reduces the areal coverage to 2.44 % and 1.13 %, respectively. Relaxing the aspect ratio does not have any impact on AAR's areal coverage. ARs, on the other hand, account for ∼2 % more of the Earth's surface area than AARs (not shown), likely because the filter based on object mean IVT direction in the AR algorithm preferentially filtered out smaller objects compared to the revised filter in the AAR algorithm based on IAT magnitude.