2.1 Model description and simulation setup

The Goddard Earth Observing System (GEOS) can simulate atmospheric circulation and chemistry, oceanic circulation and biogeochemistry, land surface processes, and data assimilation at horizontal resolutions as small as 12 km (Molod et al., 2015; Rienecker et al., 2008). In this study we use GEOS as an atmospheric general circulation model, resolving the atmospheric circulation and composition, but using prescribed sea surface temperatures and sea ice fractions. GEOS can be used either in replay or in free running configurations. In the replay mode, simulations are constrained by the MERRA-2 reanalysis fields, while in the free-running mode the evolution of the atmosphere is not constrained to the observations. The results shown in this study were produced in free-running simulations with the Icarus 3.3 version of GEOS. The meteorological fields for the model initialization on the 13 August (00:00 Z) were obtained from version 2 of the Modern-Era Retrospective Analysis for Research and Applications (Gelaro et al., 2017).

Simulations were run on a cubed-sphere horizontal grid with horizontal resolution of ∼50 km with 72 hybrid vertical sigma levels extending from the surface to 0.01 hPa (Rienecker et al., 2008). Aerosol concentrations and aerosol processes are simulated with the GOCART module, a bulk aerosol module simulating the evolution of black carbon (BC), brown carbon (BrC), organic carbon (OC), nitrates (NO₃), sulfates (SO₄), dust, and sea salt (Chin et al., 2002, 2009; Colarco et al., 2010, 2017). In our simulations, aerosols are radiatively coupled to the dynamics and have a direct impact on the meteorological forecast. The optical properties of all the species, except dust and BrC, are determined by the OPAC dataset (Hess et al., 1998), while dust optics are treated as described in Colarco et al. (2014). As in Das et al. (2021), wildfire emissions are assigned to BC and BrC, which is more absorbing than OC in the near UV. The simulations also include the stratospheric chemistry module StratChem (Considine et al., 2000; Douglass and Kawa, 1999), which allows for the inclusion of background stratospheric sulfate from the oxidation of carbonyl sulfide. Stratospheric chemical reaction rates are impacted by the pyroCb smoke through temperature changes, but not via changes in heterogeneous chemistry.

The PNE plume is represented as a 0.3 Tg injection of carbonaceous aerosol, which is consistent with the observational findings from Peterson et al. (2018) and Torres et al. (2020). Of this, 97.5 % is BrC and the remaining 2.5 % is BC. The $\mathrm{BC}/\mathrm{BrC}$ ratio was set by Das et al. (2021) to reproduce the observed diabatic lofting of the plume, which is sensitive to the overall loading of the strongly absorbing BC. The 0.3 Tg is injected in three different locations in British Columbia, corresponding to the three most pronounced pyroCb clouds of the PNE. The aerosol have been initialized as smeared in regions centered on these locations, spanning horizontally 2^∘ longitude by 2.5^∘ latitude and vertically from 10 to 12 km. The aerosol injections occur during the first hours of 13 August, with the first two puffs occurring from 00:00 to 03:00 UTC (0.19 Tg, locations 1A and 1B, Fig. 1) and the third from 04:00 UTC to 06:00 UTC (0.11 Tg, location 2, Fig. 1). These time intervals have been obtained using cloud-top brightness temperature measurements by Das et al. (2021). The modal radius of the BrC aerosol in the model was set to 0.035 µm to achieve an Angstrom exponent of 1.3, following SAGE III measurements of the event (Das et al., 2021). In this work we present a simulation with the PNE injection, starting on 13 August 2017 and ending on 30 September 2017, which shows the formation of a SWIRL.

Figure 1Locations of the carbonaceous injections included in the GEOS simulations on 13 August 2017.

2.2 Definition of SWIRL boundaries and data analysis

We define the boundaries of the SWIRL based on the local anomaly of Ertel's potential vorticity (PV) (Ertel, 1942) with respect to the zonal mean at the same altitude and the BrC concentration. We identify the SWIRL as the region where the percent PV anomaly with respect to the PV zonal mean at the same altitude is smaller than −10 % and the mass mixing ratio of BrC is higher than 1.25 µg kg⁻¹. These criteria constrain PV anomalies to those associated with the presence of high carbonaceous aerosol concentrations. In order to obtain smooth SWIRL contours, both the carbonaceous aerosol concentration fields and the PV anomaly fields have been filtered horizontally using a Gaussian filter with a standard deviation of 1^∘ in the latitudinal direction and 1.5^∘ in the longitudinal direction. We concentrate on model levels higher or equal to 100 hPa, so as to ensure that the detected SWIRL is above the tropopause. The analysis proposed here to track the SWIRL follows the approach of Lestrelin et al. (2021), which uses the Lait PV anomaly (Lait, 1994), along with the ozone concentration anomaly, to locate the SWIRL from ERA5 reanalysis fields. A similar approach was also used by Khaykin et al. (2020) to track the SWIRL from the ANY event.

To investigate the characteristics of the SWIRL, we analyzed the following diagnostic quantities: Ertel's PV, temperature (T), potential temperature (θ), density (ρ), zonal wind speed (U), meridional wind speed (V), vertical wind speed (W), and carbonaceous aerosol (BC and BrC) concentrations. All these quantities are saved on model levels every 6 h, and are presented as absolute fields or as anomalies with respect to their zonal means at the same model level and latitude, unless specified otherwise. The geopotential of the model levels that is present in our discussion is approximated by the mid-layer geometrical height multiplied by the gravitational acceleration g. In our analysis we also used the temperature tendencies due to the diabatic processes resolved by the model (radiation, moist physics, friction, turbulence, and gravity wave drag) and to the dynamics. Additionally, the model calculates the contribution of the aerosol to the temperature tendency due to radiation through two calls to the radiation model, with and without the inclusion of aerosols. In order to better capture the relationship between the SWIRL and its closest surroundings, we restricted our analysis to a region centered on the SWIRL spanning 43.75^∘ longitude by 35^∘ latitude, and we calculate the diagnostic variables as horizontal averages over the portions of this region inside and outside the SWIRL. The SWIRL geometry was described considering its volume, which is calculated by summing the volumes of the grid cells inside the SWIRL contours over the height of the SWIRL, and its average area, which is calculated by vertically averaging the areas of the SWIRL computed for every layer by summing the areas of the cells' lower boundaries inside the SWIRL selection. The thickness was obtained by considering the height difference between the highest and lowest levels at which the SWIRL was detected.

To visualize and describe the evolution of the bulk SWIRL properties, we considered averages and extremes over the SWIRL volume of our meteorological fields and temperature tendencies.

3.1 SWIRL tracking

In the days following the PNE aerosol injection in the UTLS, the smoke plume drifted eastward and the diabatic self-lofting promoted its vertical movement. A brief description of the smoke transport is available in Appendix A, where we also include a comparison of the plume transport in the free-running and relaxed replay configurations of the GEOS model. This comparison shows how the relaxed replay mode is a good proxy for the observed smoke transport, but has the drawback of damping the effects on the dynamics of the radiative heating of the aerosol. On the other hand, the free-running simulation does not faithfully reproduce the observed smoke transport because the ambient meteorology in the free-running simulation differs from the observed transport. In this work we use the free-running configuration to resolve directly (i.e., with no contribution from reanalysis fields or measurements) the dynamical and thermodynamical impact of the aerosol radiative heating, which is crucial in the formation of the SWIRL.

In our free-running simulations the plume reached 100 hPa on 16 August, consistent with OMPS LP observations (Das et al., 2021), marking the start of the SWIRL detection using the algorithm described in Sect. 2.2. By the end of 18 August the detection algorithm revealed that the bulk of the SWIRL was above the 100 hPa limit, spanning between the levels 100 and 61 hPa. For our analysis, we mark 18 August as the first day of the stratospheric SWIRL. Before this date some signs of SWIRL formation were visible at levels lower than 100 hPa and above the simulated tropopause height (about 150–200 hPa in the days following the injection in the regions corresponding to the plume). We have chosen not to include this initial part in the analysis, since we are interested in the life of the SWIRL when it is well into the stratosphere.

Figure 2Maps showing the path followed by the simulated and observed PNE aerosol plume. The period considered spans from 13 August to 15 October (loss of vortex A in the observations). The orange lines show the evolution in time of the locations of the maximum BrC concentrations in the simulation, considering the regions centered on the SWIRL positions (Sect. 2.2); the red circle over Canada represents the injection location. Panel (a) shows the time evolution of the simulated SWIRL starting on 18 August. The closed contours show the regions where the SWIRL is detected in at least two vertical layers for several selected days. The contours are color-coded and labeled according to the number of days passed since the PNE event. Panel (b) shows paths and timing of the observed vortex O (full circles, dark blue line) and vortex A (full triangles, light blue line) for several selected days corresponding to the ones presented in (a). The full circles and triangles are color-coded and labeled according to the number of days passed since the PNE. The data on the observed stratospheric vortices were kindly provided by Bernard Legras and were obtained tracking the vortices' ozone anomalies in ERA5 fields, as described in Lestrelin et al. (2021).

The SWIRL is first detected by the algorithm over the Atlantic on 18 August (Fig. 2a) and it traveled southeastward until it reached the northern Atlantic. Here it stalled for 7 d (days 6–12 after the PNE, 19–25 August) before starting its westward movement, which eventually brought it to the Arabic peninsula (day 28 after the PNE, 10 September). The next day (11 September, day 29), the SWIRL started losing compactness and that marked the end of its stratospheric lifetime. After this date, the detection algorithm was still able to detect its fragmented remains until at least 16 September, when they reached the Caribbean Sea. The contours of the fragmented SWIRL have not been included in Fig. 2a.

The simulated development of the SWIRL differs from the observed one in several aspects. First, the transport is different, as shown in Fig. 2. The simulated stalling of the vortex on days 6–12 is absent in the observations (Lestrelin et al., 2021), where the vortex moves more quickly eastward. Second, Lestrelin et al. (2021) document the formation of offspring from the initial vortex (vortex O) following its stretching by the ambient meteorology: around 22 August a first splitting of the plume occurs, resulting in the formation of vortex A (light blue line, Fig. 2b) from vortex O (dark blue line, Fig. 2b). Moreover, on 1 September, vortex O split into two vortices (B1 and B2), whose trajectories are not reported here. This splitting is not reproduced by the simulations. Lastly, the simulated SWIRL lasted 25 d overall, which is shorter than in observations Lestrelin et al. (2021), where vortex A remained visible until 15 October. The differences between the simulated and observed development of the SWIRL are due to the underlying meteorology. The generation of offspring vortices and the development of stalling periods are caused by the interaction of the vortex with surrounding meteorological features, such as jets and troughs. Since the underlying meteorology is different in the simulation and observations, its influence on the smoke plume and on the anticyclonic vortices will be different, and we do not expect the evolution of the plume to be the same. Additionally, differences in the amount and properties of the carbonaceous aerosols in the simulated vortex with respect to the observed ones result in differences in its radiative interaction and, therefore, in its evolution.

3.2 Analysis of the SWIRL on 23 August

In this section we describe the structure of the simulated SWIRL on 23 August. On this day the vortex is well into the stratosphere (detected in the model levels between 72 and 43 hPa), with tangential wind speed anomalies of about 10 m s⁻¹, and well organized, with a clear anticyclonic circulation closed around its center.

Figure 3Horizontal sections at different heights of the SWIRL on 23 August, 18:00 Z, when the maximum heating occurs. The first to the fourth columns show 43, 52, 61, and 72 hPa altitudes, respectively. The first row (a–d) shows the potential vorticity anomaly [%], the second (e–h), the vertical velocity, the third (i–l), the potential temperature anomaly, the fourth (m–p), the temperature tendency due to radiative heating or cooling of the aerosol, and the fifth row (q–t), the geopotential anomaly (color shades) and the wind field anomaly (white arrows). All anomalies are calculated with respect to the zonal mean. The geopotential anomaly was made positive in the considered region by summing the absolute value of the minimum of the geopotential anomaly to the entire field. This was done for representation purposes. The black contours represent the boundary of the detected SWIRL.

Download

Figure 4Same as Fig. 3, but for 23 August 00:00 Z, when the aerosol is cooling due to the lack of incoming solar radiation.

Download

Inside the borders of the SWIRL lies the deep negative PV anomaly, which indicates the presence of an anticyclone (Figs. 3a–d and 4a–d). Indeed, the horizontal wind anomaly structure of the SWIRL shows an anticyclonic circulation centered over the PV anomaly minimum, with tangential wind magnitudes of about 10 m s⁻¹ (Figs. 3q–t and 4q–t). The magnitude of the tangential wind speed in the SWIRL simulated here is comparable to those presented in the literature for the SWIRL developed after the ANY event (Kablick et al., 2020; Allen et al., 2020). The analysis of the geopotential shows a positive anomaly corresponding to the SWIRL (Figs. 3q–t and 4q–t): this indicates that the effect of the aerosol is to trigger an outward pressure gradient force that, together with Coriolis acceleration, produces the anticyclonic motion of the SWIRL.

The pyroCb aerosol causes a strong internal heating of the SWIRL during daytime (Fig. 3m–p). During nighttime the aerosol does not absorb SW radiation but emits LW radiation, inducing a net cooling of the plume (Fig. 4m–p). Vertical velocities within the SWIRL are comparable to those outside the SWIRL (Fig. 4e–h). From this analysis there is no evidence of an organized structure in the vertical wind field corresponding to the SWIRL and there is no sign of its upward movement. This, however, is expected considering that the vertical displacement of the SWIRL, which can be quantified in hundreds of meters per day, should be driven by a vertical velocity of the order of fractions of cm s⁻¹, which cannot be visible in the noisy vertical velocity background. The lofting of the SWIRL will be made evident later, considering the average of the vertical velocity in the SWIRL (Fig. 7).

While the higher levels of the SWIRL are approximately elliptic, the lowermost levels present a peculiar shape, consisting of an ellipsoid to which a protrusion is attached in the downwind direction. This protrusion can be seen as the tail of the SWIRL, which gradually loses material in time, leaving a trail of air with negative PV anomaly and high aerosol concentration that is classified as SWIRL by our detection method. Such tails have been also observed by Lestrelin et al. (2021).

Another characteristic feature of SWIRLs is the vertical potential temperature anomaly dipole. This is visible considering the maps of potential temperature anomaly (Figs. 3i–l and 4i–l), which indicate a negative anomaly above the center of the SWIRL and a positive anomaly below. As stated by Allen et al. (2020), this signature in the potential temperature is consistent with the vertical expansion of the plume due to the diabatic heating of the aerosol.

Figure 5Vertical cross section at fixed longitude (−59^∘ W) of the SWIRL on 23 August, 18:00 Z. Panel (a) shows the percent relative potential vorticity anomaly, (b) the temperature anomaly, (c) the temperature tendency due to the aerosol heating, (d) the vertical velocity, (e) the zonal wind anomaly, and (f) the potential temperature anomaly. All anomalies are calculated with respect to the zonal mean between −180 and +180^∘ longitude.

Download

Figure 6Same as Fig. 5 but for a fixed latitude (36^∘ N). Panel (e) contains the meridional velocity anomaly.

Download

In the vertical sections of the SWIRL (Figs. 5–6e) a strong anticyclonic motion is visible as well as the vertical structure of the diabatic heating provided by the aerosol (Figs. 5–6c), which is particularly intense in the middle of the SWIRL. The analysis of the meridional wind anomaly curtain (Fig. 6e) reveals also the tilting of the SWIRL, whose axis lies at an angle with respect to the vertical. As described by Allen et al. (2020), the tilting of the vortex is given by the wind shear in the vertical direction. The characteristic vertical temperature and potential temperature anomaly dipoles that characterize SWIRLs are also visible in Figs. 5b–f and 6b–f, with negative anomalies in the upper part of the SWIRL and positive anomalies below. The magnitude of the temperature dipole (10 K) is comparable to that reported for the ANY case (Allen et al., 2020), which reached 15 K. The temperature and potential temperature anomalies extend well below and above the PV anomaly, as also reported in Allen et al. (2020); Lestrelin et al. (2021) and Kablick et al. (2020). As in the horizontal sections, the vertical velocity field does not exhibit any detectable difference between the interior of the SWIRL and its surroundings.

Figure 7Vertical profiles at four different times (00:00–06:00–12:00–18:00 Z) for 23 August. Solid lines represent vertical profiles of the horizontal averages inside the SWIRL selection, while dotted lines represent vertical profiles of the horizontal averages outside the SWIRL selection in a rectangular region 43.75^∘ longitude by 35^∘ latitude centered on the SWIRL. The panels show (a) absolute temperature, (b) vertical velocity, (c) Ertel's potential vorticity, (d) potential temperature anomaly with respect to the zonal mean, (e) horizontal speed anomaly with respect to the zonal mean, (f) density anomaly, (g) temperature tendency due to diabatic processes, (h) temperature tendency due to the aerosol, and (i) temperature tendency due to the dynamics.

Download

The temperature anomaly dipole shown in Figs. 5b and 6b is reflected in a reduced lapse rate inside the SWIRL compared to the surroundings (Fig. 7a), with the internal temperature profile intersecting the external one roughly in the middle of the SWIRL. The same can be observed in the potential temperature anomaly (Fig. 7d). Figure 7h shows the diurnal heating and nocturnal cooling of the SWIRL due to aerosols and Fig. 7b its resulting vertical motion. The difference in the average vertical velocity between the interior and exterior of the SWIRL is crucial, since it explains how the SWIRL actually moves upward despite being immersed in a noisy vertical velocity field.

The profile of the horizontal wind speed anomaly (Fig. 7e) shows stronger wind speeds in the middle part of the SWIRL; this was visible also in Fig. 3. The reason for this is that the concentrated heating that takes place inside the SWIRL produces a strong rotation which does not interact with the background flow. Indeed, the similarly strong heating taking place at the bottom of the SWIRL (Fig. 7h) does not result in a similarly strong rotation due to the SWIRL interaction with the background flow. This fact is also supported by the shape of the SWIRL, which presents tails in its lower levels caused by the dispersive action of the background flow.

The radiative heating or cooling due to the aerosol is the main factor among the physical diabatic processes (Fig. 7g–i). During the day, the SWIRL is subject to strong diabatic heating, which promotes the vertical motion from buoyancy. The vertical motion is accompanied by the expansion of the SWIRL, which partially compensates for this heating: this is evident in the temperature tendency due to the dynamics (Fig. 7b and i, red and dark red curves). During the night, the opposite situation occurs. The diabatic cooling drives a downward vertical motion of the SWIRL and its subsequent heating due to expansion (Fig. 7b and i, yellow and orange curves). The nocturnal cooling is less intense in absolute value than the diurnal heating. This points toward a net ascent of the SWIRL, given the enhanced intensity of the diurnal lofting with respect to the nocturnal descent (Fig. 7h).

3.3 Evolution in time of the SWIRL properties

The volume of the SWIRL increases until day 10–11 (22–23 August) and then decreases steadily in time (Fig. 8a). The daily modulation of the volume, corresponding to the expansion–compression cycle caused by the diurnal radiative cycle, is visible especially in the first 11–12 d. Accordingly, the calculated radius (Fig. 8b) shows the same time evolution, peaking around day 11 at 700 km. The peak radius of the simulated SWIRL is comparable to the maximum radius of the SWIRL observed following ANY (peaking at 750 km, Allen et al., 2020); instead, the simulated SWIRL is consistently larger in comparison with the main SWIRL observed following PNE, whose horizontal dimensions (diameter) where estimated at 700 km from composite analysis (Lestrelin et al., 2021). After day 28 (11 September) the SWIRL volume increases significantly, following its dispersion and fragmentation. The detection algorithm still categorizes the remains of the plume as SWIRL even thought they lack the characteristic compactness of SWIRLs. This process is also visible in the calculated average radius, with a steep increase after day 28. Moreover, the SWIRL thickness (Fig. 8c) indicates that the SWIRL reaches a maximum thickness of 5 km between day 11 (24 August) and 16 (29 August); before and after this period, the thickness value fluctuates between 3 and 4 km; the magnitude of the SWIRL thickness is comparable to the one observed in the ANY SWIRL, that peaked at 6 km in the moment of its maximum expansion (Allen et al., 2020), and it is also close to the thickness of 3 km of the PNE SWIRL obtained from the composite analysis of Lestrelin et al. (2021). The step-like fluctuations of the SWIRL thickness is due to the vertical resolution of the model at the considered levels, which is around 1 km. The ascent of the simulated SWIRL can be divided in two phases: the steep initial ascent, during which the volume of the SWIRL increases, and a second slower ascent phase, characterized by a decrease in volume of the SWIRL. The potential temperature θ increases approximately linearly (Fig. 8d) during both ascent phases, with a slope of about 10 K d⁻¹ during the first phase and of 6 K d⁻¹ during the second. The ascent speeds are consistent with the observed ones during the ANY SWIRL: in Allen et al. (2020) the ascent speed during the first 27 d of the SWIRL life is estimated at 7.8 and at 5.5 K d⁻¹ afterwards. The potential temperature of the simulated vortex is about 50 K higher than in Lestrelin et al. (2021) (Fig. 8d), possibly because of the faster and more consistent simulated ascent of the plume in the first days following the PNE (see also Appendix A). Despite this offset, between days 5 and 29 from the PNE the ascent rates of the simulated SWIRL are compatible to the rate of vortex O observed by Lestrelin et al. (2021). In the first 11 d of its life, vortex O has an ascent rate of about 9 K d⁻¹, while after its first splitting (formation of vortex A) it rises in the stratosphere at about 4.7 K d⁻¹. These estimates have been obtained with linear fits of the potential temperature curve of vortex O. Before its dissipation 30 d after the PNE, the simulated SWIRL reaches an average potential temperature of 590 K, comparable to the final potential temperature reached by vortex A (570 K) before its loss 60 d after the PNE (Lestrelin et al., 2021). Lastly, the average absolute temperature T within the SWIRL (Fig. 8f) shows little variation in time, with values between 206 and 209 K.

Figure 8Evolution in time of the SWIRL (a) volume, (b) mean radius, (c) thickness, (d) average potential temperature, (e) carbonaceous aerosol loading, and (f) absolute temperature. In (d) the red line shows the result of the linear fit between day 5 and 11 of the potential temperature, while the blue line is between day 11 and day 30. The steepness of the red line is 10.6 K d⁻¹, while for the blue line it is 6.2 K d⁻¹. The green dots and the blue triangles represent the potential temperature of vortices O and A, respectively, obtained using ERA5 meteorological fields in Lestrelin et al. (2021). The data were kindly provided by Bernard Legras.

Download

Figure 9Evolution in time of (a) maximum (red line) and minimum (blue line) temperature anomalies inside the SWIRL with respect to the zonal mean; the black line represents the magnitude of the temperature anomaly dipole computed as the distance between the red and blue lines. (b) Maximum (red line), minimum (blue line), and average (black line) speed anomalies with respect to the zonal mean inside the SWIRL (red line). (c) The minimum percent relative Ertel's PV anomaly with respect to the zonal mean inside the SWIRL. (d) The average vertical velocity in the SWIRL (black line) and outside the SWIRL (green line). (e) Average temperature tendency in the SWIRL due to diabatic processes (red line), average temperature tendency due to aerosol inside the SWIRL (blue line), and average temperature tendency outside the SWIRL due to diabatic processes (green line). (f) Averages inside the SWIRL of dynamical (black line) and total diabatic (red line) temperature tendencies. The green lines have been computed as follows: a region centered in the SWIRL of 43.75^∘ longitude by 35^∘ latitude was selected and the averages were carried out over the portion of this region at the same height of the SWIRL but outside the SWIRL.

Download

The magnitude of the temperature anomaly dipole in the SWIRL is approximately stable until day 10–11 (22–23 August), and then gradually decreases as the SWIRL ages (Fig. 9a). The simulated temperature dipole (peaking at 7 K) is significantly more pronounced than the one calculated for the main PNE SWIRL via composite analysis by Lestrelin et al. (2021) (peaking at 2 K). The maximum horizontal wind speed anomaly in the SWIRL (Fig. 9b) first decreases for 3 d, in the same period in which the radius of the SWIRL increases significantly; then it stabilizes around 12–13 m s⁻¹ before decreasing again after day 15 (28 August) and reaching figures between 6 and 8 m s⁻¹. The same behavior is observed in the average horizontal velocity anomaly that decreases in the first days from 10 to 5 m s⁻¹, then remains constant at 5 m s⁻¹ until day 15, and subsequently decreases gradually reaching 3 m s⁻¹ at the end of the SWIRL lifetime.

The minimum of the PV anomaly (Fig. 9c) abruptly increases at day 6–7 from about −140 % to about −90 % and remains roughly stable until day 11, when it starts increasing at a slower rate. The average vertical velocity of the SWIRL (Fig. 9d) is positive (indicating upwelling) at the end of the daily heating period and negative during the night, with maximum values peaking at 0.7 cm s⁻¹. The average vertical velocity of the SWIRL reaches more pronounced peaks than those of the surroundings, indicated by a green line in Fig. 9d.

The diabatic temperature tendency (Fig. 9e) is initially dominated by the radiative heating of the aerosol. The mean aerosol heating rate within the SWIRL is significantly larger than the average background diabatic temperature tendency until day 22–23 (4–5 September). After this day, the diabatic temperature tendency inside the SWIRL becomes progressively closer to the diabatic temperature tendency outside. The quick loss of coherence of the SWIRL that takes place after day 28 is driven by the loss of this differential heating between its inner region and the surroundings. The dynamical temperature tendency is anticorrelated with the diabatic temperature tendency (Fig. 9f), which is dominated by the aerosol heating, and is directly linked to the vertical velocity of the SWIRL: when the SWIRL is heated by radiation, it rises and expands, meaning that its average vertical velocity is positive and the dynamical temperature tendency is negative. During the night this tendency is reversed along with an average negative vertical velocity. This means that while the SWIRL is cooling, it moves downward while being subject to compression.