A weather regime characterisation of winter biomass aerosol transport from southern Africa

17 During austral winter, a compact low cloud deck over South Atlantic contrasts with clear sky over 18 southern Africa, where forest fires triggered by dry conditions emit large amount of biomass burning 19 aerosols (BBA) in the free troposphere. Most of the BBA burden crosses South Atlantic embedded in 20 the tropical easterly flow. However, midlatitude synoptic disturbances can deflect part of the aerosol 21 from the main transport path towards southern extratropics. 22 In this study, a characterisation of the synoptic variability controlling the spatial distribution of BBA 23 in southern Africa and South Atlantic during austral winter (August to October) is presented. By 24 analysing atmospheric circulation data from reanalysis products, a 6-class weather regime (WR) 25 classification of the region is constructed. The classification reveals that the synoptic variability is 26 composed by four WRs representing disturbances travelling at midlatitudes, and two WRs 27 accounting for pressure anomalies in the South Atlantic. The WR classification is then successfully 28 used to characterise the aerosol spatial distribution in the region in the period 2003-2017, in both 29 reanalysis products and station data. Results show that the BBA transport towards southern 30 extratropics is controlled by weather regimes associated with midlatitude synoptic disturbances. In 31 particular, depending on the relative position of the pressure anomalies along the midlatitude 32 westerly flow, the BBA transport is deflected from the main tropical route towards southern Africa 33 or the South Atlantic. 34 This paper presents the first objective classification of the winter synoptic circulation over South 35 Atlantic and southern Africa. The classification shows skills in characterising the BBA transport, 36 indicating the potential for using it as a diagnostic/predictive tool for aerosol dynamics, which is a 37 key component for the full understanding and modelling of the complex radiation-aerosol-cloud 38 interactions controlling the atmospheric radiative budget in the region. 39 dry season is presented. By using atmospheric circulation data from a 329 reanalysis product, a robust classification with 6 WRs is defined for August-to-October in the period 330 2003-2017. Four WRs (WR1, 3, 4 and 5) represent the fingerprint of midlatitude propagative 331 disturbances, while two WRs (WR2 and 6) are characterised by persistence and represent the 332 oscillation of the pressure field in the South Atlantic. In particular, WR2 is associated with a 333 reinforced South Atlantic anticyclone, and is the dominant WR during the dry season. The 334 stationarity of the WR2/6 system suggests a connection with the synoptic variability of the SAM, 335 which is also consistent with the South Atlantic Oscillation pattern firstly identified by Chen (2014). 336 At the interannual time scale, the occurrence of persistent WR2 and 6 also shows a strong 337 connection with the El Niño/Southern Oscillation through a tropical-extratropical Rossby wave 338 pattern. characterising for the first the the The characterisation of the routes in the to the the BBA, the predictive tools for the BBA spatial distribution in the region. In particular, by using reliable long coverage reanalysis products a classification for past decades can be built, and the BBA spatial distribution can be reconstructed where observations are not available. Furthermore, the WR characterisation can be used in climate model projections to estimate the future evolution of the

In this study, a characterisation of the synoptic variability controlling the spatial distribution of BBA 23 in southern Africa and South Atlantic during austral winter (August to October) is presented. By 24 analysing atmospheric circulation data from reanalysis products, a 6-class weather regime (WR) 25 classification of the region is constructed. The classification reveals that the synoptic variability is 26 composed by four WRs representing disturbances travelling at midlatitudes, and two WRs 27 accounting for pressure anomalies in the South Atlantic. The WR classification is then successfully 28 used to characterise the aerosol spatial distribution in the region in the period 2003-2017, in both 29 reanalysis products and station data. Results show that the BBA transport towards southern 30 extratropics is controlled by weather regimes associated with midlatitude synoptic disturbances. In 31 particular, depending on the relative position of the pressure anomalies along the midlatitude 32 westerly flow, the BBA transport is deflected from the main tropical route towards southern Africa 33 or the South Atlantic. 34 This paper presents the first objective classification of the winter synoptic circulation over South 35 Atlantic and southern Africa. The classification shows skills in characterising the BBA transport, 36 indicating the potential for using it as a diagnostic/predictive tool for aerosol dynamics, which is a 37 key component for the full understanding and modelling of the complex radiation-aerosol-cloud 38 interactions controlling the atmospheric radiative budget in the region. 39 The mean atmospheric conditions over South Atlantic and southern Africa in ASO are illustrated in 179 The WR classification shows two synoptic patterns (WR2 and 6) accounting for the oscillation of the 188 pressure filed in the South Atlantic and four synoptic patterns (WR1, 3, 4 and 5) accounting for 189 midlatitude pressure anomalies (Fig. 4). These four WRs represent the fingerprint of propagative 190 disturbances travelling along the midlatitude mean westerly flow with wave number 8-12, as 191 demonstrated by the EOF analysis (see Section S1). The synoptic variability is dominated by the WR2,192 which occurs at a frequency of 22.3% and is characterised by a high pressure anomaly in the South 193 Atlantic accompanied by a reinforcement of the midlatitude westerlies (Fig. 4a). Its symmetric 194 counterpart is represented by WR6, which occurs at a frequency of 17.7% and is characterised by a 195 low pressure anomaly and a weaker westerly flow in the midlatitudes (Fig. 4f). WR2 occurs mainly in 196 September-October, while WR6 does not show a clear seasonality (Fig. 5). The analysis of the 197 transitions from a WR into the others reveals that WR2 and 6 are dominated by persistence (Table  198 2). The remaining 60% of the synoptic variability in the region is characterised by eastward travelling 199 disturbances of the westerly flow, represented by WR1, 3, 4 and 5 (Fig. 4). WR1 and 3 occur more 200 frequently in August-September, while WR4 and 5 are more frequent in October (Fig. 5). In this case, 201 the analysis of the transition rates shows similar persistence ratios (from 0.39 to 0.46) and high rates 202 for preferred transitions (WR1 into 5, WR3 into 4, WR4 into 1, WR5 into 3, see At the global scale, the variability of the atmospheric circulation south of 20°S is dominated by the 205 southern annular mode (SAM), which consists of out-of-phase surface pressure and geopotential 206 height anomalies between the Antarctic region and the southern midlatitudes, resulting in the 207 modulation of the location and intensity of the westerly wind belt (Baldwin, 2001;Limpasuvan and 208 Hartmann, 1999). Pohl and Fauchereau (2012)  relationship between the WR occurrence and the SAM daily index is investigated using both the C2E 213 and the E2C approach. WR6 shows a statistically significant association with positive SAM phases 214 (not shown), coherently with expected weaker westerlies at midlatitudes (see Fig. 4f). However, the 215 WR2-SAM relationship results statistically weaker, and no relationship at all is found with WR1, 3, 4 216 and 5 (not shown). 217 The WRs describing propagative disturbances at midlatitudes (WR1, 3, 4 and 5) are characterised by 218 the longitudinal displacement of high-low pressure anomalies modulating the meridional circulation, 219 which in turn drives the poleward BBA transport above the South Atlantic and southern Africa. In 220 particular, WR3 favours the recirculation of BBA from the ocean towards Namibia and South Africa, 221 leading to significant positive AOD anomalies above all the continental stations ( Fig. 4c), while WR5 222 pushes the BBA recirculation above the South Atlantic and inhibits the BBA transport towards the 223 Indian Ocean (Fig. 4e). Conversely, WR1 and 4 are associated with a weaker BBA transport above 224 Namibia and South Africa, leading to significant negative anomalies above the continental stations, 225 and larger transport towards the Indian Ocean (Fig. 4ad). BBA transport along the Atlantic coast of 226 Namibia and South Africa is also anomalously high during WR6, which is characterised by a low 227 pressure anomaly in the South Atlantic inhibiting the transport towards subtropical South Atlantic, 228 and leading to significant negative anomalies above Ascension Island, and favouring a poleward 229 route driving anomalous BBA concentrations above the continental stations (Fig. 4f). WR2, 230 characterised by a high pressure anomaly in the South Atlantic strengthening the easterly flow in the 231 Tropics, is the only WR associated with a reinforcement of the main BBA transport route in the 232 tropical South Atlantic, and positive AOD anomalies only affect the Ascension Island station (Fig. 4b). 233

Synoptic characterisation of aerosol optical depth in-situ observations 234
The robustness of the synoptic characterisation of the BBA transport obtained from the CAMS data 235 is assessed by linking the WR classification to the observed AOD from AERONET stations in the 236 region (Table 1). 237 https://doi.org/10.5194/acp-2021-337 Preprint. Discussion started: 20 April 2021 c Author(s) 2021. CC BY 4.0 License.

9
The C2E characterisation of the AOD observations is presented in Fig. 6; the associated statistical 238 analysis is summarised in Table 3. AOD anomalies above Ascension Island during WR1-5 are evenly 239 distributed between positive and negative values, while AOD anomalies during WR6 show a 240 preference for negative values (Fig. 6a). The significance of this characterisation is confirmed by the 241 ANOVA with a level of confidence higher than 99%. Just south of the source region in Bonanza, 242 significant positive anomalies are observed during WR4 (Fig. 6b). However, the statistical significance 243 of this characterisation only reaches 93%. AOD variability at central Namibia stations (Gobabeb,244 Henties Bay and HESS) is dominated by WR1, leading to significant negative anomalies ( with WR4, however these anomalies are poorly significant (Fig. 6c,d). The ANOVA supports this 248 characterisation, indicating that the null hypothesis can be rejected with a level of confidence higher 249 than 99%. Similarly, the continental station in Upington shows significant negative anomalies during 250 WR1 and 4, and significant positive anomalies during WR3 (Fig. 6g), and the ANOVA indicates the 251 rejection of the null hypothesis with 99% level of confidence. In South Africa, the southernmost 252 station in Simon's Town does not show significant anomalies in association with any WR, and the 253 ANOVA confirms that the WR classification is not able to characterise the AOD variability (p=0.09). 254 The C2E characterisation performed using observed AOD data confirms the relationship between the 255 WRs associated with midlatitude disturbances (WR1, 3 and 4) and the BBA transport above the 256 AERONET continental stations, and between WR6 and the BBA transport above Ascension Island, as 257 shown by the CAMS data (cf. Fig. 4). The comparison with the synoptic characterisation performed 258 using a WR classification with 7 clusters highlights that the latter is less robust, showing poorer 259 ANOVA performances. Moreover, the additional WR, accounting for a strengthening of the 260 continental high, does not provide further characterisation of the AOD anomalies (see Section S2 for 261 details). 262 The E2C characterisation of the BB AOD station data is presented in Fig. 7; the associated statistical 263 analysis is summarised in Table 4. AOD anomalies are divided in quartiles, with quartiles from 1st to 264 4th representing anomalies from the largest negative to the largest positive, and the relative change 265 in WR occurrence is displayed for each quartile. In Ascension Island, the 3rd quartile is characterised 266 by a significant change in the WR frequency the distribution, with increased occurrence of WR4 ( positive AOD anomalies associated with more frequent WR3, 5 and 6, and negative anomalies 274 associated with more frequent WR1, 4 and 6 ( Fig. 6f,g). The E2C characterisation confirms the 275 importance of the midlatitude disturbances (WR1, 3 and 4) in controlling the AOD anomalies at the 276 AERONET continental stations, in particular by driving the largest anomalies (1st and 4th quartiles). 277 However, this approach shows some inconsistencies: WR4, which is characterised by a southerly 278 anomaly in the BBA transport along the Atlantic coast (Fig. 4d), is associated with positive AOD 279 anomalies in HESS instead; similarly WR6, characterised by a northerly BBA transport anomaly along 280 the coast (Fig. 4f), is associated with both positive and negative anomalies in HESS and Upington. 281 The origin of this ambiguities is likely due to the location of these stations at the margin of the BBA 282 transport path associated with the WR circulation patterns, making them highly sensitive to the 283 variability of the circulation around the centroid. The comparison with the synoptic characterisation 284 performed using a 7 cluster WR classification highlights the same ambiguities when the AOD 285 anomalies in the continental stations are associated with the WR describing a low pressure anomaly 286 in the South Atlantic (see Section S2 for details). 287

Interannual variability 288
The WR frequency in ASO is also analysed at the interannual time scale. All WRs show similar 289 interannual variability in the frequency of occurrence (2-4% standard deviation), with the exception 290 of WR2 showing the larger interannual variability (6% standard deviation) (Fig. 8). No trend is found 291 in the WR occurrence (a Mann-Kendall test is performed at 95% level of confidence), not surprisingly 292 due to the short time coverage of the reanalysis. Possible teleconnections controlling the WR 293 interannual variability are analysed by computing the linear correlation between the WR frequency 294 and the SST variability at the global scale (Fig. 9). WR1, 3, 4 and 5 do not show significant correlation 295 patterns at the global scale, with the exception of localised SST anomalies in the South Atlantic 296 (WR1, 4 and 5) and Warm Pool (WR1). Conversely, WR2 and 6 show a strong relationship with El 297 Niño/Southern Oscillation (ENSO)-like patterns. In particular, WR2 occurrence is associated with La 298 Niña conditions, while WR6 is associated with El Niño conditions. The linkage with La Niña conditions 299 can also explain the larger interannual variability of WR2 during the analysed period, mainly due to 300 the peak in 2010 associated with a strong La Niña event (Boening et al., 2012), and the minimum in 301 2015, associated with an extreme El Niño event (Hu and Fedorov, 2017). The analysis of the WR-SST 302 correlations performed by using NOAA ERSST data show similar teleconnection patterns (see Section 303 S3). Differently from the WRs associated with travelling disturbances, WR2 and 6 represent a sort of 304 stationary South Atlantic oscillatory pattern, which might interact at the synoptic time scale with 305 https://doi.org/10.5194/acp-2021-337 Preprint. Discussion started: 20 April 2021 c Author(s) 2021. CC BY 4.0 License. Ambrizzi, 1993). The teleconnection mechanisms are explored by computing the correlation 307 between the WR occurrence and the global geopotential at 200 hPa (Fig. 10), the level where 308 teleconnection signals are the strongest. Wave patterns connecting the Pacific to South Atlantic are 309 found for both the WR2 and 6, though significance for WR6 is weak. A similar modulation by the 310 ENSO of synoptic regimes in the Southern Hemisphere is also reported during austral summer by 311 Fauchereau et al. (2009) and Pohl et al. (2018). The correlation between the occurrence of WR1, 3, 4 312 and 6 and the global geopotential does not show organised patterns at mid-to-high latitudes (Fig.  313 10). The comparison with the 7 cluster WR classification shows similar results, however the 314 teleconnection patterns are less evident (see Section S2). 315 The impact of the WR interannual variability on the BBA transport is assessed by computing the 316 linear correlation with the CAMS organic matter mixing ratio and the BBA transport at 700 hPa (Fig.  317 10). The WR interannual variability affects the mid-tropospheric circulation in the subtropics, 318 modulating the BBA transport on both the zonal and the meridional direction. However, the 319 correlation analysis reveals that the WR variability has weak impact on the BBA transport at the differences (see Section S2). In particular, the additional WR, accounting for a strengthening of the 325 continental high, shows no significant impact on the BBA transport (see Fig. S11a). 326

Conclusions 327
In this paper, the first objective classification of the synoptic circulation over South Atlantic and 328 southern Africa during the dry season is presented. By using atmospheric circulation data from a 329 reanalysis product, a robust classification with 6 WRs is defined for August-to-October in the period 330 2003-2017. Four WRs (WR1, 3, 4 and 5) represent the fingerprint of midlatitude propagative 331 disturbances, while two WRs (WR2 and 6) are characterised by persistence and represent the 332 oscillation of the pressure field in the South Atlantic. In particular, WR2 is associated with a 333 reinforced South Atlantic anticyclone, and is the dominant WR during the dry season. The 334 stationarity of the WR2/6 system suggests a connection with the synoptic variability of the SAM, 335 which is also consistent with the South Atlantic Oscillation pattern firstly identified by Chen (2014). The synoptic classification is used to characterise the transport of BBA from equatorial Africa, which 340 dominates the aerosol atmospheric content in the region during the dry season. By analysing 341 reanalysis data, it is found that WR2 and 6 modulate the easterly transport from tropical Africa 342 sources, which is the main climatological transport route. The synoptic characterisation also shows 343 that midlatitude propagative disturbances modulate the BBA transport from equatorial Africa, 344 elucidating the mechanism responsible for the BBA transport to the extratropics, which is peculiar in 345 this period of the year. Specifically, WR3 drives enhanced transport above the continent, while WR1 346 inhibits the transport; WR5 drives the BBA recirculation over the ocean, which is inhibited by WR4. 347 The BBA transport characterisation is also tested by using AOD observations from AERONET stations,      (Table 1). Red lines display the linear regression between CAMS and AERONET data, and the coefficients of the regression models are also reported in the plots, along with the correlation coefficient and the p-value. In titles, the size of the sample used in the linear regression model is reported in brackets (see Section 2 for details).
https://doi.org/10.5194/acp-2021-337 Preprint. Discussion started: 20 April 2021 c Author(s) 2021. CC BY 4.0 License. Figure 6. Circulation to environment characterisation: distributions of the AOD anomalies at 500 nm at the AERONET stations (Table 1), and as a function of the WRs. Probability density functions are estimated by using a normal kernel density; red lines represent 25th, 50th and 75th percentiles. For each WR, the p-value of a Kolmogorov-Smirnov test used to assess the difference with the total sample is reported. In titles, in brackets the number of available daily observations and the p-value of the ANOVA used to assess the WR characterisation are reported. . Environment to circulation characterisation: WR frequency anomaly as a function of the quartiles of the AOD anomalies at 500 nm at the AERONET stations (Table 1). Values are percentage changes relative to climatological frequencies. In brackets, the number of available daily observations are indicated.