Observation of absorbing aerosols above clouds over the South-1 East Atlantic Ocean from the geostationary satellite SEVIRI 2 Part 1 : Method description and sensitivity 3 4

13


9
(2) Met Office, Exeter, UK High temporal resolution observations from satellites have a great potential for studying the 15 impact of biomass burning aerosols and clouds over the South East Atlantic Ocean (SEAO). 16 This paper presents a method developed to retrieve simultaneously aerosol and cloud properties 17 in aerosol above cloud conditions from the geostationary instrument Meteosat Second 18 Generation/Spinning Enhanced Visible and Infrared Imager (MSG/SEVIRI). The above-cloud 19 Aerosol Optical Thickness (AOT), the Cloud Optical Thickness (COT) and the Cloud droplet 20 Effective Radius (CER) are derived from the spectral contrast and the magnitude of the signal 21 measured in three channels in the visible to shortwave infrared region. The impact of the 22 absorption from atmospheric gases on the satellite signal is corrected by applying 23 transmittances calculated using the water vapour profiles from a Met Office forecast model. 24 The sensitivity analysis shows that a 10% error on the humidity profile leads to an 18.5% bias 25 on the above-cloud AOT, which highlights the importance of an accurate atmospheric 26 correction scheme. In situ measurements from the CLARIFY-2017 airborne field campaign are 27 used to constrain the aerosol size distribution and refractive index that is assumed for the 28 aforementioned retrieval algorithm. The sensitivities in the retrieved AOT, COT and CER to 29 the aerosol model assumptions are assessed. Although an uncertainty of 31.2% is observed on 30 the above-cloud AOT, the retrieval of the absorption AOT and both cloud properties is weakly 31 sensitive to the aerosol model assumptions, with biases lower than 7% and 3% respectively. 32 The stability of the retrieval over time is analysed. For observations outside of the 33 backscattering glory region, the time-series of the aerosol and cloud properties are physically 34 consistent, which confirms the ability of the retrieval to monitor the temporal evolution of 35 aerosol above cloud events over the SEAO. 36 37

39
The South East Atlantic Ocean (SEAO) provides a natural laboratory for analysing the full 40 range of aerosol-cloud-radiation interactions. During the fire season, large amounts of particles 41 from African biomass burning are transported above the semi-permanent deck of stratocumulus 42 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-1333 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 10 January 2019 c Author(s) 2019. CC BY 4.0 License. covering this oceanic region. As a result, an important contrast is expected in the Direct 43 Radiative Effect (DRE) of aerosols (i.e. the direct impact of aerosol scattering and absorption 44 of radiation). On one hand, the aerosol scattering above the ocean typically increases the local 45 albedo which leads to a negative DRE at the top of the atmosphere. On the other hand, the sign 46 of the DRE above clouds depends on the underlying cloud albedo and the aerosol absorption. 47 Positive instantaneous radiative forcing up to +130W m -2 has been observed by satellite 48 instruments over the SEAO (De Graaf et al., 2012; Peers et al., 2015). There are many poorly 49 constrained variables, such as the aerosol and cloud properties, vertical structure of aerosol and 50 clouds (Peers et al., 2016), which result in a large spread in the DRE derived from climate 51 models in this region (Zuidema et al., 2016). In addition, the absorption of radiation by aerosols 52 leads to a modification of the atmospheric stability and consequently on the formation, 53 development and dissipation of clouds, i.e. semi-direct effect. Studies have shown that the 54 overlying African biomass burning aerosols are associated with a cloud thickening (Wilcox, 55 2010(Wilcox, 55 & 2012. This negative semi-direct effect partly compensates the positive DRE of 56 aerosols above clouds over the SEAO. However, as an aerosol plume moves away from the 57 coast and descends into the boundary layer, the heat due to the aerosol absorption could lead 58 to a reduction of the cloud thickness (Koren et al., 2004). Biomass burning particles may also 59 interact with cloud droplets leading to a modification of the microphysics of the cloud, its 60 lifetime and precipitations (Twomey, 1974;Rosenfeld, 2000). Recent  Generation) and provides a full-disc observation every 15 minutes, offering a unique 94 opportunity to monitor the evolution of the cloud cover and to track aerosol plumes over the 95 SEAO. The objective of this two-part paper is to demonstrate the potential of this instrument 96 to retrieve simultaneously aerosol and cloud properties in the case of absorbing aerosols above 97 clouds. In this first contribution, we describe the approach used to derive the above-cloud 98 Aerosol Optical Thickness (AOT), the Cloud Optical Thickness (COT) and the Cloud droplet 99 Effective Radius (CER) and discuss the accuracy of the retrievals. The algorithm, as well as 100 the atmospheric correction scheme and the assumed aerosol model, are presented in Section 2. 101 The sensitivities in the retrieved quantities to the water vapour profile and the aerosol property 102 assumptions are assessed in Section 3. The evaluation of the stability of the retrieval is shown 103 in Section 4 and conclusions are drawn in Section 5. In a second companion paper, we will 104 compare our SEVIRI-based retrievals of cloud and aerosol properties with those from MODIS 105 products (Meyer et al., 2015) and also in situ aircraft observations from the CLARIFY-2017 106 field campaign. 107 108 The approach used to retrieve aerosol and cloud properties from satellite spectral radiance 113 measurements relies on the colour-ratio effect (Jethva et al., 2013). The signal backscattered 114 by a liquid cloud is almost spectrally neutral from the UV to the Near Infra-Red (NIR). On the 115 other hand, the absorption from biomass burning aerosols is typically larger at shorter 116 wavelengths. Therefore, the presence of absorbing aerosols above clouds modifies the apparent 117 colour of clouds. This enhancement of the spectral contrast can be detected by any passive 118 remote sensing instrument with two channels with enough separation in the UV/NIR region. 119

Retrieval method
The SEVIRI instrument, aboard the MSG satellite (Aminou et al., 1997), has channels centred 120 at 0.64 and 0.81µm. Figure 1  azimuth. The cloud is located between 0 and 1 km and the aerosol layer is between 2 and 3 km. 127 Aerosols have a refractive index of 1.54 -0.025i and the size distribution follow a lognormal 128 with an effective radius of 0.1µm. The cloud droplets have an effective radius of 10 µm. Rayleigh scattering has been accounted for but the simulations do not include the absorption 130 from atmospheric gases. A Lambertian surface with an albedo of 0.05 is assumed. For AOT = 131 0, the radiance ratio is around 1 and is largely invariant as a function of COT. As the AOT 132 increases, the radiance at 0.81µm as well as the radiance ratio decreases, indicating that the 133 attenuation from the aerosol layer is larger at 0.64 µm. 134 135 As in the Nakajima and King technique (1990), the sensitivity of the retrieval to the CER is 136 brought by the Short-Wave Infra-Red (SWIR) channel of SEVIRI, centred at 1.64µm. Figure  137 2 shows the radiances at 0 contrast between the VIS and the NIR, and therefore, on the above-cloud AOT retrieval. The 173 atmospheric correction, and especially the water vapour one, is essential to accurately retrieve 174 the aerosol and cloud properties from SEVIRI. 175 176 In order to correct the SEVIRI measurements for atmospheric absorption, the transmittances 177 T including radiosondes and remote sensing sounding data from many meteorological satellites. 185 The forecast is run every 6 hours and the humidity profile used for the atmospheric correction 186 comes from the latest time-appropriate forecast field available. The profiles of the remaining 187 gases -including ozone, carbon dioxide and methane -are those implicitly assumed by the 188 RTTOV calculations (Matricardi, 2008). The radiance measured by SEVIRI R atm,l is finally 189 corrected using: 190 where R l is the radiance corrected from the gaseous absorption. 38.6%. According to Magi and Hobbs (2003), the light scattering coefficient of an aged African 215 biomass burning plume only increases by a factor of 1.01 for a relative humidity of 40%. For 216 this reason, the impact of humidity on the PCASP and EXSCALABAR measurements is 217 neglected. 218 219 The aerosol properties needed for the SEVIRI retrieval are the size distribution and the complex 220 refractive index. The normalized number size distribution (dN/dlnr) is commonly represented 221 by a combination of lognormal modes: 222 where N i , r i and s i are the number fraction, the geometric mean radii and the standard deviation 223 of the mode i, respectively. As in most remote sensing applications, it has been chosen to 224 represent the particle size distribution for the aerosol during CLARIFY-2017 with a fine and a 225 coarse mode contributions. The aerosol model and resulting optical parameters are selected by 226 fitting simultaneously the PCASP measurements (Fig. 4a) and the SSA from EXSCALABAR 227 (Fig. 4b)  Lambertian with an albedo of 0.05 at all wavelengths which is typical of the sea-surface albedo 262 under diffuse radiation conditions. The aerosol and cloud properties assumed for the LUT are 263 summarized in Table 1. The truncation of the cloud droplet phase function has been done using 264 the delta-M method (Wiscombe, 1977) and the TMS correction (Nakajima and Tanaka, 1988) 265 has been applied. The cloud layer is located between 0 and 1 km and the aerosol layer between 266 2 and 3 km. The sensitivity of the algorithm to the altitudes of the aerosol and cloud layers is 267 expected to be negligible due of the small contribution of the Rayleigh scattering to the signal 268 at the SEVIRI wavelengths. The cloud droplets follow a gamma law distribution characterised 269 by an effective variance of 0.06. When the cloud is optically thin and/or the cloud droplets are 270 too small, it is not possible to separate the contribution to the optical signal arising from 271 aerosols from that of clouds. Therefore, the minimum values for the CER and the COT in the 272 LUT are 4 µm and 3, respectively. This also justifies the assumption of a relatively simple sea-273 surface reflectance parameterisation as, at COTs exceeding 3, the sea-surface has little impact 274 on the upwelling radiances above clouds. Clouds associated with lower COT and/or CER are 275 rejected. The aerosol model corresponds to the CLARIFY-2017 model mentioned above, 276 assuming the same refractive index at the 3 SEVIRI wavelengths. 277 278 The retrieval of the above-cloud AOT, COT and CER is performed simultaneously. The result 279 corresponds to the parameters that minimise the difference e between the simulated radiances 280 R sim and the corrected satellite signal R l : 281 When the simulated signal is not close enough to the satellite measurements (i.e. e > 0.0006), 282 the result is rejected. The retrieval of the above-cloud AOT is highly uncertain at the cloud 283 edges and for inhomogeneous clouds. In order to remove these results, the products are 284 aggregated onto a 0.1 ´ 0.1° grid and the standard deviation of the AOT and the CER are 285 calculated. Note that each grid cell represents around 12 SEVIRI observations. The 286 inhomogeneity parameter r is defined by the ratio of the standard deviation of a parameter to 287 the average value of this parameter. The results corresponding to a standard deviation of the 288 AOT larger than 0.7 and/or r CER > 0.2 as well as grid cells associated with less than 9 successful 289 retrievals are rejected. The atmospheric transmittances above clouds used to correct the SEVIRI measurements from 318 the gas absorption are calculated based on forecasted water vapour profiles. In order to assess 319 the sensitivity of the retrieval to the atmospheric correction, new transmittances have been 320 calculated for the event studied here, modifying the specific humidity by +/-10%, which can 321 be considered as an upper limit for the error in the forecast model. The aerosol and cloud 322 properties retrieved with the modified atmospheric corrections are aggregated on a 0.1 ´ 0.1° 323 grid. Figure 7 compares the retrieved aerosol and cloud properties from SEVIRI-measured 324 radiances using the original RH forecast with the perturbed RH (+10% in orange and -10% in 325 blue). The uncertainty on the water vapour content impacts mainly the retrieval of the above-326 cloud AOT, and then the COT, because of its effect on the radiance ratio. A +10%/-10% bias 327 on the humidity leads to an overestimation/underestimation of the AOT and COT respectively.  difference of 2.2% on the COT and 1.0% the CER. The effect associated with a change in the 358 fine mode radius is even lower than 1%. As expected, the above-cloud AOT is more sensitive 359 to the aerosol size distribution used for the inversion and differences up to 11.8% have been 360 observed when the fine mode standard deviation is decreased by 0.1. However, the retrieval of 361 the AOT is based on the detection of the aerosol absorption of the light reflected by the clouds. 362 Therefore, the impact of an error on the aerosol size distribution on the AAOT retrieval is 363 reduced to 5.4% for the standard deviation and 1.4% for the fine mode radius. 364 365 To assess the impact of the assumed aerosol refractive index on the retrieved aerosol and cloud 366 properties of interest, variations of +/-0.02 and +/-0.008 have been applied to the real and 367 imaginary parts of the refractive index, respectively. Figure 10 and 11 compare the retrieved 368 aerosol and cloud properties from SEVIRI radiance data for the CLARIFY-2017 aerosol model 369 with those retrieved when the aerosol refractive index parameters are perturbed. The influence 370 of refractive index is similar to the one of the modified aerosol size distribution in that 371 differences of <1% are observed in both COT and CER and a larger impact is found on the 372 AOT with differences up to 39% where the imaginary refractive index is decreased by 0.008. 373 The magnitude of the impact on the AOT is correlated to the difference of SSA between the 374 CLARIFY-2017 and the perturbed aerosol model. Therefore, the retrieval of the AAOT is also 375 less sensitive to the assumption on the aerosol refractive index, with an impact lower than 376 6.5%. 377 378 In order to evaluate the uncertainty u aer of the retrieved aerosol and cloud properties due to the 379 aerosol model assumptions, we combined the uncertainty u i from the above sensitivity studies 380 using: 381 The uncertainty has been estimated at 31.2% on the AOT, 2.3% on the COT and 1.2 % on the 382 CER. Owing to the sensitivity of the retrieval to the aerosol absorption above clouds, a 6.1% 383 uncertainty has been obtained on the AAOT, which is, together with the cloud albedo, the main 384 parameter for the estimation of the DRE of absorbing aerosols above clouds. 385 386 4. Assessing the stability of the retrieval 387 388 One of the major benefits from using SEVIRI is the ability to track both aerosol and cloud 389 events at high temporal resolution. Therefore, it is important to evaluate how consistent the 390 retrieval is over time. For that purpose, two days of continuous observations (i.e. 5 th and 6 th 391 September 2017) have been analysed and the retrieved properties have been averaged over 392 20˚S and 10˚S, and 5˚E and 15˚E. The above-cloud AOT, COT and CER time series are 393 presented in Figures 12a, b and c. The studied area is located next to the coast, where the AOT 394 is typically the highest. The above-cloud AOT is around 0.66 and 0.72 for the 5 th and the 6 th 395 September, respectively. As expected, the transport of the aerosol plume from east to west is 396 slow, resulting in a small evolution of the above-cloud AOT that can be expressed as a linear 397 trend. In order to assess the variability of the retrieved AOT, the linear trend +/-2 times the 398 standard deviation have been plotted on figure 12a (dashed lines). On both days, a peak is 399 observed at 12:15pm with an anomaly larger than the AOT variability. The evolution of the 400 cloud properties is slightly more complex. A small decrease is observed on both the COT and 401 CER until 2pm. After 3pm, both properties sharply increase. The clouds are strongly affected 402 by the diurnal cycle and a shoaling of the cloud cover is expected from early morning to late 403 afternoon. As the thinnest clouds vanish, the cloud fraction decreases together with the number 404 of retrievals in the area. This results in a larger contribution of the thickest clouds to the mean 405 value in the late afternoon. As for the above-cloud AOT, large variations of the CER are 406 observed around noon. At that time, the sun and the satellite are almost aligned and the 407 scattering angle ( fig. 12d) reaches values larger than 175˚ which corresponds to the region 408 where the glory phenomenon is typically observed. Several reasons can explain why the 409 retrieval does not perform well in backscattering direction. The first one is the uncertainty in 410 the LUT due to the truncation of the cloud phase function. Although the TMS correction gives 411 good results, biases still remain in the glory aureole (Iwabushi and Suzuki, 2009). Also, the 412 radiances in the glory are more sensitive to the cloud droplet microphysics (Mayer et al., 2004). 413 The assumption on the variance of the droplet size distribution may induce biases in the 414 retrieval. Therefore, the accuracy of the retrieval cannot be guaranteed within the glory aureole 415 and these observations should be discarded. In Figure 12, the timespans corresponding to the 416 MODIS Aqua and Terra overpasses in the region are highlighted in orange. This shows that 417 MODIS measurements are typically performed before and after SEVIRI observes the glory 418 backscattering over the SEAO, usually allowing comparisons between these instruments. 419 Except from the glory backscattering, the stability observed on the retrieved aerosol and cloud 420 properties reinforces the reliability of the algorithm.

424
Recently, progress has been made in the remote sensing field in order to fill the lack of aerosol 425 above cloud observations. Techniques have been developed to retrieve aerosol and cloud 426 properties over the SEAO from passive remote sensing instruments. These algorithms take Compared to other satellite instruments, the SEVIRI measurements are more sensitive to the 438 absorption from atmospheric gases because of the wider spectral bands. Therefore, an efficient 439 atmospheric correction scheme is essential in order to separate the aerosol absorption from the 440 atmospheric gas contribution. Atmospheric transmittances are calculated with the fast-radiative 441 transfer model RTTOV based on the cloud top height observed by SEVIRI and the forecasted 442 water vapour profiles from the Met Office Unified Model. The water vapour correction has the 443 largest impact on the above-cloud aerosol retrieval. The impact of errors in the atmospheric 444 correction has been evaluated by modulating the humidity profile for a case study. A positive 445 bias of both the AOT and the COT is observed when the water vapour is overestimated, and 446 vice versa. On average, an 18.5% bias on the AOT and a 5.5% bias on the COT are expected 447 for a 10% error on the water vapour profile. Although a good accuracy is expected from the 448 forecast model, this limitation should be kept in mind when utilising or further developing 449 SEVIRI products. In the companion paper, the humidity from the forecast will be compared 450 against the dropsonde measurements from the CLARIFY-2017 campaign. 451 452 The choice of the aerosol model used to produce the LUT is also a key feature of the method. 453 In situ measurements of aerosols above clouds have been performed off the coast of Ascension 454 Island during the CLARIFY-2017 field campaign. An aerosol model optimised for the SEVIRI 455 spectral bands has been obtained by analysing the vertical profiles of extinction and absorption 456 from EXSCALABAR together with the size distribution from a PCASP. A bimodal lognormal 457 distribution has shown to adequately reproduce the observations. A fine mode radius of 0.12 458 µm has been obtained, which is in good agreement with the biomass burning measured over Retrievals have been performed considering aerosol models with modified size distributions 467 and refractive indexes. It has been shown that the sensitivity of the retrieved cloud properties 468 to the aerosol model is small with errors lower than 3% on the COT and the CER. As expected 469 the impact of the aerosol model assumption is much larger on the above-cloud AOT, with an 470 uncertainty estimated at 31.2%. Owing to the sensitivity of the method to the aerosol absorption 471 above clouds, a better accuracy is obtained on the retrieved AAOT, with an error of 6.1% only. 472 This indicates that the estimated above-cloud AOT can be easily converted from one aerosol 473 model to another and that the results can be used to estimate the aerosol DRE above clouds. 474 475 Despite the wider channels and the narrower spectral range of SEVIRI, it has been 476 demonstrated that the geostationary instrument has the potential to detect and quantify the 477 absorbing aerosol plumes transported above the clouds of the SEAO. Except from observations 478 within the glory backscattering for which the retrieval has shown to be unstable, a good 479 consistency has been observed on the aerosol and cloud properties. The stability of the results 480 during the day is promising for future uses of the SEVIRI algorithm. In the companion paper, 481 the reliability of the retrieved aerosol and cloud properties will be further assessed by analysing 482 the consistency with the MODIS retrievals and comparing with direct measurements from the 483 CLARIFY-2017 field campaign. The potential of such a retrieval is obvious. The 15-minute 484 resolution will aid in tracking the fate of above cloud biomass burning aerosol and will prove 485 invaluable for assessing models of the emission, transport and deposition of biomass burning 486 aerosol, with implications for accurate determination of the direct radiative effects of biomass 487 burning aerosol at high temporal resolution.

Cloud model
Size distribution Gamma law r eff from 4 to 60 µm v eff = 0.06