Retrieval of Venus' aerosols vertical distribution: a Bayesian approach

With the most complex cloud structure among the terrestrial planets of our solar system, Venus atmosphere has been studied for decades, but many questions are still far from being answered. What is clear, however, is that understanding clouds is key to comprehending the overall behaviour of this dense atmosphere. Venusian aerosols are frequently modeled as a four mode distribution with sizes ranging from 1 to more than 3 microns located mainly between 45 and 75 km in altitude, although its upper bounder varies latitudinally [1,2]. Its composition is still under discussion, especially for the biggest particles [3], but a concentrated solution of sulfuric acid seems to be the main component. Sulfuric acid is photochemically produced from SO2 and H2O at the upper cloud altitude [4], where the temperature structure shows strong latitudinal variability [5]. The radiative energy balance is also affected by the cloud structure since the cloud top altitude affects both cooling and heating rates [6]. Furthermore, cloud-tracking based on infrared nightside images constitutes the most usual way to study both lower cloud dynamics [7] and winds in the upper clouds [8].

The Visible and Infrared Thermal Imaging Spectrometer (VIRTIS) onboard Venus Express mission was designed to study both Venus atmosphere and surface through its three channels [9] providing invaluable data to study the vertical structure of the clouds and its variability [1,2,10]. However, the retrieval of atmospheric properties through radiative transfer calculations is usually an inverse multiparametic problem in which the effect of different parameters on the simulated radiance is coupled. This is the case for Venus infrared windows, from which it is not easy to disentangle the effect of temperature, cloud, minor gases and surface [11,12]. Moreover, different models in which different parameters have been retrieved can also lead to similar fits to the data.

We wanted to delve deeper into this problem, focusing on the vertical distribution of the aerosols since there is no previous exhaustive analysis on the number of free parameters required and supported by these observations. One of the most widely used parameterisations is the use of ‘mode factors’, i.e., altitude-independent multiplicative values that affect the number density but not the altitude distribution [5]. Aerosol exponential profiles in which the aerosol scale height and cloud top are taken as free parameters are also frequently present in the literature [8,10]. In our work, we used Haus et al. [5] description of the aerosol vertical distribution and we studied whether varying individual parameters actually provided more information than using ‘mode factors’.

To do so, we used archNEMESIS [13] as radiative transfer code, which implements MultiNest [14], a Bayesian inference tool based on nested-sampling algorithms and studied individual spectra convering mid-latitude, cold collar and South Polar Vortex regions to analyse the information content on the different aerosol modes contained in each of them to find their most informative description. This is feasible thanks to the Bayesian evidence, which is a measure of the probability that the model truly represents the observations. This allows us to compare models with different parameterisations of the aerosols and to choose the one that achieved the highest Bayesian evidence. Furthermore, this technique also allows us to study the correlation between the parameters and, as traditional retrieval schemes, to determine atmospheric parameters and their uncertainty. With the parameterisation proposed here to retrieve the aerosol vertical distribution, we will re-analyse the temperature and cloud structure variability from the entire VIRTIS-M-IR nightside dataset in a forthcoming work. This future study will include the retrieval of instantaneous maps of atmospheric properties for each observation thanks to a clustering preprocessing applied to the measurements. It will also serve as an initial validation of the use of Bayesian methods to determine Venusian atmospheric properties, which could be very useful for analysing data from future missions such as Envision.

References:

[1] Ignatiev et al. (2009), JGR 114.

[2] Haus et al. (2014), Icarus 232,232-248.

[3] Mogul et al. (2025), JRG 130.

[4] Titov et al. (2018), SSR 214, 126.

[5] Zasova et al. (2006), CR 44, 364-383.

[6] Garate-Lopez and Lebonnois (2018), Icarus 313, 1-11.

[7] Hueso et al. (2015), PSS 113-14,78-99.

[8] Garate-Lopez et al. (2015), Icarus 245, 16-31.

[9] Piccioni et al. (2007), ESA Special Publication

[10] Lee et al. (2012), Icarus 217, 599-609.

[11] Haus and Arnold (2010), PSS 58, 1578-1598.

[12] García Muñoz et al. (2013), PSS 81, 65-73.

[13] Alday et al. (2025), JORS 13.