Convective organization and 3D structure of tropical cloud systems deduced from synergistic A-Train observations and machine learning

Abstract

Abstract. We are building a 3D description of upper tropospheric (UT) cloud systems in order to study the relation between convection and cirrus anvils. For this purpose we used cloud data from the Atmospheric InfraRed Sounder and the Infrared Atmospheric Sounding Interferometer and atmospheric and surface properties from the meteorological reanalyses ERA-Interim and machine learning techniques. The different artificial neural network models were trained on collocated radar–lidar data from the A-Train in order to add cloud top height, cloud vertical extent and cloud layering, as well as a rain intensity classification to describe the UT cloud systems. The latter has an accuracy of about 65 % to 70 % and allows us to build objects of strong precipitation, used to identify convective organization. This rain intensity classification is more efficient to detect large latent heating than cold cloud temperature. In combination with a cloud system analysis, we found that deeper convection leads to larger heavy rain areas and a larger detrainment, with a slightly smaller thick anvil emissivity. This kind of analysis can be used for a process-oriented evaluation of convective precipitation parameterizations in climate models. Furthermore, we have shown the usefulness of our data to investigate tropical convective organization metrics. A comparison of different tropical convective organization indices and proxies to define convective areas has revealed that all indices show a similar annual cycle in convective organization, in phase with convective core height and anvil detrainment. The geographical patterns and magnitudes in radiative heating rate interannual changes with respect to one specific convective organization index (Iorg) for the period 2008 to 2018 are similar to the ones related to the El Niño–Southern Oscillation. However, since the interannual anomalies of the convective organization indices are very small and noisy, it was impossible to find a coherent relationship with those of other tropical mean variables such as surface temperature, thin cirrus area or subsidence area.