A ML-based perspective on spatio-temporal patterns of convective organisation

Convective clouds play a vital role in Earth's hydrological cycle. In the tropics, these clouds often form extensive, spatially connected structures known as mesoscale convective systems (MCSs). MCSs are significant contributors to severe weather and are linked to potential changes in precipitation extremes. However, they are still connected to uncertainties, particularly regarding the intensity and variability of their spatio-temporal clustering (convective organisation).

This study aims to characterise regional patterns of convective organisation and explore their connections to convective cloud microphysics. The analysis covers a region in tropical Africa between 30° W – 30° E and 30° N – 30° S with a focus on the spring-to-summer period. Convective clouds and the intensity of their organisation are detected using 3D radar reflectivities. These spatio-temporal contiguous predictions are derived through a machine learning (ML) based extrapolation of observations from passive (MSG SEVIRI) and active (CloudSat) 2D remote sensing sensors. Furthermore, three organisation indices are used to quantify the organisational state of the atmosphere. They are leveraged to examine the relationship between convective cloud development and large-scale organisation. Using an object-based algorithm, we identify convective core and anvil regions in the predicted 3D radar reflectivities at each time step. These cloud objects are tracked over time to construct seamless 4D trajectories that capture cloud movement in three dimensions. Then, we calculate the indices to characterise the degree of organisation at each time point. The study evaluates regional statistics for convective organisation and analyses the key features of the observed systems.

Our findings highlight regional hotspots of convective organisation over the Gulf of Guinea, continental West Africa, and the Atlantic Ocean. These areas frequently host long-lasting, highly active cloud systems, such as MCSs. We observe seasonal variations in convective cloud development lead to a modest 5 % increase in organisation during summer. For example, differences in landmass distribution and the influence of extratropical dynamics in the southern hemisphere contribute to greater variability compared to the northern hemisphere. Over the ocean, organisation indices are approximately 5–10 % higher than over land. Overall, the results highlight the importance of regional characteristics in assessing convective organisation. Integrating data from multiple remote sensing instruments offers valuable insights, potentially enhancing climate risk assessments. However, our study emphasises that the overlapping effects of isolated and clustered convection may impact the statistical analysis. Addressing this issue requires an adapted organisation index.