Machine Learning methodologies for identifying atmospheric deep convective events in ERA5

Atmospheric deep convection can occur when the warming of the Earth’s surfaceby solar radiation leads to buoyant plumes that break through the mixed layer andproduce vertical clouds reaching the tropopause. This phenomenon is associated withthunderstorms, heavy precipitation, hail, strong winds and other events that cause se-vere damage to life and property. However, representing deep convection and its as-sociated events in models is challenging as they depend on many high-resolution sub-grid processes which are difficult and expensive to simulate. As a consequence, someapproaches based on artificial intelligence and specially Machine Learning (ML) haverecently emerged to bypass some of these limitations of physical models. Here we discusssome of the ML methodologies implemented in the Convective Day Detector (CDD), astatistical model designed to identify hazardous convective events at ground level basedon ERA5 reanalysis data.First, we will describe the CDD, which is a ML classifier based on meteorologicalvariables from ERA5 reanalysis which are associated with deep convection, such asconvective available potential energy, vertical wind velocity or specific humidity. TheCDD is trained to find the relationship between these variables and the occurrence ofsevere weather events such as hailstorms and severe wind from observation-based reportsdatabases. The trained CDD is subsequently employed to infer the probability of theoccurrence of these convective events beyond the the training region, where observationsare more limited or inconsistent, if available at all.However, this modelling approach presents many challenges that need to be over-come. To start with, hazardous convective events are rare and difficult to measure in aconsistent manner. This leads to a very unbalanced training dataset, with many posi-tive unlabelled data. Therefore, we will discuss some ways to address these problems,such as under sampling, artificially filtering the storm database or positive unlabelledlearning methodologies. Additionally, the meteorological conditions that lead to thedevelopment of convective events are different depending on the location. As a conse-quence, we will also discuss transfer learning methodologies to apply a classifier trainedin North America to different regions of the world such as Europe, and how to validatethe results with very scarce and inconsistent observations.