Nature vs Nurture in Convective-Scale Ensemble Spread: Relative Influence of Inherited and Environmental Factors

Convective-scale ensembles are routinely used in operational centres around the world to produce probabilistic precipitation forecasts, but the added value provided by these models is limited by a lack of spread between members. While much work has been done to understand the spatial spread of precipitation patterns in the convective-scale ensemble itself, relatively little work has been done to explore the culpability of the driving ensemble for the lack of spread. Here, we borrow terminology used in psychology to describe these different sources of convective-scale spread: in the "nature" scenario, the spatial spread of precipitation patterns is simply inherited from the driving ensemble, while in the "nurture" scenario the environment is the dominant source of spread, and there is the potential for large differences to emerge between the two ensembles.

This study explores which of these explanations is more appropriate for different leadtimes and under different regimes using data from an operational ensemble run during summer 2023 over the UK. Using a novel extension of the Fractions Skill Score, we show that the similarity between precipitation patterns in the convective and driving ensembles depends on whether the initial conditions/model dynamics or the boundary conditions are dominating the evolution of each convective-scale member. When the initial conditions are responsible for the evolution of each member, the similarity to the corresponding driving member decays exponentially with leadtime, and the convective-scale ensemble has a better spread-skill relationship. When the boundary conditions are responsible for the evolution of each member, the similarity to the corresponding driving member is approximately constant in leadtime, and the spread-skill relationship is similarly poor for each ensemble. We also find a strong regime dependence to the similarity of precipitation patterns between the two ensembles, with pressure patterns more associated with convective activity showing less similarity and a worse spread-skill relationship than other regimes

This work may also have operational use as a tool to signal periods when the driving and nested forecasts diverge, thereby improving the efficiency of guidance production. Additionally, understanding the behaviour of these driving and nested ensembles is especially important given recent experiments of hectometric, turbulence-scale ensembles which require multiple layers of forecast nesting.