Exploring the Intrinsic Predictability Limit of a Localized Convective Rainfall Event near Tokyo, Japan, Using a High-Resolution EnKF System

Abstract Severe weather related to convective clouds can greatly affect society. Accurately predicting such events is challenging due to the small-scale, nonlinear, rapid increase in forecast errors at the convective scale. Such error growth would strictly limit convective-scale predictability, even if nearly perfect initial and boundary conditions and models are available. This study investigates the intrinsic predictability limit of a localized convective rainfall event in Japan using a high-resolution ensemble Kalman filter (EnKF) system that assimilates phased-array weather radar observations. Analyzed ensemble initial conditions are rescaled to reflect a nearly perfect or a practical level of initial-condition uncertainty. High-resolution ensemble forecasts reveal that the intrinsic predictability limit of the target convective precipitation event could be longer than 20 min, while its current practical predictability limit is likely no more than 10 min. This difference indicates significant potential for improving forecast accuracy in the current numerical weather prediction system. Further ensemble forecasts indicate that reducing initial-condition uncertainty in wind variables could be more beneficial for improving forecast accuracy than doing so in precipitation-related variables. Significance Statement While it is known that predicting localized torrential precipitation related to convective clouds is challenging, our understanding of their predictability limits is still limited. This study investigates the predictability limits of a localized precipitation event through high-resolution simulations. We assessed the intrinsic predictability limit for the target event under idealized initial conditions with minimal uncertainty. We found that the intrinsic predictability limit could be much longer than the practical predictability limit. Our results indicate that reducing initial-condition uncertainty in wind could be more effective for improving forecast accuracy than doing so in precipitation-related variables.