Assessing the Predictability of Mesoscale Moist Airflow Using the Ensemble Dimension

This paper explores the characteristics of the ensemble dimension (E dimension), which quantifies the effective number of dominant directions of ensemble perturbations, in assessing the predictability of mesoscale moist airflow. Focusing on equivalent potential temperature at 950 hPa, the E dimension was calculated over a 200 km × 200 km local domain centered at each grid point using an operational regional ensemble prediction system. Statistical analysis revealed that high values of the E dimension corresponded to the tail of the joint probability distribution between the ensemble spread and the E dimension, where the ensemble spread tended to underestimate the root mean square errors of the ensemble mean forecast. Despite the underestimation of error amplitude, the error structure was well captured by the ensemble in that distribution tail. Such regions were often observed over the ocean south of Japan, a source of moist subtropical airflow. Case studies supported these findings, showing high E dimension values over areas of heavy rain in western Japan, where the deterministic forecast failed. Although the magnitude of meso-α scale errors was not captured, their structure was represented by the ensemble subspace. These results highlight the potential of E dimension as a diagnostic tool for identifying uncertainty in error structures. This study also suggests possible applications for selecting better ensemble members for moist airflow and precipitation through the E dimension. In summary, incorporating the E dimension provides forecasters with a novel perspective on ensemble products and highlights its potential contribution to the improvement of operational forecasting.