Investigation of the medium-range forecast errors for the extreme rainfall event in North China during July 19–20, 2016

The extreme precipitation event that occurred in North China during July 19–20, 2016, resulted in serious casualties and property loss. The persistence of a strong cyclone originating in the Huang-Huai region led to this event. The cyclone center predicted by almost all operational numerical models tended to the south or east compared to the analysis, and the predicted intensity was weaker beyond the 4-d lead-time. The combination of these two factors meant that the medium-range forecasts for North China did not warn of heavy rain. Ensemble sensitivity analysis (ESA) and comparisons between member subgroups can be used to diagnose the potential sources and evolution of forecast errors. Here, the analysis is based on the 51-member European Centre for Medium-Range Forecast (ECMWF) Ensemble Prediction System initialized 5 days prior to the cyclone′s maturity. The ESA calculates the correlation coefficients between the principal components of the leading empirical orthogonal functions on the ensemble sea level pressure forecast of the cyclone and the upper circulation and finds the high correlation (i.e. sensitive) areas. This method can identify sensitive weather systems at earlier lead times and is capable of investigating uncertainties in the cyclone intensity and its track separately, allowing the potential error sources to be distinguished. The results of the ESA indicate that the uncertainty in the cyclone intensity forecast, which explains 54.4% of the variance, is related to the strength of two sensitive systems: the short wave that separated from the deep trough to the west of Xinjiang and the height field over regions south of the Yangtze River. In particular, the uncertainty in the east-west component of the cyclone position has a strong connection to the phase of these sensitive systems. Two groups are selected for further analysis: an accurate group composed of the eight members that had the smallest errors associated with cyclone track, intensity, and precipitation distribution; and a failed group consisting of the eight members that predicted the southernmost and weakest cyclone. The composite normalized differences between these two subgroups are employed to investigate the dynamical processes associated with forecast errors. Three significant difference areas (SDAs) associated with the sensitive systems were identified, including the front and rear positive SDAs of the potential vorticity disturbance (PVD) that split from the deep cyclone over West Siberia, and the negative SDA to the west of the PVD that originated from Western Inner Mongolia. These SDAs can be traced back to the model initialization or early lead times, which means that initializing and short-lead errors have important influences on the model′s medium-range prediction. The original SDAs display a maximum at the upper troposphere and present a positive-negative interval array; these SDAs tend to propagate along the high PV gradient and spread out to the lower troposphere. To diagnose the physical mechanism by which these errors (i.e. difference areas) influenced the cyclone development in the model, statistical PV inversion is applied to derive the wind field associated with the PV difference between the accurate and failed subgroups. The results of this analysis show that the two positive SDAs accompanying the split PVD affect the cyclone′s formation intensity, its track, and the strength of the downstream ridge. The negative SDA, after arriving at the lower-middle reaches of the Yangtze River, affects the strength of the southerly warm, moist airflow during the cyclogenesis. Overall, the members of the accurate group present more intensive upper PVD and carry the warm, wet air further north, which is conducive to the rapid increase in strength and the more northerly track of the cyclone. In addition, the reasons for large forecast uncertainties are discussed. These may relate to the quality of data assimilation in the warm conveyor belt of the West Siberia deep cyclone, imperfect model parameterizations, and the complex topography of the Qinghai-Tibet Plateau. Finally, the above analysis shows that the ESA and the subgroup difference analysis can complement each other. The former can easily identify the upstream sensitive weather systems that may accompany forecast errors, while the latter can be applied to further diagnose the dynamic processes and mechanisms at play.