Even if a numerical weather prediction model is capable of predicting an extreme weather event, several questions remain such as the confidence level of the predicted event and the reliability of the information related to details such as timing, location, and magnitude. In this paper, a method known as Ensemble Anomaly Forecasting, which combines ensemble forecasts with climatology, is introduced and demonstrated by using a case of extremely heavy rain occurring in Beijing on July 21, 2012. The results show that these two questions can be effectively addressed through this method and ensemble forecasts by providing more reliable and consistent information than that provided by a single forecast. Therefore, we strongly recommend that forecasters apply this method in their daily operations to improve their prediction capability of rare high-impact weather events.
The following three aspects are discussed in detail in this study. (1) By comparing a forecast with climatology, the potential rarity of the predicted variable can then be quantitatively measured in terms of standardized anomaly (SA), which normally indicates an extreme event when the departure of a forecast from its climatology mean exceeds three standard deviations. By combining further with ensemble forecasts, the confidence of such an anomaly forecast can also be estimated on the basis of the SA of individual ensemble members, which provides critical information that enables a forecaster to make a more reliable forecast of a potentially rare weather event. A combination of the anomaly and confidence then defines a "societal impact matrix," which can be used to quantitatively measure a forecast's potential impact on society. (2) Because the synoptic scale pattern associated with this heavy rain event in Beijing is quite classical for extreme flooding events, it was a highly predictable event from the large-scale pattern perspective. For example, the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS) model quite successfully predicted a rainfall event of approximately 115 mm over Beijing with a lead time of approximately six days (0600, July 16). However, the detailed information such as rainfall location and intensity were highly variable or uncertain in subsequent GFS forecasts, thus resulting in low predictability. Such shifting of the model solutions from one cycle to another significantly limits the usefulness of a forecast because it is difficult to follow. In contrast, and as demonstrated by this study, ensemble-based—particularly multi-model ensemble-based—ensemble mean and probabilistic forecasts can mitigate some of the issues associated with model shifting by providing more consistent information to greatly increase forecast utility. Additionally, ensemble-based forecasts may extend the practical predictability length. For example, the predictability length of rainfall exceeding 100 mm over or near Beijing can be extended for approximately two days by using THORPEX Interactive Grand Global Ensemble (TIGGE) based probabilistic forecasts as compared to that by a single GFS forecast. (3) If observation or analysis is used instead of forecasts in the calculation of SA, SA can also help to determine the possible causes responsible for an extreme event. In this case, the spatial distribution of the SA reveals that the immediate short-range synoptic cause of the extreme rainfall is the merging of a cold front from the northwest and strong ridge extending from a tropical system in the southeast, which formed a favorable moisture, convergence, and vertical lifting environment for the development and maintenance of meso-and small-scale convective systems. The time evolution of the SAs further reveals that the medium-range background cause is the flow's meridional development to form and maintain a large-amplitude low-high alternating wave train in high latitudes, particularly the development of a blocking system to the northeast of Beijing and a deepening trough to the west, resulting in a strong cold front that enhanced the north-south exchange including the northward advancement of a tropical system.