10B.3 A New Ensemble Simulation Analysis Considering Water Vapor Update History for Line-Shaped Rainband Heavy Rainfall Forecasting

Wednesday, 15 January 2020: 3:30 PM
258A (Boston Convention and Exhibition Center)
Nana Kuroda, Graduate school of Engineering, Kyoto University, Kyoto city, Japan; and K. Yamaguchi and E. Nakakita

In Japan, linear quasi-stationary mesoscale convective systems (MCSs), which are linearly organized by cumulonimbus clouds to produce line-shaped rainbands, are often observed in early summer in the rainy season, and called Baiu. On 5 July 2017, above Asakura City in Fukuoka Prefecture, the line-shaped rainbands brought severe rainfall of 700 mm in 9 hours to cause serious floods and sediment disasters. This disastrous heavy rainfall event is called “2017 Northern Kyushu Heavy Rainfall” (abbreviated as NKHR). To minimize the damage of disasters triggered by such heavy rainfall, it is necessary to improve the accuracy of rainfall prediction. In recent years, some methods of ensemble forecast of MCSs have been studied and developed. The previous study [Yamaguchi, K., Horiike, Y. and Nakakita, E. (2018): Analysis of Predictability and Development Mechanism of Line-shaped Rainbands Heavy Rainfall of 2017 Northern Kyushu Heavy Rainfall, Journal of the Japan Society of Civil Engineers, B1 (water engineering), Vol.74, No.5, pp277-282.] reports that NKHR was the extreme event exceeding the scenario of any present ensemble members. For such a difficult-to-predict event, a new advanced ensemble forecast method is required.
The particular feature of our method is to use not only the latest forecast but also the past ones. This study aims at the quantitative prediction of the occurrence and duration of this heavy rainfall composing of line-shaped rainbands by utilizing “update history,” which is defined as the change of the predicted results of forecast by updating initial conditions. We hypothesize that two distinct patterns of characteristic change exist in ensemble forecast of this difficult-to-predict event when updating values of the initial condition. Pattern 1 denotes that the ensemble mean does not approach true values even when forecast is updated; On the other hand, Pattern 2 is that the dispersion of ensemble does not decrease when forecast is updated.
For NKHR, we conducted an ensemble forecast by taking multiple initial time (00:00, 03:00, 06:00, 09:00, 12:00, 15:00, on July 5th 2017 (JST)). As the numerical weather model, we used CReSS [Tsuboki, K. and Sakakibara, A. (2002): Large-Scale Parallel Computing of Cloud Resolving Storm Simulator, High Performance Computing, Springer, pp. 243-259.]. The initial perturbations were generated by the method of Breeding of Growing Modes, and 33 ensemble members including one control run were made for each initial time. We examined the patterns in the forecast of water vapor, which is regarded as a potential factor for the mechanism of rainfall initiation and development. With these information of water vapor forecast, we could create an ensemble scenario which could forecast heavy rainfall in the amount exceeding the previous scenarios can do and approaching the true values. Then, we examined water vapor mixing ratio at 750m height not only in the area around Asakura, where severe disasters occurred, but also in the upstream area of water vapor inflowing to Asakura.
For Pattern 1, we investigated the time change of the deviation from the ensemble mean to the true values, of which the objective analysis values are obtained by simulations using Meso-Scale Model developed by Japan Meteorological Agency. As a result, the Pattern 1 is clearly observed around Asakura, where the newer forecasts are under-estimated and have more significant differences from the analysis values. However, Pattern 1 is not observed in the upstream area of water vapor because its inflow into the area above Asakura cannot be revealed. This means the prediction of water vapor in the upstream area is better than that of Asakura. On the other hand, for Pattern 2, we used the value subtract the ensemble minimum from the ensemble maximum as a dispersion index expressing the maximum width of prediction. The dispersion index is chosen because we want to include information from a small number of members which predict extreme values. Figure-1.jpg is the spatial distribution of ensemble dispersion of water vapor mixing ratio at 750m height. The figures are arranged horizontally in the order of the forecast time and vertically in the order of the initial time. Pattern 2 is observed both in the area around Asakura and the upstream area of water vapor inflow, where the dispersions of newer forecasts are bigger rather than the same as that of past ones. These two patterns show fairly good agreement with the time duration of heavy rainfall, which lasted from about 12:00 to 21:00.
Although it holds roughly low possibility of prediction using Pattern 2, we also verified the capability of Pattern 2 on real-time quantitative prediction due to a positive correlation between the forecast of inflow water vapor in its upstream area with greater dispersion index in Pattern 2 and the real rainfall intensity in the area around Asakura.
Patterns 1 and 2 are observed and matched with the time duration of our target heavy rainfall event. Our analysis successfully revealed the possibility of utilization of update history of water vapor information in ensemble forecast on predicting the occurrence, duration, and amount of heavy rainfall triggered by line-shaped rainbands. By using the approach of update history in ensemble forecast, we can create new scenarios which could forecast heavy rainfall in the amount exceeding the previous method can, and is convinced to be very useful for disaster prevention.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner