Test of the Synthetical Multilevel Analog Forecast Technology in Short-term Rainstorm Prediction

A new synthetical multilevel analog forecast technology (SMAT) is developed to make the analog forecast trial of different pattern rainstorm process in a selected region, including cold front pattern, typhoon pattern, quasi-stationary front pattern, cyclonic pattern, inverted trough pattern etc. The new forecast system's research process, the study results and its application are introduced. "Synthetical" represents the combination of various meteorological elements, combination of large scale weather condition and meso-scale weather condition, combination of static simulation and dynamical process simulation. Multilevel indicates three level forecast flows by which different aspects are described and are embodied in a harmonious body. The basic element field is used to reflect macro-atmospheric circumstance (large scale) similitude, local physical elements are used to reflect local climate trait (meso-scale) similitude, numerical model integral products are used to reflect dynamical process similitude. "The reducing FAR (vacant-forecast rate) " technology and extremum check method are included in SMAT, which are useful in selecting analog terms and optimizing forecast conditions' combinations. Multi-meteorology terms and physical conditions' combination are also included in SMAT, which are useful in each analog level trail. This is a step forward than the former single element analog. The science problem analog criterion alters a lot with different analog elements and ranges and it is pointed out, the following method is used to resolve this problem. Analog degree in a more general view can be described by evaluating various good elements and their combination samples. The key analog range is selected from some possible ranges. In the 3rd level analog process, assimilation numeric products are imported. Moreover, based on double-times rolling forecast, losing-forecast events can be decreased. This is good to improve forecast capability in disastrous weather. The following conclusions can be drawn. The historical testing CSI (forecast successful index) of each pattern is more than 0.4 (some are even more than 0.6), model testing average CSI is 0.37, this is better than other work in the same field. Better indexes can be gotten in the 3rd level forecast than double levels forecast. The selection and operation way of critical analog deviation suggest the idea of "false alarm better than miss hit". This leads to the high false alarm index. The 3rd level forecast based on the model products can be used in reducing the false alarm index. After revising the model continuously, model average forecast ability can be improved. Comparing with other current forecast methods (CSI is about 0.35), a revised model testing average CSI (0.392) is obtained. SMAT is also good at COR (forecast precise rate) and POD (miss hit forecast rate) index. Results show that successful forecast in various rainstorm process is achieved. SMAT model has a stronger forecast capability.