Methods and Platform Realization of the National QPF Master Blender

With the development of the weather forecast modernization, the forecasters are facing challenges brought by meteorological data explosion, the increasing demand of the meteorological services as well as the wide use of post-processing techniques. Traditional Quantitative Precipitation Forecast (QPF) routines, which mainly based on the precipitation contours drawn by forecasters can no longer assist forecasters in demonstrating added values at a higher level.

To support the forecasters' central role in the QPF operation, a subjective and objective merging QPF Platform (the QPF Master Blender, Version 1.0, hereafter referred to as "the Blender") has been designed and developed in China. This platform helps forecasters to control of the whole process of digital forecast from the following aspects: selection and integration of multi-source QPF, adjustment and correction of QPF, grid post-processing and service-oriented products generation (Fig. 1).

Fig.1 The flow chart of the QPF Master Blender.

The intelligence of the Blender is secured by the development of a number of key supporting techniques, including multi-model QPF dataset construction techniques, multi-model QPF integration techniques, QPF field adjustment and correction techniques and gridded QPF post-processing techniques. Generally, these techniques provides the following advantages. Firstly, precipitation simulations from all available models are interpolated into unified spatial and time resolution by the multi-model QPF dataset construction techniques. Secondly, the multi-model QPF integration techniques allows forecaster to assign subjective weight to each chosen model result based on subjective weights. Thirdly, multiple algorithms, such as frequency fitting and rain targets moving, are applied to adjust QPF on gridded fields depending on the QPF field adjustment and correction techniques. Fourthly, the Gridded QPF post-processing techniques allow forecasters to perform grid-to-contour/contour-to-grid transformation, which produces the merging QPF by combining objective QPF and subjective corrected QPF with different weights. And then the spatial/time resolution can be improved by applying downscaling and time-split tools to the merging QPF.

The main functions of this QPF platform were implemented based on MICAPS4 (the fourth generation of Meteorological Information Comprehensive Analysis and Process System). The Blender Version 1.0 was released in May 2017 and has been put into operational use in the National Meteorological Center (NMC), China Meteorological Administration (CMA) since then, which has yield improved effectiveness and received good feedback. The quality of the high-resolution grid products from the Blender has been compared with outputs of the ECMWF and T639 operational models through the Threat Score verification of heavy rainfall(daily rainfall over 50mm) during the operational period (May 1st to August 31st, 2017). The results show that forecasters' threat score by using the Blender reach 0.204, which is 22% and 32% higher than 0.167 of ECMWF model and 0.154 of T639, respectively. The improvement is the highest among the same period of the past three years.

It should be pointed out that the improvement ratio of TS compared with model forecast is also related to the climate background and the training of forecasters, but it also shows that the adaptability of the Blender to the new operational routine. With the help of the Blender, forecasters can produce higher resolution QPF products; in addition, the main role of the Blender is to provide an intelligent forecasting tool, as for the forecaster to play the best value, it needs to be stronger from other support. In the future, we are planning to develop more functions for this platform, including the development of numerical model verification tools to support forecasters to make the best judgments and fusion technologies of multi-scale model information.