气象   2019, Vol. 45 Issue (4): 445-457.  DOI: 10.7519/j.issn.1000-0526.2019.04.001

### 引用本文 [复制中英文]

[复制中文]
JIN Ronghua, DAI Kan, ZHAO Ruixia, et al, 2019. Progress and Challenge of Seamless Fine Gridded Weather Forecasting Technology in China[J]. Meteorological Monthly, 45(4): 445-457. DOI: 10.7519/j.issn.1000-0526.2019.04.001.
[复制英文]

### 文章历史

2018年11月30日收稿
2019年3月5日收修定稿

Progress and Challenge of Seamless Fine Gridded Weather Forecasting Technology in China
JIN Ronghua, DAI Kan, ZHAO Ruixia, CAO Yong, XUE Feng, LIU Couhua, ZHAO Shengrong, LI Yong, WEI Qing
National Meteorological Centre, Beijing 100081
Abstract: This paper reviews the development of the technology for seamless fine gridded weather forecasting in China since 2014. And the key technical difficulties in the future development are analyzed. It is pointed out that the high spatio-temporal resolution observations capturing the fine structure of weather systems, the analysis products by multi-source data fusion, the real-time rapid updating assimilation and prediction system, the high resolution regional model providing short-time and short-term weather prediction, the global numerical forecast model providing 10 days' weather forecasting, and the ocean-atmosphere coupled ensemble prediction system providing 46 days' weather prediction, have jointly established the premise and foundation of the seamless gridded weather forecasts. After nearly 5 years' exploration and constant efforts, the technology system of seamless fine gridded forecasting with different temporal resolutions has been established. The high-frequency lagrangian extrapolation skills are used for 0-4 h forecasting based on GRAPES-Meso model forecast products and radar data over China. For the 4 h to 30 d lead-time forecasting, it mainly depends on the downscaling, error correction, model output statistics and post-processing methodologies based on regional and global models of different spatio-temporal resolutions to improve forecast skills and resolution. At the same time, automatic and intelligent interactive forecasting platform is developed to meet the demand of combining efficient objective forecasting with forecas-ters' subjective intelligence. In order to assess and track the performance of high resolution gridded forecasting, a spatial analysis verification method based on gridded observation data is developed. It is also stressed that the future gridded forecasting technology system should be able to reflect the latest technology development including the artificial intelligence application, more advanced statistical post-processing skills, key technics for consistency forecasting and unified complete technical architecture and standards.
Key words: gridded weather forecasting    technology development    technical framework    gridded forecast editor    verification methodology    technical difficulty

* 2018年全国智能网格预报业务工作部署会议报告。

* 2018年成都智能网格预报和实况数据分析业务研讨会议交流报告“全国智能网格实况分析产品研制与评估应用”。

1 主要技术进展 1.1 无缝隙精细化网格预报技术框架

 图 1 无缝隙精细化网格预报技术框架 Fig. 1 Technological framework of seamless fine gridded weather forecasting

1.2 不同预报时效的客观后处理技术 1.2.1 0~24 h网格客观预报技术 1.2.1.1 0~4 h预报时效

 图 2 2018年7月1日00时至9月30日23时0~4 h时效的外推雷达回波TS评分对比 Fig. 2 Comparison of extrapolated radar echo TS scores from 00:00 BT 1 July to 23:00 BT 30 September 2018

* 2018年10月全国气象台长会议交流报告《大数据与人工智能技术在天气预报中的应用》。

1.2.1.2 4~24 h预报时效

 ${x_c} = F_o^{ - 1}[{F_m}({x_m})]$ (1)

 图 3 2016年4月13日07—08时累积降水量的预报与观测实况对比 (a)GRAPES-Meso模式02时起报的降水预报，(b)05时滚动更新的动态频率订正后的降水预报，其频率分布曲线利用03和04时的实况和预报构建，(c)降水实况 Fig. 3 Comparison of the forecast with corresponding observation of accumulated 1 h precipitation from 07:00 BT to 08:00 BT 13 April 2016 (a) direct output precipitation forecast of GRAPES-Meso model initialed at 02:00 BT; (b) precipitation forecast after adaptive frequency correction updated at 05:00 BT, during the process of the correction, the frequency distribution curve is constructed using the observation and forecast at 03:00 BT and 04:00 BT; (c) observed precipitation
1.2.2 1~10 d网格客观预报技术

1.2.2.1 降水预报

 图 4 2018年1—9月四种定量降水客观预报方法与ECMWF模式直接输出降水的24 h预报时效暴雨(≥50 mm)TS(a)和Bias(b)评分对比 Fig. 4 Comparison of 24 h heavy rain (≥50 mm) TS (a) and Bias (b) scores of four QPF calibration methods and direct output precipitation of ECMWF from January to September 2018

1.2.2.2 其他气象要素预报

1.2.3 10~30 d延伸期网格客观预报技术

1.3 主客观融合的网格预报平台技术

Hoffman et al(2017)指出，利用最先进的探测手段也不可能对大气进行完整观测，同时基于最高级的计算机系统也不能完美预报天气。因此，天气预报是人和技术相互依赖、协同的结果。

 图 5 主客观融合的定量降水预报平台的不同等级工作模式 Fig. 5 Operation modes at different levels for the quantitative precipitation prediction in the subjective and objective forecasting blending platform
1.4 高时空分辨网格预报产品的检验评估技术

1.4.1 传统城镇预报检验技术的沿用

1.4.2 精细网格预报检验的参照实况方案的改进

1.4.3 解决时空精细度变化的检验技术

Gilleland et al(2009)将上述升尺度检验、尺度分离检验、邻域空间检验、基于目标的检验和变形场检验等方法统称为空间检验方法，上述方法侧重于空间上的对比，对天气过程在发生、持续和结束等时间特征的检验则非常薄弱，但对预报产品的用户而言，了解预报在时间上的误差幅度也是非常必要的。为此有必要开展针对精细化网格预报中天气过程的时间分布特征的检验研究。牛若芸等(2018)提出了降水过程的识别方法，未来此类方法可以引入到精细化网格预报的时间特征检验当中。表 1归纳了目前网格预报中的常用检验方法。

2 未来发展面临的关键技术难点

3 结论与展望