基于静止轨道气象卫星云顶属性的快速发展对流监测与多源资料验证研究

This project establishes efficient Geostationary Operational Environment Satellite (GOES) multi-channel infrared data-based Rapid Developing Convective clusters (RDC) monitoring methods as a principal framework, including cloud clusters(CC) identification, the CC tracking technique, the multi-spectral cloud-top properties of RDC , interest fields and the determination of likelihood. The RDC includes two parts: Convective Initiation（CI） and Developing Deep Convective Clusters(DDCC).This study also utilizes daytime visible (VIS) and near-infrared (NIR) reflectance from 0.6 to 3.9 mm, as well as high-resolution visible (HRV) local texture fields and Time changes as new interest fields for understanding various aspects of growing convective clouds in the GOES imagery and capturing the CI information earlier. These interest fields are initially assessed for growing cumulus clouds, with correlation and principal component analyses used to highlight the fields that contain the most unique information for describing principally cloud-top glaciations, as well as the presence of vigorous updrafts. Materials specially Lighting image sounder (LIS) and weather Radar echoes are collected to validate the interest fields prior to RDC events, whose statistical likelihood relation would be classified into different datasets as a reference.CI rate and DDCC rank detection approach based on maximum posterior probability estimation method is proposed. Verification and transmission validations are also proposed by some other data sources (including Microwave Imager ,Precipitation Radar, Cloud Profile Radar, MODIS and Lidar as well) for better understanding micro-physical process of growing convective clouds which includes updraft strength, effective radii of frozen hydrometeors and cloud-top glaciations. In order to reveal various features of physical process at different stage of Convective clusters lifecycle and improve accuracy of determination , a detailed analysis for Convective clusters profiles signatures for growing cumuli at the stage of CI and DDCC would be made.

本课题建立静止轨道气象卫星红外多光谱通道监测快速发展对流（包括对流c初生与发展的深对流）主体框架，主要涉及对流的判识、跟踪技术、快速发展对流的多光谱云顶特征、对流初生的判别问题等内容。研究拓展了白天可见光、近红外以及甚高分辨率反射比构造新快速发展对流判据（如局部纹理特征及其时间变化率）的方法，以深入理解快速发展对流云顶特征与物理过程之间的关系，尝试更早地捕捉到对流初生。根据主成分分析法对判别进行去冗余处理，帅选对云顶冻结、气流上升敏感的判据条件。分类研究快速发展对流案例，将地基雷达等资料与设计出的判据条件进行匹配以构造概率统计模型数据集以及基于贝叶斯最大后验快速发展对流检验方法。引入主被动微波等多源资料建立快速发展对流传递验证方法，深入理解对流云物理属性与构造的判别条件之间的关系。在初生、发展成熟阶段对流云系结构进行精细化分析，提高快速发展对流监测精度从而更好地为临近预报提供参考。

本课题建立静止轨道气象卫星红外多光谱通道监测快速发展对流（包括对流初生与发展的深对流）主体框架，主要涉及对流的判识、跟踪技术、快速发展对流的多光谱云顶特征、对流初生的判别问题等内容。研究对象快速发展对流包括对流初生与仍然活跃的成熟的对流。目前新一代静止轨道气象卫星大中国区域局地产生的对流初生存在高虚警率的问题，很多浅对流并没有发展成为致灾的强对流。为了甄别出能够发展成为强对流的浅对流，本项目根据新一代静止轨道气象卫星风云四号AGRI的资料，发展了新的0-2小时的快速发展对流监测与预警系统。快速发展对流监测系统引入了基于TV-L1范数的光流法以辅助弱小对流单元的跟踪，同时使用了基于监督学习的方法对潜在对流单元进行甄别。通过大中国四个典型区域（青藏高原、华东、华南、东北）的案例分析与样本统计可知，对流初生产品的提前量（相对于雷达回波反照率因子35dBZ），同时对流初生的准确率约为80%，虚警率约为34%。另外，基于Himawari-8与数值预报资料以及随机森林的机器学习方法被用来发展追踪与识别三类不同（弱、中等以及强）的对流性系统。GPM的网格化降水资料被用来标注以及构造不同的对流性强降水数据集，最终形成了对流性可降水预警模型。