基于Tetrolet非线性融合红外和水汽通道卫星资料的热带气旋客观定强方法

China is greatly affected by tropical cyclone (TC). The track prediction of TC has been greatly improved recently. Because it is very difficult for the real time observation and quantitative intensity estimation, this results in little improvement in TC intensity prediction. This is also an unsolved important problem of TC domain in the world. Currently, TC intensity is mainly estimated by satellite data. Time resolution of the microwave data from the orbiting satellite is low and the data is easily interfered by strong rainfall. TC intensity is estimated by using stationary satellite data with Dvorak technique in actual operational forecast. But the Dvorak technique is subjective. Although the Dvorak technique is amended for many times, it cannot get a good result for a weak TC. Most of methods, which use the stationary satellite data to estimate the TC intensity, focus on using the brightness temperature of infrared channel data to build intensity estimation model for a TC. But they little consider to combining the brightness temperature with the inner core scale of the TC and other channel data which are strongly related to the TC intensity. This will result in a big error. The high time resolution infrared and water vapor channel data from the stationary satellite are fused in our project. The brightness temperature, inner core scale of the TC and latitude are combined to construct the feature factors. Tetrolet transformation and RVM (relevance vector machine) are used to build the TC intensity estimation model to improve the accuracy. Then the estimated TC intensity is assimilated into numerical prediction model so as to improve the analysis and prediction level for the TC intensity. This will provide an efficient technique to reduce the disaster which is caused by strong TC.

我国受热带气旋（TC）灾害影响严重，尽管近年来我国TC路径预报技术有大幅提升，但因TC强度实时探测和定强难度极大，阻碍了我国TC强度预报技术的发展，这也是国际TC研究领域的难点。目前，TC定强主要基于卫星资料。由于极轨卫星主要依赖的微波资料时间分辨率过低和易受强降水污染，业务上主要基于静止卫星资料和Dvorak技术定强，但Dvorak技术主观性较强，虽几经改进但仍不适用于弱TC。在用静止卫星定强时，大多方法集中于用红外通道亮温资料构造线性定强模型，很少兼顾内核尺度和其他通道等与TC强度密切相关的重要信息，导致误差较大。本项目融合连续高时间分辨率的静止卫星红外和水汽通道资料，综合考虑亮温、内核尺度和纬度等因子，并用Tetrolet变换和RVM等非线性智能技术，尝试改进TC定强精度，并期望用所改进的TC强度改进TC数值预报模式初始场，提高TC强度分析和预报水平，为我国防台减灾提供科学技术支撑。

近年来，国内外各类观测手段和数值预报技术的发展推动了TC路径预报水平的不断提高，但与此同时，TC强度的预报能力却进展非常缓慢。其主要原因一方面是TC定强能力不足，另一方面是对影响TC强度的机制仍不清楚。长期以来，TC定强主要借助于Dvorak方法。目前业务TC定强方法较Dvorak技术有了一定的改进，但仍存在一些问题，第一、已有研究主要集中于利用红外通道资料，对其它通道的资料应用较少；第二、定强时考虑的因素不够全面，未充分考虑与TC强度有关的结构信息（如内核、尺度等），因而仅对某类TC定强有效。第三、TC定强建模时多采用线性模型，即使有少数非线性模型也常常要借助一些主观经验确定模型中的参数。导致目前对TC强度的估计仍主要基于Dvorak技术定性描述，定量能力较弱。因此，我们需要尝试基于静止卫星资料，考虑融合多通道卫星资料并结合更多表征TC强度的特征因子，利用非线性建模工具，研究新的客观方法定量估计TC强度。本项目基于Tetrolet变换域偏微分方程结合广义交叉确认原理的卫星图像去噪算法，能够为后续多通道卫星云图融合提供良好的图像源；利用多尺度几何分析工具融合红外和水汽通道卫星云图，能够提高后续基于融合云图的TC中心定位精度；基于卫星云图的梯度、纹理和结构等信息的TC中心定位方法，定位精度在5km-80km之间；基于静止卫星云图亮温、结构和尺度等信息结合相关向量机（RVM）的热带气旋客观定强算法，定强的平均绝对误差在3m/s-10m/s之间；基于静止红外卫星云图的热带气旋内核表面风场结构的客观估计方法，能够为TC内核表面风场反演提供资料；基于静止红外卫星云图结合深度学习技术热带气旋区域客观检测方法，能够自动TC识别跟踪提供技术支撑。综上所述，本项目完成了融合静止卫星的红外和水汽通道资料，结合TC内核、尺度和中心纬度等特征因子，利用Tetrolet变换和RVM等建模技术，改进TC定强，精度高于现有Dvorak方法。