基于“深度学习”的台风海域上层海洋热含量准实时估算方法研究

It is the urgent needs of national disasters prevention and mitigation to strengthen the understanding of the interaction between ocean and typhoon as well as to improve the level of the typhoon forecast. However, we can’t obtain spatial and temporal variation of the upper ocean heat content in near real-time because of the observation data insufficiency in the upper ocean under the influence of typhoon, which becomes a bottleneck in the typhoon research. In this study, we will select Northwest Pacific Ocean as study area and use data-driven approach. Deep learning method in the artificial intelligence will be introduced into the estimation of upper ocean heat content. Argo profile data and satellite altimeter remote sensing data will be extracted rapidly and integrated in near real-time. Then deep nonlinear network will be built to train the massive observation data in a depth way with layer by layer, in order to fuse features from multi-source data. We will adjust model parameters dynamically by using particle swarm optimization when training, to improve the generalization ability of the model in different waters. We aim to obtain temperature field and the heat content of the upper ocean in different depths under typhoons. On this basis, the impact of the upper ocean heat content on typhoon intensity change will be explored and analyzed. The completion of this study will provide a better understanding of the interaction between the upper ocean and typhoon, as well as improve the accuracy of the typhoon forecast.

加强海洋与台风相互作用的认识，提高台风预报水平是国家防灾减灾的迫切需求。但是由于台风海域上层海洋观测资料在时空尺度上的不足，无法准实时获取上层海洋热含量的时空变异情况，已成为制约台风研究的一个瓶颈。本项目拟以西北太平洋台风源地海域为主要研究海区，借助数据驱动技术，将人工智能中的“深度学习”方法引入到上层海洋热含量准实时估算方法研究中。通过快速提取和准实时集成台风海域Argo剖面浮标观测资料和卫星高度计遥感资料，构建深层非线性网络结构对海量观测数据进行分层和深度训练，融合多源数据特征，并通过粒子群算法在模型训练过程中根据新的数据源动态调整模型参数，提高模型在不同海域的泛化能力。最终实现准实时获取台风下垫面上层海洋不同深度处的温度场和热含量，并初步探讨分析上层海洋热含量对台风强度变化的影响。本项目将进一步加强对上层海洋与台风相互作用的认识，提高台风预测预报的精度。

台风海域上层海洋热含量的准实时获取，对于加强海洋与台风相互作用的认识，提高台风预报水平十分重要。本项目基于“深度学习”开展西北太平洋台风海域上层海洋热含量的准实时估算研究。具体内容包括：（1）台风海域海量观测资料高效、统一集成与存储机制；（2）台风海域实时海洋环境基础资料的快速查询以及热含量估算研究；（3）探讨分析热含量对台风强度变化的影响。项目构建了一个面向海量海洋环境数据的分布式数据库，实现了对大规模异构Argo数据、台风数据及其他海洋环境数据的高效管理；研制了台风海域实时海洋环境智能服务系统，实现了对台风路径附近Argo剖面浮标的快速查询、浮标位置的实时监控、和热含量等衍生产品的快速制作等功能。西北太平洋台风海域实时海洋环境智能服务平台可为国内相关研究项目及海洋、气象服务部门提供丰富的台风海域实时海洋环境数据与信息服务，进一步促进Argo资料在台风基础研究和业务化预测预报中的广泛应用。