Radar Research, Results & Data to Decisions - R3D2

R3 D2项目将开发先进的综合数据产品，这些产品来自目前遥感水深图和海流的研究和开发级波浪反演输出。国家海洋和大气管理局开发波浪反演方法已有二十多年，现在能够从静态（陆基）雷达和最近的移动船只获得长达6公里的深度和海流矢量图。对雷达后向散射图像序列进行处理，以提取符合线性波理论的波长、周期和方向。在理想条件下，结果与ADCP测量的深度和海流吻合良好。潜在用户对这些功能作为商业服务的需求不断增长，NOC已与Marlan Maritime Ltd合作提供这一服务。在用户需求所决定的作业条件下，要求越来越多地试图在风力和海况越来越低的情况下作业，以及在雷达成像机制难以探测到足够的波浪特征作为算法输入的地区作业。这不可避免地导致了确定算法何时产生与现实一致的深度和海流估计的挑战-如果用户社区继续采用这些破坏性方法，则需要解决这种情况。NOC的未发表工作先前表明，一系列质量和置信度措施与机器学习元素相结合，有助于稳定和质量控制时间序列，结果作为算法“学习”每个站点的特征。

The R3D2 project will develop advanced synthesised data products from what are currently research and development grade wave inversion outputs of remotely sensed water depth maps and currents.Wave inversion methods have been developed at the NOC for over two decades and are now able to derive depth and current-vector maps over ranges up to 6km from both static (land based) radars, and more recently from moving vessels. Sequences of radar backscatter images are processed to extract wave lengths, periods and directions that are fitted to linear wave theory. Under ideal conditions, results match well with survey depths and currents measured by ADCP. There is a developing demand by potential users for these capabilities to be available as a commercial service and the NOC have partnered with Marlan Maritime Ltd to deliver this. Under the operational conditions dictated by user needs, the requirement has been increasingly to attempt to operate under lower and lower wind and sea states and in areas in which the radar imaging mechanism struggles to detect adequate wave signatures as inputs to the algorithms. This has inevitably led to challenges determining when the algorithms are producing estimates of depth and currents that are consistent with reality - a situation that needs to be addressed if the user community is to continue adopting these disruptive methods.Unpublished work at the NOC has previously shown that a range of quality and confidence measures combined with elements of machine learning help to stabilise and quality control time series of results as the algorithms 'learn' the characteristics of each site.