Extreme rainfall forecasting: new statistical simulation and Big Data methods for making sense of rainfall radar and rain gauges

降雨雷达在城市和小型集水区实时进行洪水预报和风险评估方面的巨大潜力尚未实现，因为在全分辨率下观测高降雨强度和数据量时存在不确定性。传统的雷达校准程序只能部分地考虑到强降雨造成的波束衰减，因此使产品在其最具潜力的应用中失效。纽卡斯尔安装了一种新的高分辨率降雨雷达，用于洪水风险研究，并与遍布城市和周边地区的密集遥测雨量计网络相结合。该项目将开发大数据和统计模拟方法，以有效校准和使用雷达和雨量计数据，实时和长期评估洪水风险。此外，传统的校准方法试图在最小误差意义上提供最佳估计，从而导致方差减小。雷达图像通常被解释为两次扫描之间的降水积分，但实际上它们是快照。我们将根据雷达测量的正确特征来解释它们。将利用时空条件随机模拟的新进展，开发一种新的数据密集型方法，并将其应用于新型降雨雷达。风险-雷达数据提供洪水灾害实时发生的信息，档案允许估计极端情况。通过结合来自城市天文台的降雨和洪水对交通和建筑物的影响和损害的其他信息，可以预测和回顾性地分析洪水对人类和基础设施的风险：与传统的天气和洪水预报完全不同，风险将用于指导和更好地定位校准和预报程序。缓解——风险评估和预测将使该地区的利益攸关方（如Northumbrian Water Ltd、NEXUS transport和University Estates Services）能够更好地对对流风暴洪水进行预警和调整。大数据——将从雷达和遥测雨量计中生成和分析高容量和高速率的数据。到项目结束时（2016年6月开始），将有5年的数据可供分析。然而，用于校准和不确定度估计的蒙特卡罗集合统计模拟方法将产生数量级更高的数据量（TB）。高时空分辨率的蒙特卡罗集成仿真产生了大量需要分析的数据。

The huge potential for rainfall radar for flood forecasts and risk assessments in real time for cities and small catchments has not yet been realised due to uncertainties in observing high rainfall intensities and data volumes when used at full resolution. Conventional radar calibration procedures only partly account for do not account for beam attenuation by intense rainfall, so invalidating the product in its use for the very application where it has most potential. A new high resolution rainfall radar has been installed in Newcastle for use in flood risk studies in combination with a dense telemetered rain gauge network across the city and surrounding area. This project will develop big data and statistical simulation approaches to calibrating and using the radar and rain gauge data effectively for real time and long term assessments of flood risk. Further, conventional calibration approaches attempt to provide a best estimate in the minimum error sense, thus leading to a variance reduction. The radar images are usually interpreted as integrals of precipitation over the time between two scans, however in reality they are snapshots. We will to interpret radar measurements according to their correct characteristics. A new data-intensive approach will be developed using new advances in space-time conditional stochastic simulation and applied to the new rainfall radar. Risk - the radar data provide information on the occurrence in real time of flood hazard and the archive allows estimation of extremes. By combination with other information on impacts and damages from rainfall and flooding on e.g. transport and buildings from the Urban Observatory, flood risk to people and infrastructure can be analysed both predictively and retrospectively: in a radical departure from conventional weather and flood forecasting, risk will be used to guide and better target calibration and forecasting procedures. Mitigation - risk assessment and forecasting will allow better warning and adaptation of operations by stakeholders in the region (e.g. Northumbrian Water Ltd, NEXUS transport and University Estates Services) who have significant exposure to convective storm flooding. Big Data - high volume and rate data will be generated and analysed from the radar and telemetered rain gauges. 5 years of data will be available for analysis by the end of the project (started June 2016). However, orders of magnitude higher data volumes (TB) will be generated by the Monte Carlo ensemble statistical simulation methods to be used for calibration and uncertainty estimation. The Monte Carlo ensembles simulation in high space time resolution generates a huge amount of data to be analysed.