UQ4FM: Uncertainty Quantification for Flood Modelling

目前，英国有640万人以及公路、铁路和电力网络等关键基础设施面临洪水风险，预计到2080年，这一数字将上升至1080万人，并涵盖更多关键资产。2020年国家风险登记册将洪水列为英国面临的最大风险，仅次于流行病和大规模袭击。尽管如此，用于规划、开发和适应目的的常规洪水风险评估使用确定性方法来评估洪水风险，使用基于水动力过程的模型，这些模型计算量很大(运行时间从几小时到几周)。这一既定进程未能承认、量化和捕捉该进程中固有的连锁不确定性，这些不确定性来自各种来源，包括气候情景、流量测量、极值估计和水文模型。低估当前和未来的洪水风险可能导致政府气候变化风险评估(CCRA)所称的“锁定”和工程不足的适应措施，而过高估计可能导致财务上不可行的方案和不适当的开发。洪水分析行业必须迫切地转向承认和量化级联不确定性的概率方法；但这需要尚未开发的算法来捕捉过程中的关键不确定性，并减少与正向不确定性量化(UQ)相关的计算负担。该项目将通过开发洪水淹没模型的新的和定制的不确定性量化算法，并通过展示其在一系列尺度上对当前和未来洪水风险的预测的适用性，在模型链中纳入广泛的不确定性，来加快有力地评估洪水风险不确定性所需的速度。成功将带来洪水分析行业接受向UQ评估的必要过渡所需的步骤变化，从而使英国处于洪水研究和未来应对气候变化适应的前沿。

Currently 6.4 million people, as well as critical infrastructure such as road, rail and power networks, are exposed to flood risk across the UK, and this is expected to rise to 10.8 million people and encompass further critical assets by 2080. The 2020 National Risk Register places flooding behind only pandemics and large-scale attacks as the most significant risks to the UK. Despite this, routine flood risk assessments for planning, development and adaptation purposes use deterministic methods to assess flood hazard, using hydro-dynamic process-based models which are computationally heavy (~hours to ~weeks run time). This established process fails to acknowledge, quantify and capture the cascading uncertainties inherent in the process, which manifest from a wide range of sources including climate scenarios, flow gauging, extreme value estimates and hydrological models. Under estimation of current and future flood hazard could lead to what the Government's Climate Change Risk Assessment (CCRA) terms 'lock-in' and under-engineered adaptation measures, whilst over-estimation could lead to financially non-viable schemes and inappropriate development. The flood analytics industry must urgently move towards probabilistic methods which acknowledge and quantify cascading uncertainties; but this requires yet-to-be developed algorithms which capture the critical uncertainties within the process and reduce the computational burden associated with forward Uncertainty Quantification (UQ). This project will deliver the speed up required to robustly assess flood hazard uncertainty through the development of novel and bespoke uncertainty quantification algorithms for inundation modelling; and by demonstrating their applicability to the prediction of current and future flood hazards at a range of scales, incorporating a wide range of uncertainties in the modelling chain. Success will deliver the step change needed by the flood analytics industry to embrace the necessary transition to UQ assessment, thus placing the UK at the forefront of flooding research, and future proofing climate change adaptation.