Modelling the frequency of extreme event forecasts for the management of funds for humanitarian action

基于预测的融资（Forecast-based Financing，简称FbF）是一项人道主义举措，它在灾害发生前根据预测发放资金，以支持人道主义行动，而不是依赖于灾害应对。该倡议发展迅速，目前包括若干系统性筹资机制，如红十字会与红新月会国际联合会（红十字与红新月联会）救灾紧急基金。最近捐助者对联邦储备基金的投资有所扩大，因此有必要探讨管理此类基金破产风险的备选办法（例如保险产品或巨灾债券），成功管理FbF融资机制和类似基金需要了解预期提款情况（即满足预测触发条件和发放资金的频率）。目前，预期缩编是基于以下假设：（例如）10年重现期事件发生概率为50%的条件平均每5年满足一次。本说明书忽略了以下问题：- 有一个有限的样本，从10年回报期的定义，因此它可能会发生更多或更少的频率比预期-事件将在空间和时间上聚集，导致意外低或高提取基金-概率预测是不完美的，将高估或低估事件发生的可能性，而任何偏差或噪音也会因预测变量、提前时间和事件的严重程度而变化，因此，本研究工作旨在探索预测触发支付的频率以及支付频率的不确定性有多大。该研究项目将使用欧洲中期天气预报中心制作的预报档案，例如重新预报，重点是具体的预报产品，包括全球洪水意识系统（www.globalfloods.eu），以探索目前存在的一些预报触发因素（例如乌干达和孟加拉国的洪水事件），以及极端预报指数。作为该项目的一部分，研究生将有机会在欧洲中期天气预报中心进行为期3个月的实习，以制定识别和表征事件预测频率偏差的指标。通过跟踪ECMWF工作人员的业务“每日报告”的职责，学生将获得一个深刻的理解的情况下，在事件预测频率的偏差可能是有问题的。他们将与预测部门的主要研究人员合作，制定一项战略，以诊断这种偏见的原因，以支持模型开发，并将这种方法应用于选定的案例研究。学生将与红十字与红新月联会和红十字与红新月气候中心合作，提供有关触发频率的信息，这些信息将被纳入“DREF的FbA”早期行动协议的制定中，从而产生宝贵的现实影响。

Forecast-based Financing (FbF) is a humanitarian initiative which releases funds on the basis of a forecast to support humanitarian actions before a disaster strikes, rather than relying on disaster response. The initiative has grown rapidly and now consists of several systematic financing mechanisms, such as the International Federation of Red Cross and Red Crescent Societies' (IFRC) Disaster Relief Emergency Fund (DREF). The recent expansion in donor-investments in FbF has necessitated the exploration of options for managing the risk of insolvency of such funds (e.g. insurance products or catastrophe bonds), and one such risk relates to how often the forecast triggers are met.Successful management of this FbF financing mechanism and similar funds requires knowledge of the expecteddrawdown (i.e. how frequently the conditions specified in the forecast triggers will be met and funding released).Currently, the expected drawdown is based on the assumption that (for example) the condition of a 50%probability of a 1 in 10 year return period event would be met on average once in every 5 years. This assumptionoverlooks the following issues: - There is a limited sample from which the 1 in 10 year return period is defined, and so it could occur more or less frequently than expected- Events will be clustered in both space and time, leading to unexpectedly low or high drawdowns from the fund- Probabilistic forecasts are not perfect and will overpredict or underpredict the likelihood of an event occurring, and any bias or noise will also vary by forecast variable, lead time and severity of eventTherefore, this research work aims to explore how often the forecasts will trigger a payout, and how muchuncertainty there is in the frequency of payouts. The research project will use archives of forecasts produced by the European Centre of Medium-Range Weather Forecasts such as re-forecasts, focusing on specific forecast products including the Global Flood Awareness System (www.globalfloods.eu) to explore some of the forecast triggers currently in place (such as for flood events in Uganda and Bangladesh), and also the Extreme Forecast Index. As a part of this project, the research student will have the opportunity to undertake a 3-month placement at the European Centre for Medium Range Weather Forecasting - to develop metrics for identifying and characterizing biases in event forecast frequency. By shadowing ECMWF staff on operational 'daily report' duty the student will gain a deep understanding of the circumstances where biases in event forecast frequency might be problematic. They will work together with leading researchers in the Forecast Department to develop a strategy for diagnosing the causes of such biases to support model development, applying this approach to selected case studies. Working with the IFRC and the Red Cross Red Crescent Climate Centre, the student will offer information about the frequency of triggers, which will be incorporated into the development of Early Action Protocols for the 'FbA by the DREF', thereby having a valuable real-world impact.