Probabilistic bias adjustment of remotely sensed quantitative precipitation estimations to improve urban flood risk assessments in southern Canada

点击翻译按钮获取中文摘要

Extreme precipitation events have significant social, economic and environmental consequences over Canada particularly in urban domains. This project, conducted in collaboration with industry-partner the Institute for Catastrophic Loss Reduction (ICLR), will develop a novel probabilistic framework to evaluate, bias-correct and blend recently generated remotely sensed (RS) precipitation products including Multi-satellitE Retrievals for GPM (Global Precipitation Measurement) (IMERG) and Multi-Radar Multi-Sensor (MRMS) to improve the robustness of urban flood risk assessments. We will analyze the intensity and duration of extreme precipitation events and the corresponding biases by comparing RS estimations with in-situ observations. The project will characterize the spatial dependencies of biases through the development of their marginal and joint distributions using Copulas. Monte Carlo samples from the joint distribution will be taken to bias correct RS for areas where in-situ observations do not exist. The probabilistic bias adjusted RS products will be merged based on a Bayesian approach which, combined with the insurance data, will be used for flood risk analysis. We will assess the reliability of the adjusted precipitation estimations by driving a hydrological model using raw and adjusted data and assessing streamflow simulations against observed records. The proposed probabilistic approach is a major step towards robust flood hazard and risk analysis and has considerable socio-economic and environmental benefits for municipalities, insurance industry and other stakeholders in Canada.