Property level pluvial flood risk estimation from large-scale flood models - Identifying potential erroneous locations using a geospatial postprocessing algorithm

Abstract

<div> <p><span data-contrast="auto">A distinctive characteristic of 2D flood models used in catastrophic modelling in industries such as re/insurance is that they are often developed for a national, continental or even global scale (hereafter referred to as large-scale models). Traditionally these models were used at portfolio-level loss estimations. However, the increased need for understanding flood risk at a much smaller scale (e.g., at the property level), especially for parametric insurance products demands these models to deliver high-resolution, high-accurate flood inundation mapping. Although technological advancements in remote sensing and cloud computing have made this possible to a certain extent, the underlying models are still large-scale models with inherent uncertainties caused by unaccounted small-scale dynamics in both hydrometeorological and terrain data that are impossible to be captured by these large-scale models. Although this uncertainty exists in all three sources of floods (fluvial, pluvial and coastal), pluvial floods' sensitivity to drainage structures and small-scale dynamics in terrain - both of which are poorly represented in large-scale models - make them the most vulnerable flood type to this uncertainty. </span><span data-ccp-props="{}"> </span></p> </div> <div> <p>We attempt to develop a novel and practical geospatial approach to identify cases where large-scale 2D flood models are likely to underestimate/miss pluvial flood risk at the property level using easy-to-calculate proxies. These proxies are derived from the inundation clusters created using the 2D flood map itself - hence they do not need any additional data. To test the method, we used data collected from FloodFlash’s novel and low-cost flood depth sensors installed at properties across the UK. Initial results show that the method is able to identify more than 75% of the cases where the large-scale flood maps underestimated flood risk at the property level (Hits). This method is currently being improved to reduce the number of false positives by finetuning the selected proxies. </p> </div>