AbstractFlooding presents a challenge as a natural hazard, with around 1.1bn people threatened worldwide. Global flood models are useful in managing this risk because of their complete coverage. However, these models allocate flood risk based on national level income disaggregated by population density. Global flood risk maps as a result do not adequetly account for local variations in social vulnerability. The construction of high-resolution social vulnerability is challenging in low and middle income countries due to a lack of reliable household data. Informal settlements are particularly socially vulnerable environments to flooding due to factors related to a lack of legal tenure and poor provision of services. However, there are no comprehensive, global informal settlements maps that could be integrated with existing global flood models.
But informal settlements are an environment clearly identifiable from satellite imagery, a task which has recently been dramatically improved using deep learning in the form of Convolutional Neural Networks (CNNs). I use a CNN to map informal settlements using Slum Dwellers International (SDI) Know Your City as a proxy for social vulnerability. Both a pixel-based and area-based approach are used because the benefits of each method are recognised.
Using Cape Town as a case study, the flood risk map using the informal settlement layer derived from the CNN is contrasted to a traditional population-based flood risk map. The flood inundation layer for each is derived from the Fathom Global Flood Model Version 2. Different methods of spatially allocating flood risk using the informal settlement layer are produced to inform policymakers of the various methods of producing maps which spatially allocate risk to socially vulnerable environments. I also calculate the number of additional informal settlement dwellers demonstrated to be disproportionately at risk by the new flood risk mapping method.
|Date of Award||23 Mar 2021|
|Supervisor||Sean Fox (Supervisor) & Jeff Neal (Supervisor)|