Understanding and Enhancing Housing Flood Resilience: a data-driven approach

Abstract

Flooding ranks among the most destructive natural disasters posing serious risks to housing, especially in flood-prone areas. Despite the efforts made to date, the weakness in the resilience of housing in communities has sustained consequential losses to flood. The situation is exacerbated by the increasing frequency and severity of floods resulting from climate change, urbanisation, and infrastructure development. These circumstances have made the need for improved understanding, assessment, and the deployment of effective flood resilience strategies more critical than ever. This research introduces a data-driven approach aimed at strengthening understanding, assessing, and informing decisions towards enhancing the flood resilience of housing in communities, utilising Bayesian network models. Before developing these models, challenges in housing data were identified across the stages of the data life cycle, that is, acquisition, classification, management, and utilisation. These challenges were identified through semi-structured interviews, with 12 carefully selected professionals. The strategies to overcome these challenges were also explored, through the interviews. Thereafter, data were collected from 18 zones across three United Kingdom regions of Manchester, Cumbria, and York, representing high, medium, and low flood risk levels in each location. The originality of this research lies in its systematic use of Principal Component Analysis (PCA) to identify influential resilience variables for each flood risk level, and Pearson correlation analysis to explore interrelationships among these variables, providing a structured, data-driven foundation for resilience modelling. The application of nine Bayesian network models developed using GeNIe software establishes a dynamic resilience forecasting approach.
The results reveal distinct patterns of resilience across different flood risk levels. In high-risk zones, lower levels of rapidity, redundancy, robustness, and resourcefulness were identified, highlighting the need for specific interventions. Low redundancy and robustness were found to be particularly significant in high-risk areas, indicating these zones have higher vulnerability to prolonged damage and slower recovery. Medium and low-risk zones demonstrated more balanced resilience profiles though areas for improvement were still noted. Validation of the models was conducted through scenario analysis, which confirmed the models’ ability to forecast resilience accurately across various conditions. This study’s primary contribution is the development of Bayesian models, which serve as a flexible tool for real-time flood resilience assessment and can be adapted to other regions. The significance lies in its potential to transform resilience planning for housing by empowering stakeholders with a novel data-driven approach providing insights into different attributes and needs regarding housing resilience in different zones. This is crucial for informed proactive decision-making concerning risk potentials mitigation judgements and accelerating recovery processes. This research offers a transferable methodology and adaptable models which is a significant contribution to the knowledge base for improving flood resilience globally, this sets a new standard for data integration in resilience assessment.

Date of Award	26 Jun 2025
Original language	English
Awarding Institution	Northumbria University
Supervisor	Onaopepo Adeniyi (Supervisor) & Niraj Thurairajah (Supervisor)