Statistical Mapping and Modelling of Urban Flood Using Social Media Data
Yao Li is a PhD student in the Department of Earth Observation Science. (Co)Promotors are prof.dr. A. Stein and dr. F.B. Osei from the Faculty Geo-information and Earth Observation (ITC).
Urban flooding poses significant challenges to cities worldwide, leading to substantial economic losses, infrastructure damage, and loss of life. Rapid urbanization and climate change have intensified these issues, highlighting the urgent need for effective flood mapping and management strategies. This thesis develops a comprehensive framework for urban flood mapping and management. It addresses flood susceptibility, prediction of their intensity, coupling of 1D-2D flooding and simulation models, and targeted mitigation measures across different scales. The methods, applied to the cities of Chengdu and Haining in China, integrate novel data sources, social media data with advanced machine learning models, spatial statistical methods, and hydrodynamic simulations. They offer new insights into urban flooding in rapidly urbanizing regions.
The first chapter is an introduction to the research and offers its background. The second chapter explores the potential of social media data as a novel, low-cost, and real-time source for urban flood mapping. Using environmental factors, a naïve Bayes model was developed to assess flood susceptibility, achieving high accuracy (0.95) and identifying high flood-susceptibility areas in Chengdu. The findings highlight the importance of vegetation density in flood resilience while social media data can effectively fill the gaps in traditional flood monitoring systems.
The third chapter presents a Log-Gaussian Cox Process (LGCP) model that predicts urban flood intensity. It incorporates fixed environmental effects and spatial random effects. This spatial statistical model captures unexplained spatial variability, offering more accurate predictions than traditional deterministic models. Applied to Chengdu, the results identified the central region as a flood hotspot with significant spatial random effects, emphasizing the value of spatial statistical models for understanding the complex spatial dynamics of urban flooding.
The fourth chapter presents a 1D-2D coupled model to simulate urban inundation, combining the 1D Storm Water Management Model (SWMM) with a 2D diffusion-based physical model. This coupling addresses the limitations of traditional 1D models by predicting more detailed spatial inundation distributions and depths. Applied to Haining, the model identified severe inundation areas and provided actionable insights for early warning and management. By combining hydrodynamic models with geostatistical models, the results demonstrate the advantages of integrating physical and statistical models to address uncertainties in flood simulations.
The fifth chapter analyzes the influence of landscape patterns and topographic factors on urban flooding using a multi-scale stepwise regression model. Key drivers, such as impervious surface coverage and elevation gradients, were identified, highlighting the need for spatially heterogeneous and scale-specific planning approaches. These insights inform the design of targeted flood mitigation strategies that align with urban development goals.
The last chapter summarizes the key findings and places them in a broader scientific and societal context to highlight future directions.
In summary, this thesis demonstrates the potential of integrating social media data with advanced spatial statistical and hydrodynamic models to improve urban flood mapping and management. It offers scalable solutions for cities facing similar challenges globally and has a broad application potential for addressing other natural hazards, such as landslides, earthquakes, and wildfires.
