Strategizing Building Resilience: A Big Data Driven Approach to Flood Risk Assessment and Management

The increased prevalence and intensity of flooding, exacerbated by climate change, pose significant risks to the structural integrity of buildings. A pertinent example of this was the 2022 flooding of Gangnam Station in South Korea, resulting in loss of life and amplifying concerns over flood-related damages. It is imperative to proactively assess the flood vulnerability of individual buildings to enhance public safety. This susceptibility is influenced by the building's unique characteristics and geographical location, necessitating their incorporation into flood mitigation strategies. This study endeavors to: (1) ascertain the flood risk of individual building units, and (2) suggest a management strategy to augment the flood resilience of buildings.

Our analysis encompassed 27,438 instances of flood damage in Seoul from 2016 to 2022, correlating this data with detailed building registry information. We categorized buildings into two risk groups—low and high—based on a damage threshold of 3 million won. Employing a range of variables, our study developed a flood risk analysis model utilizing the TabNet classifier, achieving an impressive predictive accuracy of 88%. Key factors in assessing flood risk included the building's function, structural design, height, and floor area ratio, with smaller buildings identified as particularly vulnerable.

The research revealed that flood hazard maps and flood risk maps display differing patterns. In certain areas, high flood probability coincides with low potential damage. This observation has two key implications: First, individuals in high flood probability but low damage areas might be exempt from stringent governmental oversight. Second, there are regions with low flood likelihood outside of government regulation that could still incur significant damage in the event of a flood.

Leveraging the power of machine learning and deep learning, increasingly applied across various fields, this study integrates building attribute data with a plethora of spatial and socio-environmental factors. This integration has facilitated the creation of a comprehensive list of buildings particularly prone to flooding, utilizing public datasets and advanced deep learning techniques. Most identified high-risk buildings are small-scale structures, already under the purview of mandatory inspections by several legislations including the Building Management Act and others related to safety and fire protection. However, buildings classified as safety-vulnerable are not subject to regular inspections under current laws. Given the anticipated increase in flood events due to climate change, it is crucial to establish safety management standards tailored to specific building characteristics to effectively reduce flood-related damages.