POSTECH and Kyungpook National University research teams have successfully created a “flood risk map” of South Korea by using artificial intelligence (AI) to predict regional flood risks. This research was recently published in the journal, *Journal of Environmental Management*.
Due to climate change and rapid urbanization, flood-related damages are becoming increasingly severe. The increase in impermeable surfaces, such as concrete roads and buildings, exacerbates the damage even with the same amount of rainfall. Traditionally, the Analytic Hierarchy Process (AHP), which relies heavily on subjective judgments, had been used for predicting flood risks. However, this method is time-consuming and costly, and it is challenging to express the reliability of predictions in numerical terms.
In their study, the research team utilized AI to address these challenges. They analyzed flood damage data recorded by the Ministry of the Interior and Safety over the past 20 years (2002-2021) on a city and county basis. Based on this data, they identified four key factors determining flood risk: “hazard” (the amount of rainfall), “exposure” (population and facilities exposed to risk), “vulnerability” (the degree of susceptibility to damage), and “response capability” (the ability to cope with the risk). These were then fed into the AI models for learning.
Among various AI models, ‘XGBoost’ and ‘Random Forest’ demonstrated a high accuracy of over 77% in predicting flood damage. Interestingly, the two models identified different factors as the most significant risks. XGBoost highlighted the “impermeable surface area,” while Random Forest pointed to “river area” as the largest risk factor.
Nonetheless, both AI models identified major cities like Seoul and Incheon as “high flood risk areas.” These cities have high population density, extensive concrete-paved areas, and significant infrastructure concentrated along rivers, making them more susceptible to damage.
The research is significant in that it allows for the numerical evaluation of “prediction uncertainty” regarding flood risks. Regions commonly identified as high-risk by all AI models can be prioritized in disaster management policies, while areas with conflicting assessments among models can be flagged for further investigation. This assists in devising effective flood countermeasures within a limited budget.
The research team also proposed practical solutions. Given that “impermeable surface area” and “river area” were confirmed as major risk factors through AI analysis, they emphasized the necessity of ecological urban development policies, like increasing green spaces where water can be absorbed into the ground naturally and restricting development around rivers to mitigate flood damage.
The research was supported by the National Research Foundation of Korea’s Science and Engineering Basic Research Support Program and the Hyundai Motor Chung Mong-Koo Foundation.