Analysis of Major Rainfall Factors Affecting Inundation Based on Observed Rainfall and Random Forest

Abstract

It is important to analyze urban floods according to uncertain rainfall patterns. This study was carried out to determine key statistical characteristics of rainfall affecting the occurrences of flooding by using observed rainfall, one- and two-dimensional hydraulic analysis models, and random forests. The target area selected was the Centum City area of Busan, where flooding occurs frequently with heavy rainfall. For analysis, the rainfall data from the Automatic Weather System (AWS) and radar observatory in the entire city of Busan were collected. Using the results from the one- and two-dimensional numerical models, the occurrence of flooding was determined according to the observed rainfall events. The random forest was used to classify the presence or absence of flooding according to the statistical characteristics of rainfall, and the importance evaluation function was applied to determine the impact of each statistical factor. The three main statistical factors of rainfall with the greatest impact on the occurrence of flooding were analyzed as the 1-hr maximum rainfall, the average rainfall, and the skewness. On the other hand, the three main factors affecting the maximum flood depth were the 3-hrs maximum rainfall, the 1-hr maximum rainfall, and the total rainfall. It is judged that the results of this study will be useful in suggesting quantitative urban flood forecasting standards when rainfall is examined in real-time by using various observation equipment.