Development of AI-Based Short-Term Heavy Rain Damage Prediction 1SP-Model
By Jun-hak Lee, et al.
In order to reduce disasters such as heavy rains, floods, and urban flooding caused by climate change, it is important to determine in advance whether disasters caused by natural disasters occur. The special weather warning operated by the Korea Meteorological Administration are largely divided into two categories: one is heavy rain advisory and the other is heavy rain warnings. However, the heavy rain special warning, which applies consistent standards across the country, cannot reduce damage caused by disasters in advance. Therefore, it is necessary to study the differentiated hazard-triggering rainfall setting in consideration of regional heavy raindamage characteristics and rainfall characteristics. The area to be studied was selected as the Gyeonggi-do area where heavy rain damage occurred most frequently. In addition, the heavy rain special warning standards were reset in consideration of regional characteristics of each region using rainfall data and heavy rain damage data.
In this study, a model for predicting rainfall was developed using long short-term memory (LSTM) and deep neural network (DNN) models. As a result of evaluating predictive power by model, the DNN model showed the best predictive. In order to predict the occurrence of heavy rain damage, an optimal heavy rain damage classification prediction model was developed using hazard-triggering rainfall and predicted rainfall. As a classification model, two models, a decision tree (DT) model and a random forest (RF) model, were applied and compared. As a result of the classification model evaluation, the random forest model had the best predictive performance. Through this, if the 1SP-Model developed in this study is used, it is judged that appropriate preparation can be made according to the risk stage of heavy rain damage before a disaster occurs.
Key Words : Disaster Management, Heavy Rainfall, Damage Prediction, Heavy Rain Special Warning Standards
Development of AI-Based Short-Term Heavy Rain Damage Prediction 1SP-Model
By Jun-hak Lee, et al.
In order to reduce disasters such as heavy rains, floods, and urban flooding caused by climate change, it is important to determine in advance whether disasters caused by natural disasters occur. The special weather warning operated by the Korea Meteorological Administration are largely divided into two categories: one is heavy rain advisory and the other is heavy rain warnings. However, the heavy rain special warning, which applies consistent standards across the country, cannot reduce damage caused by disasters in advance. Therefore, it is necessary to study the differentiated hazard-triggering rainfall setting in consideration of regional heavy raindamage characteristics and rainfall characteristics. The area to be studied was selected as the Gyeonggi-do area where heavy rain damage occurred most frequently. In addition, the heavy rain special warning standards were reset in consideration of regional characteristics of each region using rainfall data and heavy rain damage data.
In this study, a model for predicting rainfall was developed using long short-term memory (LSTM) and deep neural network (DNN) models. As a result of evaluating predictive power by model, the DNN model showed the best predictive. In order to predict the occurrence of heavy rain damage, an optimal heavy rain damage classification prediction model was developed using hazard-triggering rainfall and predicted rainfall. As a classification model, two models, a decision tree (DT) model and a random forest (RF) model, were applied and compared. As a result of the classification model evaluation, the random forest model had the best predictive performance. Through this, if the 1SP-Model developed in this study is used, it is judged that appropriate preparation can be made according to the risk stage of heavy rain damage before a disaster occurs.
Key Words : Disaster Management, Heavy Rainfall, Damage Prediction, Heavy Rain Special Warning Standards