3:00 PM - 3:15 PM
[MIS19-06] Prediction Method for Heavy Rainfall Using High Spatial and Temporal Resolution Radar Observation Data, CNN and LSTM
Keywords:Phased Array Radar, Rainfall Forecast, Deep Learning
In recent years, severe events that rapidly develop cumulonimbus clouds and bring torrential rains with gusty winds and lightning strikes have been increasing in Japan, and the damage caused by such events has also been increasing. Therefore, real-time forecasting of heavy rainfall disasters is becoming increasingly important.
Therefore, a dual-polarization phased-array weather radar (Multi Parameter- Phased Array Weather Radar: MP-PAWR) has been developed, which can perform higher-density observations in 30 seconds. However, precipitation forecasting with MP-PAWR using huge amounts of data is computationally expensive, and real-time performance may be impaired. Therefore, a rainfall forecasting method using Convolutional Neural Network (CNN) and Long-Short Term Memory (LSTM) network was proposed.
In this study, we propose several methods based on the proposed method, and compare the discriminant results of each method and study the usefulness of each method for predicting heavy rainfall.
Therefore, a dual-polarization phased-array weather radar (Multi Parameter- Phased Array Weather Radar: MP-PAWR) has been developed, which can perform higher-density observations in 30 seconds. However, precipitation forecasting with MP-PAWR using huge amounts of data is computationally expensive, and real-time performance may be impaired. Therefore, a rainfall forecasting method using Convolutional Neural Network (CNN) and Long-Short Term Memory (LSTM) network was proposed.
In this study, we propose several methods based on the proposed method, and compare the discriminant results of each method and study the usefulness of each method for predicting heavy rainfall.