PRECIPITATION PREDICTION BY DEEP LEARNING USING AMeDAS DATA

As heavy rainfall disasters occur frequently, it is important to predict precipitation with high accuracy in order to perform evacuation and dam operation properly before heavy rainfall occurs. Precipitation observed at a point is affected by weather conditions around the point, such as advection of precipitation area and surrounding air temperature. In this study, we tried to predict the rainfall several hours ahead by learning the relationship between the precipitation at an observation point and the surrounding weather conditions using deep learning. Although deep learning model tends to be able to generally predict the precipitation start/end times, it has been confirmed that it is more difficult to predict gusty rainfall than GPV. In addition, by visualizing the weights of the input data, it shows the possibility of estimating the meteorological factors that affect the precipitation.