Machine Learning Prediction of Precipitation in Metro Manila, Philippines

It is difficult to predict the occurrence and rain volume of torrential rainfalls, such as guerrilla rain, rain band with typhoon and linear precipitation zone. As heavy rain area is spatially localized and the parent thunderstorm tends to develop within a short time, it makes difficult to accurately predict the occurrence location/time and rain volume. Recently, the machine learning technique is remarkably developed with the improved processing speed of computers and with a huge amount of the data. In addition to this, the application of the machine learning methods to the meteorological fields is intensively tried in the world. Since 2017, we started installing the automatic weather observation system (AWS) named as P-POTEKA in Metro Manila, the Philippines, which is one of the cities suffering from the torrential rainfall and related flood. So far, we installed 35-P-POTEKAs in Metro Manila and continue the weather observations (rain volume, temperature, air pressure, humidity, wind speed, wind direction and solar radiation) with the time resolution of 1 min. In this study, we used both P-POTEKA rain volume data and machine learning model (ConvLSTM: Convolutional Long-Short Term Memory) in order to predict the near future (< 1hour) rain volume and distribution. At the presentation, we will show the results derived from the machine learning prediction of the rain volume and distribution more in detail.