Author(s): Sunmin KIM, Tsuguaki Suzuki, Yasuto TACHIKAWA
Linked Author(s): Sunmin KIM, Yasuto TACHIKAWA
Keywords: Convolutional Neural Network; Rainfall prediction; Atmospheric variables;
Abstract: A rainfall occurrence prediction model is developed based on the convolutional neural network algorithm, which is one of the representative machine learning algorithm in image recognition. As an image data, a spatiotemporal data array is created from the time series of related atmospheric variables from multiple ground gauge observation sites. By feeding the atmospheric data array to the CNN algorithm as an input, the algorithm is trained to classify whether there will be rain in a certain lead time, such as 30 minutes or 60 minutes ahead. The trained model shows promising results with 71% of the detection ratio and 0.40 of critical success index for 30-min of prediction lead time. The high false alarm ratio is a remaining task that should be improved in furtherresearch. This paper illustrates the basic concept of the developed model and the results from modeling testswith variant model structures and input data combinations.
Year: 2019