Rainfall Nowcasting Based On Ground-Based Cloud Images Using Machine Learning (A Case Study In Majalaya)
Frank Leonard Tansaulu1, Yanuar Rizky Ramadhan1, Muhammad Rais Abdillah2, Edi Riawan2,3, Wendi Harjupa4, Riki Waskito5, Mimid Suamid5, Yadi Mulyadi5, Indra Mustofa5

1Department of Meteorology, Faculty of Earth Sciences and Technology, Institut Teknologi Bandung, Indonesia
2Atmospheric Science Research Group, Faculty of Earth Sciences and Technology, Institut Teknologi Bandung, Indonesia
3Center for Water Resource Development, Institut Teknologi Bandung, Indonesia
4Research Center for Climate and Atmosphere, National Research and Innovation Agency (BRIN), Indonesia
5Garda Caah, Community Volunteers, Majalaya, Indonesia

Abstract

Majalaya is one of the area that often experiences flooding problems due to rain that occurs in the upstream area of the Cirasea Sub-watershed. This causes the local community to independently make manual rain predictions by observing cloud patterns in the upstream areas that have the potential to produce rain and cause flooding in Majalaya. However, manual predictions made by the community have several drawbacks, namely they are very subjective, and less effective because there must always be someone who makes observations. Therefore, this research was conducted to build a model of detection and prediction of the rain events automatically using a machine learning model.

This study uses the Convolutional Neural Network (CNN) method, which is one of the methods in machine learning that aims to extract the features that exist in the image, so as to produce a function model that can determine the output of the image inputted to the model. The model is built with two schemes, namely a detection scheme for a 0 minute time lag and a prediction scheme for a 10, 20, and 30 minute time lag. To determine the model with the best performance, the model that has been built is validated with three cloud cover conditions, namely cloudy, balanced, and sunny, and tested with several test parameters such as accuracy, bias, probability of detection (POD), false alarm ratio (FAR), and threat score (TS).

The results showed that the model with a time lag of 0 minutes, with an accuracy value of 86%, had the best performance in determining whether it rained or not in the upstream area of the Cirasea Sub-watershed, compared to the models with 10 minutes time lag (80%), 20 minutes time lag (68%), and 30 minutes time lag (66%). This model also has a better accuracy value than the accuracy value of the validation results from the detection made by three observers who used to observe cloud patterns in the Majalaya area. This shows that the model built is quite promising if it is implemented to assist the community in flood disaster mitigation activities in the Majalaya area.

Keywords: Convolutional Neural Network (CNN), Machine learning model, Majalaya, Rainfall, Ground-based images

Topic: Atmospheric Sciences