Machine Learning for Flood Prediction in Indonesia: Providing Online Access for Disaster Management Control

Abstract

Climate change has altered the hydrological cycle which leads to unpredictable and extreme precipitation events, triggering a high risk of flooding across the globe. As one of the most vulnerable countries to floods, there should be a high necessity for accurate and reliable flood forecasting in Indonesia. In a megacity such as Jakarta, several proper flood forecasting methods are already available; however, the rest of the Indonesian regions are left behind. The current flood forecasting provided by the Indonesian government is on a monthly basis and made of monthly forecasts of precipitation and flood-prone area maps without mathematical or statistical analysis. The sample of current forecast from the last 3 years was evaluated and resulted in hundreds of wrong predictions. Therefore, a new prediction model using a machine learning algorithm is proposed to provide daily flood prediction in Indonesia, which consists of 33 regions. Data crawling was conducted to obtain daily precipitation, streamflow, land cover, and flood data for the last 13 years, from 2008 to 2021. The model was built using Random Forest (RF) algorithm for classification. The accuracy, specificity, precision, recall, and f1-score of the prediction model using the RF algorithm are approximately 99.91%, 99.90%, 99.47%, 99.99%, and 99.72%, respectively. Moreover, the AUC (Area Under the Curve) of the ROC (Receiver Operating Characteristics) curve results in 99.97% and an OOB (Out of Bag) error of < 0.001. The Partial Dependence Plots (PDP) indicate that flood in Indonesia is highly dependent on precipitation rate value. Furthermore, the variable importance plot shows the most to the least important variables, which are precipitation rate, forest ratio, and streamflow. Both the PDP and variable importance suggest that the vast majority of occurred flood events are flash flooding and only a few coastal, riverine, or other types of flooding in Indonesia. The precipitation forecast for June 2022 was gathered and input into the model, resulting in the flood prediction. Flood prediction for June 2022 was then deployed as an open-access website using shinyR. This flood prediction model is a fast and accurate alternative to complement the current flood forecast with extensive information.