Flood forecasting using advanced machine learning model and flood susceptibility analysis and mapping using morphometric parameters

Abstract

Flood catastrophes are among the natural disasters that have occurred regularly around the world. Malaysia is one of the countries that experience flood disasters on a yearly basis, most notably during the monsoon season, which runs from November to January. This study developed a novel flood forecasting model through the application of advanced machine learning (ML), deep learning (DL), and natural language processing (NLP) for sentiment analysis and text classification before a flood event, during a flood event and after a flood event. The morphometric ranking approach (MRA) was used to identify flood-susceptibility areas. Various data sources were collected including natural dimension such as rainfall intensity (mm), streamflow (cm/s), and water level (m) from Department of Irrigation and Drainage, and social dimension like text data extracted from Twitter platform. A digital elevation model (DEM) was used to process parameters for MRA with the application of geographic information system (GIS) for identifying flood-prone areas. General ML pipelines were used before building the model such as data pre-processing, data exploration to detect outliers, and filling missing values. The flood forecasting model used advanced machine learning and deep learning specifically long-short term memory (LSTM) which is suitable for time series data of rainfall and streamflow forecasting. Additionally, the model was developed using these three models: LSTM, ARIMA, and FB Prophet. The forecasting results indicated that the LSTM model has a root mean square error (RMSE) of 10.76, which is more accurate in comparison to the other models ARIMA and FB Prophet, which have RMSE values of 14.15, and 14.23, respectively. The accuracy of the model of text classification algorithm for predicting flood events is 0.87. Flood susceptibility mapping using MRA revealed that sub-catchments 5, 24, and 25 were highly susceptible to flooding. These sub-catchments were located in Jeli, Kuala Krai sub-district, and Gua Musang sub-district, respectively. In sum, this flood forecasting model is vital to provide flood information for early warning system to enable flood managers or decision-makers to make more informed plans during the flood preparation and mitigation phases, thereby minimizing the impact of floods on people, property, and the environment.