Bivariate Copula Model for selected stations in Peninsular Malaysia rainfall data

Abstract

Malaysia faces a high risk of flooding nearly every year, driven by seasonal monsoons and sudden flash floods. These events result in significant economic damage; in 2021 alone, as recorded by Department of Statistics Malaysia, total losses reached approximately RM 6.1 billion. Hence, predicting extreme rainfall episodes is critical for the Malaysian governments to give early warning to authorities before the severe floods happen. Without such alerts, the government faces a massive financial burden to mitigate and repair flood-related damages. Developing reliable rainfall prediction models is therefore essential for the Malaysian government to issue timely alerts. Such systems rely on the analysis of historical rainfall data to provide authorities with a clear approximation of future rainfall intensity and volume, thereby reducing the financial burden of post-flood recovery. Thus, this study aims to predict rainfall data of the selected stations in Peninsular Malaysia from 2015 to 2021 by using bivariate copula model, specifically the Elliptical and Symmetric Archimedean Copula models. The selected monitoring stations are chosen from four different states which are Johor, Selangor, Kuala Terengganu and Kelantan. These four stations were chosen for their proximity to flood-prone zones, specifically regarding monsoon and flash flood risks. Bivariate copula model was employed in this study since it could capture the complex relationship between two different rainfall stations data with different distributions. Thus, the model can provide a more flexible and accurate representation of how heavy rainfall at one location relates to another. Before the copula model is selected, the distribution analysis is performed on the rainfall data and it is found that the lognormal distribution gives the best fit for the rainfall data for all stations. Then, according to Akaike Information Criterion, Clayton Copula model outperformed the other copula models, including Frank, Gumbel, Gaussian and t student copulas. The Clayton model demonstrated a better ability to capture the bivariate dependencies between stations, showing a substantial improvement in fit ranging from 25% to 230% over the other copula models. The results indicate that the Clayton Copula model can be effectively used by authorities, particularly government agencies, to predict reliable rainfall data by capturing inter-station dependencies. This prediction can be served as a tool to make better informed decision making particularly related to managing flood risks.