Predicting Hydrological Drought Alert Levels Using Supervised Machine-Learning Classifiers

Abstract

Reliable drought prediction is a global challenge in disaster-prone regions around the world. Data-driven models such as machine-learning (ML) classifiers have recently received considerable attention from water resources planners and managers. In this study, we applied several ML classifiers, including decision tree (DT), naive Bayes (NB), random forest (RF), and support vector machine (SVM) to the prediction of hydrological drought classes. Daily data of precipitation, reservoir inflow, and reservoir volume collected from three large dams (Andong, Chungju, and Seomjin) in South Korea were used as classifier input to predict hydrological drought alert levels. A comparison of the accuracy and computation time of each ML classifier revealed that the classifiers were capable of predicting hydrological drought alert levels, with the SVM achieving outstanding performance in terms of accuracy (97%) and precision (89%) and the NB exhibiting superior computational time (0.63 sec). The results of this study indicated that the ML classifiers can be effective predictors of hydrological drought classes and can provide warnings of drought conditions.