Hydrological Drought Forecasting Using Machine Learning—Gidra River Case Study

Wael Almikaeel,Lea Čubanová,Andrej Šoltész

doi:10.3390/w14030387

Wael Almikaeel, Lea Čubanová + Show 1 more

Open Access

https://doi.org/10.3390/w14030387

Copy DOI

Journal: Water	Publication Date: Jan 27, 2022
Citations: 22	License type: CC BY 4.0

Affiliation: Slovak University of Technology in Bratislava

Abstract

Drought is one of many critical problems that could arise as a result of climate change as it has an impact on many aspects of the world, including water resources and water scarcity. In this study, an assessment of hydrological drought in the Gidra River is carried out to characterize dry, normal, and wet hydrological situations by using the Slovak Hydrometeorological Institute (SHMI) methodology. The water bearing coefficient is used as the index of the hydrological drought. As machine and deep learning are increasingly being used in many areas of hydroinformatics, this study is utilized artificial neural networks (ANNs) and support vector machine (SVM) models to predict the hydrological drought in the Gidra River based on daily average discharges in January, February, March, and April of the corresponding year. The study utilized in total 58 years of daily average discharge values containing 35 normal and wet years and 23 dry years. The results of the study show high accuracy of 100% in predicting hydrological drought in the Gidra River. The early classification of the hydrological situation in the Gidra River shows the potential of integrating water management with the deep and machine learning models in terms of irrigation planning and mitigation of drought effects.

Highlights

Drought can be characterized based on the objective of the study, which is essential when quantifying drought [1]
This study examined the performance of two well-known machine learning models, support vector machine (SVM) and artificial neural networks (ANNs), in predicting the hydrological drought of the Gidra River
This confirms the effect of splitting the data homogeneously, meaning that the training dataset should contain a variety of hydrological situations with different water bearing coefficient values, especially for values located around the limits of determining a new hydrological situation; when the input data are of one type, the daily average discharge values, as in this case

Summary

Introduction

Drought can be characterized based on the objective of the study, which is essential when quantifying drought [1]. Drought can be defined as a prolonged dry period in the natural climate cycle that can occur anywhere in the world. Considerable changes in climate components during the past decades are being connected in several cases to abnormal events such as droughts and floods [2]. Droughts generally correlate with large-scale impacts and are often driven by regional or even global-scale climate features. The historic classification of drought has emerged mainly from meteorological and hydrological studies in order to manage agricultural and socioeconomic impacts [3]. Drought classification is usually divided into four major categories: hydrological drought, soil moisture drought, meteorological drought, and socioeconomic drought [4]

Objectives

Methods

Results

Discussion

Conclusion