Flash Flood Forecasting Using Machine Learning Models: A Scientometric Analysis.

Abstract

Hydro-meteorological hazards are a major issue in many regions of the world, including Romania. Among these, flash floods are the most frequent phenomena, generating significant annual socio-economic and environmental damages. In recent years, flash flood forecasting using machine learning algorithms has become an useful tool for data-based hydrologic modeling. Machine learning allows to create mathematical relationships between the river discharge and other climatic and physico-geographic parameters from the training dataset. This paper aims to perform a scientometric analysis using open-source programs, namely ScientoPyGui and VOSviewer. The expression ‘flash flood forecasting AND machine learning’ was searched in the Web of Science and Scopus databases. After merging and removing duplicates, 112 publications were retained for analysis. Their number has increased by 60% in the past three years (after 2021) with a trend towards a sub-branch of machine learning, namely deep learning. The spatial distribution of the papers showed that China is a global leader with 25% of the total. These findings highlight the increasing role of machine learning based models (particularly deep learning) in enhancing flash flood forecasting, a nonstructural measure for the flash flood risk mitigation

Keywords

• flash flood forecasting
• scientometric analysis
• machine learning
• deep learning