Artificial Intelligence-based Regional Flood Modeling: Illustration of Learning Aspects by a Doctoral Student

Design flood estimation in ungauged catchments remains a persistent challenge in the field of hydrology. To address this challenge, the Regional Flood Frequency Analysis (RFFA) is a widely accepted method. Until now, the RFFA method has primarily relied on linear approaches, which are inept to capture the complexity of nonlinear hydrological processes. The advent of Artificial Intelligence (AI)-based techniques has demonstrated superior performance in addressing this limitation. Recently in RFFA, several studies have indicated the effectiveness of AI-based models compared to linear models. The advantage of AI-based techniques lies in their ability to learn from given datasets without adhering to predefined rules, contributing to more accurate predictions. However, application of AI to RFFA problem is not straightforward. This paper outlines the learning aspects that the first author gains in carrying out her PhD research applying AI-based techniques to RFFA problem to New South Wales flood and catchment data. It is found that learning the fundamentals of the AI-based techniques is relatively difficult. Although its application to a dataset using available software is relatively easy, the interpretation of results and understanding assumptions related to the model output need significant efforts. The findings of this study will benefit other students and researchers who would like to apply AIbased methods to RFFA problems in Australia and other countries