Regional flood modelling using artificial intelligence-based non-linear techniques

Abstract

This study made several contributions to advancing AI-based regional flood frequency analysis (RFFA) techniques. First, it explored deep learning-based methods, specifically Convolutional Neural Networks (CNN), and compared their performance against Support Vector Machine (SVM), Decision Trees (DT), and the traditional linear approach like Multiple Linear Regression (MLR). The findings demonstrated that CNN consistently outperformed the other techniques. Second, to evaluate the relative performance of linear and non-linear approaches, this research compared the widely applied Quantile Regression Technique (QRT) as a linear model with the Artificial Neural Network (ANN) as a non-linear model. The results revealed that ANN achieved superior performance in terms of median relative error across all the selected AEPs. Third, ensemble learning methods, including Random Forest Regression (RFR) and Gradient Boosting Regression (GBR), were developed and benchmarked against the standalone ANN. The outcomes showed that RFR outperformed both GBR and ANN for Q2, Q5, Q10, Q20, and Q100, while GBR exhibited better performance for Q50. Finally, an uncertainty analysis was undertaken to assess the robustness of AI-based RFR in comparison with the traditional linear MLR model. This analysis applied Monte Carlo simulation to estimate the confidence intervals of flood quantiles for each of the six AEPs. In this doctoral study, model performance was evaluated using several statistical indices, including Relative Error (RE), median Relative Error (REr), Root Mean Square Error (RMSE) and Bias thereby confirming the reliability of the models. Overall, this doctoral research is expected to make a significant contribution to the advancement of AI-based RFFA techniques and, ultimately, to provide valuable evidence for future updates of the Australian Rainfall and Runoff (ARR) national guidelines.

Date of Thesis submission	2025
Original language	English
Awarding Institution	Western Sydney University
Sponsors	Australian Government Research Training Program Scholarship
Supervisor	Ataur Rahman (Supervisor) & Sathaa Arumugam Sathasivan (Supervisor)