Machine learning insights into the evolution of flood Resilience: A synthesized framework study

Abstract

Enhancing urban resilience represented a viable strategy to mitigate flooding induced by intense human activities and climate change. However, existing studies often concentrated on system attributes or isolated resilience characteristics, failing to offer a holistic evaluation of urban flood resilience performance. Thus, it was imperative to develop a comprehensive flood resilience framework that incorporated the resilience evolution process including resistance, economic and function recovery. Consequently, this study endeavored to devise a synthesized framework for evaluating urban flood resilience, subsequently employing a Convolutional Neural Network (CNN) model for simulation. The findings indicated that: (1) Guangzhou’s maximum resistance capacity diminished from 0.52 to 0.50 as rainfall return periods altered, while Dongguan exhibited the lowest resistance, decreasing from 0.42 to 0.40. Regarding functional recovery capacity, Guangzhou ranked highest (0.35) and Foshan lowest (0.19); (2) according to Triangular Fuzzy Number-based AHP (TFN-AHP) analysis, the area classified as highest in resilience decreased from 15.6% to 12.1% of the total, whereas the low resilience area increased from 7.6% to 8.7%; (3) Zhuhai and Zhaoqing were primarily clustered along the resistance axis, in contrast, Dongguan was distinguished by its advancement along the axis of functional recovery.(4) CNN simulations yielded precise outcomes, with the Area Under the Receiver Operating Characteristic Curve (AUC) and predictive accuracy (ACC) values exceeding 0.8,respectively. The insights provided by this research were crucial for entities tasked with flood risk management.