Name
Climate-Resilience of Small Dams and Levees: A Multi-Model Approach for Design Flood Estimation
Date & Time
Wednesday, May 10, 2023, 2:30 PM - 2:45 PM
Description
Rapidly rising economic losses from floods in Canada are raising concerns about the safety of critical flood control and mitigation structures such as dams and levees. Due to these concerns and the expected impacts of climate change, it is important to evaluate the climate resilience of such structures to ensure the safety of people, the environment, and various infrastructure systems. In this study, a multi-model approach for estimating flood quantiles in current and future climates is proposed. The modelling approaches include Linear Regression, Locally Weighted Linear Regression, Generalised Least Squares Regression, K-Nearest Neighbors Regression, Support Vector Regression, Weighted Support Vector Regression, Random Forest Regression, and Weighted Random Forest Regression. The concept of Region of Influence (ROI) is utilized to create homogeneous groups of stations, while Bayesian Model Averaging (BMA) is applied to merge the outcomes from multiple modelling approaches. The ROI approach quantifies the influence of each location in the study region on the target location of interest by using physiographic and climatic variables through Principal Component Analysis (PCA). A nonlinear weighting scheme, originally proposed for regional flood frequency analysis using the ROI approach, is used for conducting locally weighted regression and informing selected machine learning approaches. Following this approach, flood quantiles of 10, 25, 50, and 100-year return periods are estimated at each site, and the performance of each modelling approach is assessed using the leave-one-out cross-validation procedure. The locally weighted regression model outperformed other options, with the BMA achieving the best score across multiple evaluation metrics.
Location Name
Lynx
Full Address
Banff Park Lodge Resort Hotel & Conference Centre
201 Lynx St
Banff AB T1L 1K5
Canada
Abstract
Rapidly rising economic losses from floods in Canada are raising concerns about the safety of critical flood control and mitigation structures such as dams and levees. Due to these concerns and the expected impacts of climate change, it is important to evaluate the climate resilience of such structures to ensure the safety of people, the environment, and various infrastructure systems. In this study, a multi-model approach for estimating flood quantiles in current and future climates is proposed. The modelling approaches include Linear Regression, Locally Weighted Linear Regression, Generalised Least Squares Regression, K-Nearest Neighbors Regression, Support Vector Regression, Weighted Support Vector Regression, Random Forest Regression, and Weighted Random Forest Regression. The concept of Region of Influence (ROI) is utilized to create homogeneous groups of stations, while Bayesian Model Averaging (BMA) is applied to merge the outcomes from multiple modelling approaches. The ROI approach quantifies the influence of each location in the study region on the target location of interest by using physiographic and climatic variables through Principal Component Analysis (PCA). A nonlinear weighting scheme, originally proposed for regional flood frequency analysis using the ROI approach, is used for conducting locally weighted regression and informing selected machine learning approaches. Following this approach, flood quantiles of 10, 25, 50, and 100-year return periods are estimated at each site, and the performance of each modelling approach is assessed using the leave-one-out cross-validation procedure. The locally weighted regression model outperformed other options, with the BMA achieving the best score across multiple evaluation metrics.
Session Type
Breakout Session