Floods are considered the most destructive natural catastrophes on the planet; therefore, studying their dynamics in the context of each region is a fundamental pillar of society. A very popular tool for studying the degree of threat is physics-based modeling using 2D models. It is imperative to include physical, hydrological, and anthropogenic variables in flood modeling approaches. The reviewed literature did not develop a methodology or strategy to facilitate this process. Based on this reason, it was considered for Riohacha, a coastal city in northern Colombia. A survey was conducted to determine social, economic, and flood depth variables during extreme events. Twenty socio-hydrological variables were obtained from the survey analysis using Kruskal–Wallis and multiple correspondence analysis (MCA) tests. Flood depth was the response variable used to select the socio-hydrological variables, among which the following stood out: age of the house, internal slope, toilet overflow, proximity to a wetland, paved road, raft area, and slope of the soil in front of the house. An iterative process supported by a Support Vector Machine (SVM) dummy was used to determine the appropriate combination of parameters and optimize the calibration process of the TELEMAC-2D hydrodynamic model. The curve number (CN) and Manning friction coefficient were used as calibration parameters. The optimization process was performed by entering the dummy SVM into the socio-hydrological variables CN and Manning's n to test up to 20,000 possible combinations of the parameters and evaluate the mean absolute error (MAE), mean error (ME), absolute relative error (RAE), root mean square error (RMSE), and inertia root mean squared error (IRMSE). The best combination of parameters to simulate an extreme flood event and determine the metrics with the observed and simulated flood depths was entered into TELEMAC-2D. This paper presents a methodological proposal for optimizing the calibration process of physics-based flood models (traditional method), considering socio-hydrological variables, to articulate it as a new step in the modeling protocol. In the traditional simulation, an RMSE of 0.48 m, MAE of 0.37 m, and IRMSE of 1.37 m are obtained, which decrease to 0.33 m, 0.27 m, and 0.92 m, respectively, when the socio-hydrological variables are included. Contrary to the previous metrics, the ME improves slightly from 0.15 m with the traditional method to 0.17 m when the socio-hydrological variables are included. These results can be improved using a digital terrain model (DTM) that more realistically represents the complexity of the urban fabric. With only 7 iterative processes, the model calibration was optimized using only seven iterative processes. Therefore, the proposed methodology was established as a tool for optimizing the flood modeling processes.
Read full abstract
Aug 1, 2023
David A Chin 
Open Access