A Hybrid Computational Intelligence Framework for Automatic Calibration of Catchment Scale Hydrodynamic Flood Models

Abstract This study presents a novel automated framework for the calibration and validation of rain-on-grid (RoG) hydrodynamic models at the catchment scale, with specific application to the TUFLOW HPC platform. The proposed approach establishes an integrated and automated workflow that links rainfall input directly to flood response, enabling efficient calibration of key hydraulic parameters, including Manning’s roughness, initial loss, and continuous loss. The framework combines ensemble-based parameter sampling, automated model execution, data-driven surrogate modelling, and evolutionary optimization algorithms (EAs) to systematically explore parameter combinations within the defined calibration ranges. Three modelling techniques—Random Forest (RF), Gaussian Process Regression (GPR), and Gradient Boosting (GB)—are employed to approximate model responses from an ensemble of hydrodynamic model simulations, thereby accelerating the calibration process and reducing computational costs. Furthermore, three EAs, consisting of Particle Swarm Optimization (PSO), Invasive Weed Optimization (IWO), and Differential Evolution (DE), are applied. The optimization objective is defined as minimizing the discrepancy between simulated and reference inundation extents, quantified through total inundated area derived from benchmark inundation datasets. The entire process is implemented in a MATLAB-based script that automatically controls TUFLOW simulations, iteratively adjusts parameters, and evaluates model performance without manual intervention. The framework is tested in a small catchment case study under multiple design rainfall events (50-, 100-, and 500-year return periods). Results demonstrate close agreement between the simulated and reference inundation areas, with differences of less than 1% across all scenarios. These findings indicate satisfactory predictive performance within the studied calibration range. The key novelty of this study lies in the integration of surrogate-assisted optimization within a fully automated calibration environment tailored for RoG models. Furthermore, the developed MATLAB code is made publicly available via GitHub, promoting transparency, reproducibility, and potential broader application in hydrodynamic flood modelling.