Implementation of Heuristic Search Algorithms in the Calibration of a River Hydraulic Model

Analyzing the responses of the rivers to different imposed conditions such as flooding is a key to understanding many river engineering problems. Historically, characterization and prediction of river flows have been based on the one- or two-dimensional hydraulic models, which are only reliable following extensive calibration by trial-and-error. This method of calibration requires a user to manually change one or multiple input parameters (usually in the forms of roughness parameters) in order to find the set of parameters that minimizes model error for at least one observable scenario. Searching for this global optimum manually, if possible, is a time-consuming and tedious job, which also requires the user to have significant hydraulic knowledge. Heuristic search algorithms provide a data-driven, alternate method to calibrate hydraulic models. In this contribution, two heuristic search algorithms, (1) particle swarm optimization (PSO) and (2) genetic algorithm (GA) are used to calibrate the iRIC, a two-dimensional hydraulic model for a test case on the Green River in Utah. The goal of the searching algorithms was similar to that of trial-and-error model calibration- to find the combination of roughness values that minimized model error compared to the measured water surface elevation. An artificial neural network (ANN), which was trained based on the simulated water surface profiles obtained from iRIC, was used to measure the fitness of each solution. Implementation of PSO and GA both improved the accuracy of the calibration by 34 and 30 percent respectively relative to the trial-and-error. We demonstrate that the implementation of heuristic search models provides an objective methodology for calibration of hydraulic models with improved performance when compared to models calibrated via trial-and-error. An additional advantage of these methods is that the data-driven nature of the approach minimizes the requirements of hydraulic knowledge for a given user and eliminates user bias in the choice of roughness values. This study demonstrates the utility of heuristic search algorithms in the objective calibration of hydraulic models and represents an advance toward the automated prediction of flooding hazards in real-time.