Efficient Calibration Technique under Irregular Response Surface

Conceptual rainfall-runoff models that aim at predicting streamflow from the knowledge of rainfall over a catchment are a basic tool for flood forecasting. The parameter calibration of a conceptual model usually involves the selection of an automatic optimization algorithm to search for the parameter values that minimize the distance between the simulated and observed data. However, practical experience with model calibration suggests that traditional optimization methods, such as the Newton-Raphson, conjugate gradient, and downhill simplex methods, are easily trapped in local minimums because the objective function surface is rough, caused by model structure errors and data errors. The across-ridge calibration method (ARC), an effective and efficient methodology for searching global optimization problem, is proposed in this paper. The method searches for all local optima in a fixed boundary, which makes it less susceptible to the irregularity of the response surface. The features and capabilities of ARC are illustrated by means of a binary function, an ideal conceptual rainfall-runoff model (Xinanjiang model) without model and data errors and a real model (Xinanjiang model) with structure and data errors. The results indicate that the proposed across-ridge calibration method can converge quickly from any starting point even when the response surface is rough. The optimized parameter sets for the Xinanjiang model (XAJ) are unique and conceptually realistic; the simulation performance of these parameters is reasonable and efficient.