A Repetitive Parameterization and Optimization Strategy for the Calibration of Complex and Computationally Expensive Process-Based Models With Application to a 3D Water Quality Model of a Tropical Reservoir

Parameter calibration is critical for modeling, especially for current process-based models that are complex with many chemical and biological processes and immeasurable model parameters. This analysis quantifies significant disadvantages of the traditional use of local or global sensitivity analysis (SA) for selecting calibration parameters of nonlinear, expensive models when there are a large number of constituents and parameters. We propose a new Repetitive parameterization and optimization (Rep-OPT) strategy that uses multiple optimization steps; and between each optimization step, a modeler picks the parameters to be optimized in the next optimization step. The modeler picks the parameters in each iteration following a suggested set of steps that analyze which processes and parameters are related to the poorly fit constituents with the current parameter set. We successfully applied the Rep-OPT strategy on a complex tropical water quality model with more than 91 parameters using real data. We demonstrate that expert knowledge with assistance of proposed postanalysis techniques (i.e., trade-off analysis, component analysis, and mass-balance analysis) can identify the right calibration parameters and obtain excellent model fit. In contrast, the traditional approach using SA with optimization (SA-OPT) does not find the right calibration parameters for our data. The solution found by Rep-OPT excellently improves manual solution by 32.7% in goodness-of-fit, and all calibrated constituents fit well to observations. The solution found by SA-OPT using global SA improves manual solution by only 13.3%. Local sensitivity by SA-OPT performs very poorly being 49.6% worse than manual solution.