Sensitivity-based guided automatic calibration of hydrological models
Loading...
Date
2019
Authors
Semnani, Mohammad
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
A new method for efficient calibration of complex hydrological models that combines Dynamically Dimensioned Search (DDS) global optimization algorithm with Global Sensitivity Analysis (GSA) methods is introduced. This approach, which is called sensitivity-informed DDS, utilizes sensitivity indices to increase the probability of perturbation for the most sensitive parameters, while giving low chance to least sensitive ones. This feature improves the efficiency and effectiveness of optimization by finding good quality solutions in a shorter time. Three different implementations of sensitivity-informed DDS are considered. The first approach is named as GSA↔DDS, in which GSA toolboxes (Morris or Sobol) are performed initially and throughout the optimization process to constantly update the sensitivity information. The second approach is called GSA→DDS. In this method, the GSA methods are only performed initially to include the results of GSA within optimization process. The final implementation is called VARS→DDS. In this method, to enhance the efficiency of sensitivity analysis and optimization, VARS toolbox is performed outside the optimization to provide the sensitivity information. The performances of GSA↔DDS, GSA→DDS and VARS→DDS are compared with original DDS by solving various optimization problems (test functions and model calibration case studies). According to the results, when calibrating complex hydrological models with enough computational budget, VARS→DDS is significantly more efficient and effective than original DDS. However, the results also show that GSA→DDS and GSA↔DDS methods do not substantially improve the convergence rate and the final best solution compared to DDS. Thus, VARS→DDS is the recommended approach for sensitivity-informed DDS in calibration of distributed and semi-distributed models, when enough computational resources are available.
Description
Keywords
Sensitivity Analysis, Single Objective Optimization, Model Calibration