Multi-objective calibration of hydrological models and data assimilation using genetic algorithms.

摘要

This dissertation investigates parameter estimation and data assimilation in the context of hydrological modeling to improve water resource decisions. Using the Non-dominated Sorting Genetic Algorithm-II (NSGA-II), the Soil and Water Assessment Tool was calibrated in a multi-objective fashion for simulations of streamflow. The resulting output is a Pareto frontier comprising a set of incomparable solutions which form a trade-off between two model evaluation objectives.;Additionally, a model characterization framework (MCF) was developed and it uses cluster analysis to examine the distribution of solutions, and conditional probability to combine linkages between the distributions of solutions in both spaces. The MCF computes two indicators: robustness and choice index - which categorizes incomparable sets of solutions to select parameter set(s) with desired properties/behaviour. The evaluation of linkages between robustness and choice index for 225 separate evaluations show that robustness is critical to the performance of solutions across several validation periods.;Furthermore, the study has improved a time series of soil moisture through a joint assimilation of satellite brightness temperature and soil moisture. The NSGA-II was applied in a data assimilation framework to merge two soil moisture estimates. One soil moisture was estimated from the Advanced Microwave Scanning Radiometer-Earth Observing System (AMSR-E) by assimilating brightness temperature into a radiative transfer model. The other estimate of soil moisture was generated from the Canadian Land Surface Scheme (CLASS). A comparison between the assimilated soil moisture and in situ dataset showed an improvement in accuracy and temporal pattern that was accomplished through the assimilation framework;Using the Pareto frontier, the study has developed an automated framework to select solutions from the trade-off surface by evaluating the distribution of solutions in objective space and parameter space. The framework selects solutions with four unique properties including a representative pathway in parameter space, a basin of attraction in objective space, a proximity to the origin in objective space, and a balanced compromise between objective space and parameter space (denoted BCOP). Evaluation of the four auto-selection methods for 15 calibration outputs which are each evaluated across 15 different validation periods show that BCOP perform consistently better than other methods.