CHALLENGES IN MULTIOBJECTIVE IDENTIFICATION AND EVALUATION OF WATERSHED MODELS FOR UNGAUGED BASINS

Open Access
Author:
Bhushan, Rashi
Graduate Program:
Civil Engineering
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
September 08, 2008
Committee Members:
  • Patrick M Reed, Thesis Advisor
  • Dr Thorsten Wagener, Thesis Advisor
Keywords:
  • Regionalization
  • Ungauged Basins
  • Calibration
  • Conceptual models
  • Optimization
  • Regional Sensitivity Analysis
Abstract:
Rainfall runoff models are standard tools for making hydrologic predictions. All hydrologic models require some degree of parameter calibration to achieve reliable predictions. However, a large part of the world remains ungauged and requires robust model identification and evaluation approaches to improve predictions in ungauged locations. This paper demonstrates both benefits and challenges associated with predictions in ungauged basins (PUB) framework that uses multi-objective optimization in combination with hydrologic indices regionalization to improve ensemble predictions at ungauged sites. Our objective is to assess how the quality of regionalized hydrologic constraints impacts the difficulty of identifying behavioral parameter sets using multi-objective optimization. This work characterizes the modes of failure in the optimization algorithm using random seed analysis and ensembles of hydrologic predictions obtained from high, intermediate and low quality regionalized constraints for representative watersheds in the United Kingdom. The results highlight that a slight increase in model complexity can lead to severe difficulties for the multi-objective PUB framework when the new model fails to produce simulated hydrologic indices within the feasible ranges attained from regionalization. Results also point to the issue that low quality regionalized constraints can still yield improved ensemble predictions given longer search durations for the optimization algorithm. Random seed analysis for the search dynamics of the multi-objective optimization algorithm and scatter plots showing the overall coverage of the behavioral indices space defined by regionalized hydrologic constraints provide key diagnostics of the quality and effectiveness of the predictions attained from the multi-objective PUB framework in the absence of observations.