4th UF Water Institute Symposium Abstract

Submitter's Name Yogesh Khare
Session Name Poster Session: Impact of changing drivers on water resources
Poster Number 55
Author(s) Yogesh Khare,  Graduate Research Assistant (Presenting Author)
  Christopher Martinez,  Associate Professor, Agricultural and Biological Engineering, University of Florida
  Rafael Munoz-Carpena, Professor, Agricultural and Biological Engineering, University of Florida
  Robert Rooney, Post-doctoral Research Associate, Agricultural and Biological Engineering, University of Florida
  A multi-criteria trajectory-based parameter sampling strategy for the screening type sensitivity analysis method of elementary effects
  Processes specific hydrologic and water quality models are often integrated to build complex computer programs that can simulate large natural systems such as watersheds. These large scale models can assess impacts of natural and anthropogenic disturbances on natural systems and have become important decision making tools. A typical feature of such models is tens to hundreds of input parameters which makes it essential to evaluate the model through global uncertainty and sensitivity analyses. However, due to the high dimensionality of the problem in hand and limited availability of the computational resources, parameter screening i.e. separation of important model parameters from unimportant ones at a low computational cost becomes very important. The method of elementary effects, a global parameter screening method, is the most extensively used methodology for parameter screening. Due to issues like inefficient parameter screening, time requirements for parameter sample generation etc. development of an effective sampling strategy has been a research focus in recent years. In this work a new sampling strategy based on the generation of exact theoretical distributions and maximizing the trajectory spread – Uniformity Sampling – is presented. The performance of the new sampling strategy relative to traditional strategy - the Morris sampling and relatively new strategies - method of optimized trajectories and its modified version was evaluated using a number of criteria such as the uniformity of generated parameter distributions, time efficiency, and the spread of trajectories. The ability of the various sampling strategies to screen important parameters was assessed using five test functions in eighteen configurations covering a range of model sizes and complexities. The Uniformity Sampling, in general, performed better than other sampling strategies across tested functions and criteria, underlining the effectiveness of multi-criteria based sampling and the need to focus future efforts on combining various sampling criteria to obtain robust parameter screening results.