Influential input classification in probabilistic multimedia models

TitleInfluential input classification in probabilistic multimedia models
Publication TypeJournal Article
Year of Publication2001
AuthorsRandy L Maddalena, Thomas E McKone, Dennis PH Hsieh, Shu Geng
JournalStochastic Environmental Research and Risk Assessment
Volume15
1
Issue1
Pagination1-17
Date Published03/2001
ISSN1436-3240
KeywordsError propagation, model development, Monte Carlo, multimedia mass balance, variance
Abstract

Monte Carlo analysis is a statistical simulation method that is often used to assess and quantify the outcome variance in complex environmental fate and effects models. Total outcome variance of these models is a function of (1) the variance (uncertainty and/or variability) associated with each model input and (2) the sensitivity of the model outcome to changes in the inputs. To propagate variance through a model using Monte Carlo techniques, each variable must be assigned a probability distribution. The validity of these distributions directly influences the accuracy and reliability of the model outcome. To efficiently allocate resources for constructing distributions one should first identify the most influential set of variables in the model. Although existing sensitivity and uncertainty analysis methods can provide a relative ranking of the importance of model inputs, they fail to identify the minimum set of stochastic inputs necessary to sufficiently characterize the outcome variance. In this paper, we describe and demonstrate a novel sensitivity/uncertainty analysis method for assessing the importance of each variable in a multimedia environmental fate model. Our analyses show that for a given scenario, a relatively small number of input variables influence the central tendency of the model and an even smaller set determines the spread of the outcome distribution. For each input, the level of influence depends on the scenario under consideration. This information is useful for developing site specific models and improving our understanding of the processes that have the greatest influence on the variance in outcomes from multimedia models.

DOI10.1007/PL00009786