Improving ecological simulation through sensitivity analysis

Thornton, Kent W.; Lessem, Allan S.; Ford, Dennis E.; Stirgus, Connie A.

doi:10.1177/003754977903200503

Cited by 13 publications

(1 citation statement)

References 1 publication

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our local sensitivity analyses were patterned after sampling designs in which all parameters of interest were changed by a small, constant percentage (e.g., Miller et al 1976, Thorton et al 1979, MacNeil et al 1985, as defined by Swartzman (1979). The intervals sampled in local analyses neglect differences in parameter variances (Swartzman 1979) and are only dependent on the magnitudes of the parameters.…”

Section: Influence Of Sampling Designmentioning

confidence: 99%

Contaminant Transport through Agroecosystems: Assessing Relative Importance of Environmental, Physiological, and Management Factors

Breshears¹,

Kirchner²,

Whicker³

1992

Ecological Applications

View full text Add to dashboard Cite

Agroecosystems can become contaminated by atmospherically released radionuclides. The subsequent concentrations of radionuclides in foods are dependent on numerous environmental, physiological, and management factors. We compared four approaches for estimating the relative importance of several of these factors in determining concentrations of I and Cs in milk. A series of sensitivity analyses with Monte Carlo and full-factorial sampling designs was conducted on the PATHWAY model, which simulates radionuclide transport through an agroecosystem. Sensitivity of time-integrated concentrations in milk was estimated as a function of the time of year that fallout was deposited and as a function of time following a spring deposition. The dominant parameters affecting time-integrated concentrations of I in milk were the initial fraction of radionuclides deposited on vegetation, timing and amount of pasture consumption, and the production rate of milk. For time-integrated concentrations of the longer-lived Cs in milk, resuspension was a dominant parameter and pasture use was less important. The sampling designs were compared by ranking the parameters to which the model output is sensitive. The three sampling designs based on parameter variances produced sets of ranks that were similar to each other but differed from the ranking produced by the sampling design based on parameter magnitude. The results indicate which data are most crucial for real-time calculations following an accident and how subsequent dose from ingestion can be most effectively reduced, provide insight into model behavior, and help prioritize future research. This paper demonstrates the importance of variance-based sensitivity analysis.

show abstract