Parameter uncertainty and interaction in complex environmental models

Abstract

Recently developed models for the estimation of risks arising from the release of toxic chemicals from hazardous waste sites are inherently complex both structurally and parametrically. To better understand the impact of uncertainty and interaction in the high‐dimensional parameter spaces of these models, the set of procedures termed regional sensitivity analysis has been extended and applied to the groundwater pathway of the MMSOILS model. The extension consists of a tree‐structured density estimation technique which allows the characterization of complex interaction in that portion of the parameter space which gives rise to successful simulation. Results show that the parameter space can be partitioned into small, densely populated regions and relatively large, sparsely populated regions. From the high‐density regions one can identify the important or controlling parameters as well as the interaction between parameters in different local areas of the space. This new tool can provide guidance in the analysis and interpretation of site‐specific application of these complex models.

Sensitivity analysis approaches applied to systems biology models

“…Previous practice with hydrological models has shown that the fraction of 'acceptable' behaviours is barely larger than 5% over the total simulations for large models (e.g. with number of input parameters .20), indicating a lack of statistical power [2,78]. (iii) If some parameters are highly correlated with each other, the marginal distributions cannot separate under the acceptable classification and thus underestimate the importance of some parameters.…”

Sensitivity Analysis of Environmental Models: A Systematic Review with Practical Workflow

“…van Werkhoven et al, 2009); (ii) investigate the parameters' influence and interactions in the regions of the parameter space associated with higher model performance, and thus provide the knowledge base for a more efficient local-search calibration in those regions (e.g. Spear et al, 1994); (iii) assess the potential for and limitations of model calibration given other uncertainty sources besides parameters, e.g. measurement errors in the observations or in the model forcing data (e.g.…”

Tree-Structured Modeling of the Relationship Between Great Lakes Ice Cover and Atmospheric Circulation Patterns

“…Some recent applications in atmospheric and hydrological sciences include: Burrows and Assel 1992, Rodionov 1994, Spear et al 1994, Burrows et al 1995, Zorita et al 1995. CART presents its results in the form of decision trees using methodology known as binary recursive partitioning.…”