Environmental ModelingFeedback

Under some circumstances environmental models are required to provide the information about spatial and temporal contaminant concentrations needed for risk assessment. In general a model is a collection of facts, observations, and assumptions relating observed information to desired unobserved information or predictions. Models are not expected to provide a perfect understanding of the system they are intended to mimic. Such a goal is not achievable. However, models are expected to prove useful in the context of the decisions or the predictions they are intended to support. Usefulness is measured in terms of insights that are gained through model development and model evaluation, and decisions that can be supported through uncertainty and sensitivity analysis. Models are tools that support prediction or decision making; they support rather than supplant the decision making process. Models are numerical approximations of reality that are intended to provide a reasonable or reasoned facsimile of the system or process under examination. Numerical models are used to capture, through quantification, the underlying conceptual models. The numerical representation facilitates analysis and understanding, which, if performed using systematic and well-defined methods can provide valuable perspectives on the underlying system and valuable support to the decision making process.

Historically modeling approaches for linking observed and unobserved (predictive) information have been categorized as either empirical/statistical or process/deterministic. Models placed in these categories are treated as fundamentally different when they actually exist on a continuum bounded on the empirical end by the mean of the response (classical statistical regression modeling and probabilistic modeling) and on the deterministic end by complex models in which uncertainty is not acknowledged. Empirical models are generally built using observed data in which a response is statistically modeled as a function of predictor variables. Inclusion of predictor variables in even the simplest functional form is a move towards the process side of the modeling continuum. Process models are generally based on experimental observation or data in an ad hoc manner. The degree of statistical rigor in which observational information is incorporated in process models defines the model's location on the modeling continuum. Acknowledging that all models are a blend of determinism and chance leads to a statistical decision analytic approach for assessing the quality of both "empirical" and "process" models.