As described on elsewhere we can categorise uncertainty into six classes:

• Parameter Uncertainty
• Model Uncertainty
• Scenario (Future) Uncertainty
• Natural Variability (Turbulence)
• Bias
• Evaluation Data Uncertainty

Uncertainty is already an important factor in determining the safety of radioactive waste disposal . Various approaches to treating uncertainty have been considered in that context, those mainly used have been 'Best estimate'; conservative (cautious) and Monte Carlo (probabilistic).

Of the six classes of uncertainty, those most readily quantifiable are Parameter Uncertainty, Natural Variability and Evaluation Data Uncertainty. Model Uncertainty and Bias (Model or Data) can be bounded but not fully quantified. Future uncertainty is difficult to estimate, and increases the further into the future the modelling is extended.

The 'Best Estimate' and 'Conservative' approach both involve undertaking calculations with a single set of parameter values, respectively the 'average' value or the maximum/minimum (whichever would give the highest pollutant concentration). Use of Monte Carlo methods, undertaking multiple runs with changes to the input parameters, gives a distribution of results. This allows an estimate of the probability that the concentration or deposition will be above any specified threshold.

In atmospheric dispersion modelling the approach that is used is almost always to undertake a deterministic calculation using the 'Best Estimate' of input parameter values. However, some uncertainty calculations have been undertaken using 'deterministic sensitivity' studies, for instance by Beychok (2000).

For modelling Gaussian Plumes it is also possible to use a Monte Carlo technique.

© 2000 Neill S Cooper
Last changed on January 5, 2004