88
5 Handling Uncertainty and Sensitivity …
of the mechanisms of action are uncertain, this reflects on uncertainty. As model
complexity increases, the uncertainty tied to the model framework is reduced, but
a more complex model uses more parameters and the data uncertainty increases
(Gaber et al. 2009). For ERA Acute there are uncertainties of the model that belong to
both the structure (model uncertainty) and the numerical parameters used (epistemic
uncertainties).
On the condition that the functions and calculation sequences in the model are
correct, if the input is not changed, the model output stays the same. In this case,
uncertainty in the model output is solely affected by variation in the input parameter.
This is called Epistemic, or reducible uncertainty (Helton et al. 2006), related to lack
of knowledge of the true value of a constant parameter (Marino et al. 2008).
For the individual parameters used in a model it is important to distinguish clearly
between:
Variability: How spread out or clustered a data set is, e.g. the (natural) variation
in the measured values found in nature and
Uncertainty—The lack of certainty or knowledge about what the value of the
parameter/data truly is. Such data uncertainty is specific to the individual parameter. As mentioned above, a more complex model uses more parameters and
data uncertainty therefore increases (Gaber et al. 2009). ERA Acute uses many
parameters.
Sensitivity Analysis (SA) tells us how the model’s response can be apportioned to
changes in model inputs. It is algorithm specific. For models with a high number
of parameters, sensitivity analyses are useful to rank the relative importance of
the factors and processes involved (Saltelli 2004).
ERA Acute is a new method and testing the sensitivity of the model to variation
in the input parameters is an important part of uncertainty handling, with the goal
of ensuring that ERA Acute does not under-estimate environmental risk. All data
sets and parameter values have inherent uncertainties and a model consisting of a
series of calculations will need some method of handling uncertainty. In the process
of developing ERA Acute, the following activities were carried out:
1. Sensitivity testing of the risk functions to the variation in input parameters
2. A pilot study to score the parameters and propose feasible uncertainty handling
The functions of ERA Acute are built so that individual parameters representing
biological or environmental characteristics can be improved as knowledge increases,
thereby reducing uncertainty by a continuous improvement process. Sensitivity
testing provides knowledge of which of the parameters that contribute most to the
final endpoint values, and therefore the testing provides information on which parameters that would be most important to improve by further research if they have high
uncertainty.
ERA Acute covers four compartments and uses a large number of functions. The
input parameters (values and datasets) that are used are based on knowledge from
few and highly diverse incidents. Validating the results of the method and applying
the results with an acceptable level of (un)certainty is therefore challenging (see
5 Handling Uncertainty and Sensitivity …
of the mechanisms of action are uncertain, this reflects on uncertainty. As model
complexity increases, the uncertainty tied to the model framework is reduced, but
a more complex model uses more parameters and the data uncertainty increases
(Gaber et al. 2009). For ERA Acute there are uncertainties of the model that belong to
both the structure (model uncertainty) and the numerical parameters used (epistemic
uncertainties).
On the condition that the functions and calculation sequences in the model are
correct, if the input is not changed, the model output stays the same. In this case,
uncertainty in the model output is solely affected by variation in the input parameter.
This is called Epistemic, or reducible uncertainty (Helton et al. 2006), related to lack
of knowledge of the true value of a constant parameter (Marino et al. 2008).
For the individual parameters used in a model it is important to distinguish clearly
between:
Variability: How spread out or clustered a data set is, e.g. the (natural) variation
in the measured values found in nature and
Uncertainty—The lack of certainty or knowledge about what the value of the
parameter/data truly is. Such data uncertainty is specific to the individual parameter. As mentioned above, a more complex model uses more parameters and
data uncertainty therefore increases (Gaber et al. 2009). ERA Acute uses many
parameters.
Sensitivity Analysis (SA) tells us how the model’s response can be apportioned to
changes in model inputs. It is algorithm specific. For models with a high number
of parameters, sensitivity analyses are useful to rank the relative importance of
the factors and processes involved (Saltelli 2004).
ERA Acute is a new method and testing the sensitivity of the model to variation
in the input parameters is an important part of uncertainty handling, with the goal
of ensuring that ERA Acute does not under-estimate environmental risk. All data
sets and parameter values have inherent uncertainties and a model consisting of a
series of calculations will need some method of handling uncertainty. In the process
of developing ERA Acute, the following activities were carried out:
1. Sensitivity testing of the risk functions to the variation in input parameters
2. A pilot study to score the parameters and propose feasible uncertainty handling
The functions of ERA Acute are built so that individual parameters representing
biological or environmental characteristics can be improved as knowledge increases,
thereby reducing uncertainty by a continuous improvement process. Sensitivity
testing provides knowledge of which of the parameters that contribute most to the
final endpoint values, and therefore the testing provides information on which parameters that would be most important to improve by further research if they have high
uncertainty.
ERA Acute covers four compartments and uses a large number of functions. The
input parameters (values and datasets) that are used are based on knowledge from
few and highly diverse incidents. Validating the results of the method and applying
the results with an acceptable level of (un)certainty is therefore challenging (see