5.2 Methods Used in Sensitivity Testing
91
input parameters to ERA Acute. Where calculations were carried out in succession,
combined formulas were used. PRCC allows independent effects of each parameter
to be determined, even when the parameters are correlated. The goal is to determine
which factor, once fixed to its true value by additional research, on average leads
to the greatest reduction in the variance of an output. The interpretation of PRCCs
assumes a monotonic relationship (relationship or function which preserves a given
trend) between parameters (Marino et al. 2008). This is the case for all the (sub-)
models used in ERA Acute. The rank-transformation is done to reduce the effect of
non-linear data, and PRCC is a robust sensitivity measure for nonlinear, monotonic
relationships (Marino et al. 2008).
The result is a sensitivity index for each input parameter to the formula, which is
the fraction of the variation in the output value that can be ascribed to the different
parameters. Note that this is given the uncertainty defined by the range of natural
variation (results based on literature search) and the weight of each value given
by the distribution (uniform—equal weight). If a different distribution for the initial
random drawing of values had been used, the result would have been different.
However, given the nature of the parameters, a uniform distribution was assumed.
The use of these statistical methods in the ERA Acute sensitivity testing is
described in further detail in the project reports by Bjørgesæter and DamsgaardJensen (2018) and Stephansen and Bjørgesæter (2017).
Impact and restoration functions were tested for each compartment and for each
relevant VEC-group within the compartment having different parameter values
and/or functions.
The results from the Spearman correlation coefficient analysis are presented in the
test reports by Bjørgesæter and Damsgaard-Jensen (2018) for surface, water column
and shoreline compartments and Stephansen and Bjørgesæter (2017) for the seafloor
compartment.
5.2.3 Example from Surface Compartment
For the deterministic testing, all parameter values used for the wildlife groups are
available in the test report (Bjørgesæter and Damsgaard-Jensen 2018), as well as
figures showing the results for each of the tested parameters. As part of the testing
it was determined that the equation which includes the exposure time (N-let2)
(Sect. 3.3.1) performs best according to the impact estimated from various field
estimates.
The individual factors comprising p let for the surface; p beh × p phy were set up with
values for high, medium and low estimates of the values for each of the 13 wildlife
groups. The assumption behind choice of probability distribution for the stochastic
drawing of values plays an important role as described in Sect. 5.2.2.
P-values and ranking according to importance from the Spearman correlation
coefficient analysis for the surface compartment are presented for the parameters
used in the initial impact calculation in Fig. 5.2. If the p-value (probability of type 1
91
input parameters to ERA Acute. Where calculations were carried out in succession,
combined formulas were used. PRCC allows independent effects of each parameter
to be determined, even when the parameters are correlated. The goal is to determine
which factor, once fixed to its true value by additional research, on average leads
to the greatest reduction in the variance of an output. The interpretation of PRCCs
assumes a monotonic relationship (relationship or function which preserves a given
trend) between parameters (Marino et al. 2008). This is the case for all the (sub-)
models used in ERA Acute. The rank-transformation is done to reduce the effect of
non-linear data, and PRCC is a robust sensitivity measure for nonlinear, monotonic
relationships (Marino et al. 2008).
The result is a sensitivity index for each input parameter to the formula, which is
the fraction of the variation in the output value that can be ascribed to the different
parameters. Note that this is given the uncertainty defined by the range of natural
variation (results based on literature search) and the weight of each value given
by the distribution (uniform—equal weight). If a different distribution for the initial
random drawing of values had been used, the result would have been different.
However, given the nature of the parameters, a uniform distribution was assumed.
The use of these statistical methods in the ERA Acute sensitivity testing is
described in further detail in the project reports by Bjørgesæter and DamsgaardJensen (2018) and Stephansen and Bjørgesæter (2017).
Impact and restoration functions were tested for each compartment and for each
relevant VEC-group within the compartment having different parameter values
and/or functions.
The results from the Spearman correlation coefficient analysis are presented in the
test reports by Bjørgesæter and Damsgaard-Jensen (2018) for surface, water column
and shoreline compartments and Stephansen and Bjørgesæter (2017) for the seafloor
compartment.
5.2.3 Example from Surface Compartment
For the deterministic testing, all parameter values used for the wildlife groups are
available in the test report (Bjørgesæter and Damsgaard-Jensen 2018), as well as
figures showing the results for each of the tested parameters. As part of the testing
it was determined that the equation which includes the exposure time (N-let2)
(Sect. 3.3.1) performs best according to the impact estimated from various field
estimates.
The individual factors comprising p let for the surface; p beh × p phy were set up with
values for high, medium and low estimates of the values for each of the 13 wildlife
groups. The assumption behind choice of probability distribution for the stochastic
drawing of values plays an important role as described in Sect. 5.2.2.
P-values and ranking according to importance from the Spearman correlation
coefficient analysis for the surface compartment are presented for the parameters
used in the initial impact calculation in Fig. 5.2. If the p-value (probability of type 1