92
5 Handling Uncertainty and Sensitivity …
pbeh, 7%
pphy, 20%
Cov, 73%
Sensitivity index for Nlet-1
pbeh, 3%
pphy, 12%
Cov, 40%
Texp, 45%
Sensitivity index for Nlet-2
Parameter
N let-1
N let-2
Spearman Corr.
Coeff.
p-values
Importance
rank
Spearman Corr.
Coeff.
p-values
Importance
rank
0.19
0.00
Cov
0.13
0.00
Texp
0.39
0.00
pphy
0.28
0.00
Cov
0.84
0.00
pbeh
0.60
0.00
pphy
-
-
-
0.68
0.00
pbeh
Pbeh
Pphy
Cov
Texp
Fig. 5.2 Example of sensitivity testing: result for the sensitivity analysis of Nlet-1 and Nlet-2 for
the following random variables: p beh , p hy , coverage (Cov) and exposure time (Texp). The relative
abundance is held constant at 1
error) is below 0.05 this means that the result is statistically significant within a 95%
confidence limit. The p-values were in this case 0.001. The pie diagram shows the
sensitivity index from the FPRV.
The population loss is more sensitive to the variation in the oil drift impact parameters than to the variation in the two model parameters. This is also the case for the
impact calculations in the other compartments. In the example from the surface
compartment, the oil coverage in the grid cell is ranked as the most important variable for Nlet-1 (equation without exposure time) and as much as 73% of the total
variance observed in Nlet-1 in Fig. 5.2 can be attributed to this parameter. Therefore,
although all parameters are initially equally important, coverage and Texp (the latter
for Nlet-2) are the most important parameters. Both values have inherent uncertainty
from input parameters’ and model soundness uncertainty of the oil drift simulation
model. Logically, since coverage and exposure time represent the spreading and
degradation of oil and the results are reported for cells at different distances from the
spill and for a multitude of simulations representing very different weather conditions, it is natural that there is a large variation in the values. An inherent property
of these tests is that if a tested parameter has a high variability it also becomes more
important.
5 Handling Uncertainty and Sensitivity …
pbeh, 7%
pphy, 20%
Cov, 73%
Sensitivity index for Nlet-1
pbeh, 3%
pphy, 12%
Cov, 40%
Texp, 45%
Sensitivity index for Nlet-2
Parameter
N let-1
N let-2
Spearman Corr.
Coeff.
p-values
Importance
rank
Spearman Corr.
Coeff.
p-values
Importance
rank
0.19
0.00
Cov
0.13
0.00
Texp
0.39
0.00
pphy
0.28
0.00
Cov
0.84
0.00
pbeh
0.60
0.00
pphy
-
-
-
0.68
0.00
pbeh
Pbeh
Pphy
Cov
Texp
Fig. 5.2 Example of sensitivity testing: result for the sensitivity analysis of Nlet-1 and Nlet-2 for
the following random variables: p beh , p hy , coverage (Cov) and exposure time (Texp). The relative
abundance is held constant at 1
error) is below 0.05 this means that the result is statistically significant within a 95%
confidence limit. The p-values were in this case 0.001. The pie diagram shows the
sensitivity index from the FPRV.
The population loss is more sensitive to the variation in the oil drift impact parameters than to the variation in the two model parameters. This is also the case for the
impact calculations in the other compartments. In the example from the surface
compartment, the oil coverage in the grid cell is ranked as the most important variable for Nlet-1 (equation without exposure time) and as much as 73% of the total
variance observed in Nlet-1 in Fig. 5.2 can be attributed to this parameter. Therefore,
although all parameters are initially equally important, coverage and Texp (the latter
for Nlet-2) are the most important parameters. Both values have inherent uncertainty
from input parameters’ and model soundness uncertainty of the oil drift simulation
model. Logically, since coverage and exposure time represent the spreading and
degradation of oil and the results are reported for cells at different distances from the
spill and for a multitude of simulations representing very different weather conditions, it is natural that there is a large variation in the values. An inherent property
of these tests is that if a tested parameter has a high variability it also becomes more
important.