Page 10 - Volume 4, Issue 1, January 2011

Makowski

8

56 h, the 10% percentile was equal to 43.9 h

and the 90% percentile was equal 128.7 h.

The high coefficient of variation of the dis-

tribution of

W

(47%) and the large differ-

ence between the 10% and 90% percentiles

showed that a moderate uncertainty on pa-

rameter values (±20% around the estimated

values) can induce a large uncertainty about

wetness duration requirement for infection.

In order to identify the main sources of

uncertainty, sensitivity indices were com-

puted for the five model parameters for sev-

eral temperatures

T

. Sensitivity of the model

output

W

to parameter values were mea-

sured by calculating correlations between

W

and parameter values using the 10,000

Monte Carlo simulations. Results are shown

in Figure 2 for all parameters in function of

T

. A correlation close to +1 or -1 indicates a

strong influence of the parameter on the

model output. A correlation close to zero in-

dicates that the parameter is not influential.

More sophisticated sensitivity indices could

had been computed (Saltelli

et al

., 2000), but

correlation-based indices were considered

here because of their simplicity and intui-

tive interpretation.

Figure 2 showed that correlation be-

Figure 2.

Sensitivity indices for the five model parameters in function of the temperature. Sensitivity indices correspond

to correlations between parameter values and wetness duration requirement estimated from 10,000 Monte Carlo simula-

tions.