4.2. Precision as function of sampling frequency
It was shown that the long-term weekly grab sampling strategy yielded the best estimate of the true EF. However, this sampling strategy took the maximum number of weeks into account. Since sampling for such a long period would be costly, simulations were made to determine if a shorter sampling frequency (i.e. taking grab samples for fewer weeks) would also be sufficient. Since weekly grab sampling on a random working day yielded equally good results as weekly grab sampling on a specific day of the week, it was decided to test the shorter sampling frequency for a random working day only. Decreasing the number of consecutive sampling weeks increased the relative error gradually. For 15 weeks of sampling, the relative error was between −25% and 25% (Fig. 11). With less than 15 sampling weeks, the relative error increased rapidly. If grab sampling was performed for 25 consecutive weeks on a random working day, an EF was estimated with a the relative error between −12% and 25% for dataset 1 and a relative error which has 95% chance of lying between −17% and 13% for dataset 2. If the same number of grab samples was taken completely at random (i.e. not systematically one measurement every week), the resulting estimated EF had a relative error which has 95% chance of lying between −10% and 24% (dataset 1) and between −16% and 15% (dataset 2). So, a similar precision was obtained with completely random grab samples as with grab samples every week. If the number of random grab samples or the number of consecutive weeks decreased even further, the random grab sampling strategy had a better precision as compared to grab sampling each week (Fig. 10 and Fig. 11).