Requirements for the completeness of ambient air quality data sets with respect to derived parameters

Monitoring and sampling of air quality data is costly and labor intensive. The necessary efforts increase progressively with increasing accuracy requirements. Also loss of data because of instrument break down, data transmission failure, or service and calibrating procedures is more or less unavoidable. Calculation of characteristic parameters like means or percentiles as necessary for information compression and also for comparison with air quality standards do not require complete data sets, since successive primary data like half-hour means are not independent from each other. Emission patterns and periodically reappearing or comparably slowly changing transmission conditions are responsible for autocorrelation of these data. Using air quality data from the Austrian public monitoring networks for various air pollutants (NO2, SO2, CO, O3) over the last decade various patterns of data loss are simulated and used to compute air quality parameters (fractiles, semi-annual means, daily means). The variation interval of these parameters is compared to equivalent parameters resulting from the complete data sets. Furthermore, autocorrelation functions of these data are calculated and discussed briefly. Finally, the applicability of the parameters obtained from truncated data sets for air quality management decisions is discussed and compared to the Austrian standard. The results indicate an error of a few percent — depending on the type of data loss — if these parameters are computed from incomplete data sets up to 50% data loss. Thus reduction of monitoring efforts without substantial loss of information is possible