AirWare Reference Manual: Model Accuracy

AirWare On-line Reference Manual

Release Level 5.4
Release Date 2008 10

Revision Level 1.0

Last modified on: Monday, 20-Jul-09 14:03 CEST

Model Accuracy

Model accuracy is usualy determined by comparing model results with air quality observations; the basic problems encountered are primarily related to

the intrinsic uncertainty of the oberservations themselves and their sample nature;
the vastly different scales involved when observing and modelling a turbulent process.

While monitoring data describe concentrations over time at a point location based on small volumes (several liter) of air samples, the model estimates average concentration within a time step (e.g., an hour) for a comparatively very large volume of air (usually several million liters), under assumptions of complete mixing within the model grid cell/volume or as resulting from a (tenous at best) steady state assumption.

For regulatory purposes, the expectations on model accuracy are defined in

Council Directive 1999/30/EC

Data Quality Objectives and Compilation of Air Quality Assessment:

The following data quality objectives for the required accuracy of assessment methods ...
are laid down to guide quality-assurance programmes.

Modelling SO2, NO2, NOx PMx, Pb
Accuracy:
Hourly averages 50 - 60 % not defined at present
Daily averages 50 % not defined at present
Annual averages 30 % 50 %

Comparison with monitoring data: model validation

AirWare offers two mechanisms to compare monitoring data and model results for the grid cells conatining the monitoring station:

Model scenario pages: the button Compliance at the model scenario levels leads to a pop-up window that providea a tabular listing of all monitoring stations in the domain with their measured values (if available for the current model time step) and the model generated values as well as the respective air quality standard for the current substance and temporal aggregation; the top of the windo shows the comparison of measured versus simulated for the 24 hourly values of the scenario.
Time series data analysis: in the time series analysis (display and analysis of a single time series) the time series of observed data can be compared with model generated values for that location if the model was run for the period selected. The two time series are shown in a common graph for direct comparison.
The tabulat summary on the left side includes a selection of (applicable) error statistics.