Report from the workshop on "Intercomparison..."

Report from the workshop on
Intercomparison of Advanced Practical Short-Range Atmospheric Dispersion Models held in Manno, Switzerland - August/September 1993)

The text below was written shortly after the workshop, and has only been slightly changed since.

In 1991, a European initiative was launched for increased cooperation and standardisation of atmospheric dispersion models for regulatory purposes. It was decided to arrange a series of workshops dealing with different aspects of this issue. The first workshop was held in 1992 in Denmark, while the second took place August 29 - September 3, 1993 in Manno, Switzerland. The second workshop was organized by ERCOFTAC (European Research Community On Flow Turbulence And Combustion). Also EURASAP is engaged in this series of activities, being a co-sponsor of the first workshop and a subsequent workshop to be held next autumn in Belgium.

In the USA, much attention has been given to the subject of model evaluation, whereas in Europe studies in this field have not been so numerous. However, a major concern of the Manno workshop was to investigate model evaluation procedures and discuss the problems involved in the process - based on practical experience. The workshop was not an attempt to perform an in-depth evaluation of models, but can be characterized as a demonstration exercise. It was the intention to seek agreement on methodologies which can later be used for studies involving more data sets and more models.

Before the workshop, data sets from three atmospheric field experiments (Kincaid, USA; Copenhagen, Denmark; Lillestrom, Norway) were distributed to the participants. They were accompanied by a package of model evaluation software (developed by Sigma Research Inc.). The software package includes the options of producing an operational evaluation as well as pursuing a diagnostic approach to model evaluation. It further includes plotting software suited to the purpose of model evaluation.

The operational (statistical) evaluation results in a number of statistical performance measures such as mean fractional bias (FB), normalized mean square error (NMSE), fraction of observations within a factor of two (FA2) etc. Further, the software package offers the possibility of computing confidence limits on these measures, making use of a bootstrap resampling procedure.

The diagnostic (scientific) approach aims at obtaining an understanding of cause-and-effect relationships by analysis of model residuals, stratified by primary parameters.

These two approaches complement each other. A regulator may have his primary interest in an operational evaluation, but the diagnostic approach should not be forgotten. In order to have confidence in a model, it should be shown to give the right answer for the right reason, which can only be accomplished by a diagnostic approach.

The data sets selected for the workshop were based on field tracer experiments where the monitoring networks were dense. Therefore, it was possible to determine maximum concentrations along crosswind arcs as well as crosswind integrated concentrations at different distances. Thus, the two basic concentration variables considered were maximum arc-wise concentrations and crosswind integrated concentrations. Modelled values of these parameters were compared to observed values.

It was agreed that the performance measures computed by the software package (FB, NMSE, FA2) are generally useful when comparing models. However, it must be recognized that model performance as quantified by these measures is very much dependent on the processing of input data, including data selection. Therefore, a fair comparison between models must be based on a common protocol.

There was general satisfaction with the software package used for model evaluation, though a few additions to it were suggested.

For the workshop, a number of modern gaussian plume models based on boundary layer parameterisation were tested (the Danish OML, the American HPDM, the British UK-ADMS). Also, a more traditional Pasquill-Turner-based model, the American ISC2 was considered.

To be very brief it can be stated that in many respects the newer models performed similarly, tending to have their highest values closer to the source than traditional Pasquill-Turner based models. The ISC2 model performed poorly, but the results presented must be considered preliminary.

During the workshop, a number of unanswered questions emerged regarding model performance on the specific data sets. Therefore, a subsequent workshop to be held in Belgium in November 1994 will continue the work of the present workshop.

At the Manno workshop, a number of presentations were given on subjects related the above mentioned theme - some, however, focusing more on models to be used in complex terrain.

The volume of proceedings from the workshop is available from C. Cuvelier (editor), JRC Ispra, TP 690, 21020 Ispra, Italy.

This page is maintained by Helge RÝrdam Olesen

It was last modified on December 9, 1997

Back to Homepage of the
Initiative on Harmonisation within Atmospheric Dispersion Modelling for Regulatory Purposes