Documentation of the Evaluation of CALPUFF and Other Long ...

a quarter of the models had FMS>55%, with a maximum FMS value of 71%. In ETEX, at 36
hours one tenth of the models had a zero FMS (i.e., no overlap of the predicted and observed
tracer cloud) and a quarter had an FMS>45%, with a maximum FMS of 67%. At 60 hours in
ATMES‐II half of the models (against only a quarter of the models of ETEX) had a FMS>30% and
the maximum FMS was 58%, while the maximum FMS for ETEX models was 52%.
Temporal Analysis: The temporal analysis was carried out at two arcs of receptors at distances
of approximately 600 and 1,200‐1,400 km from the release point. In general, the LRT models
were better at predicting the time of arrival, duration and peak concentration of the tracer
cloud for the central stations of the two arcs, and less satisfactory for the external stations. The
Figure of Merit in Time (FMT, see Section 2.4 for definition) the best performances were
observed for the central stations of the two arcs. For all the stations selected for the time
analysis, FMT of models in ATMES‐II improved when compared to the first ETEX release
exercise.
Global Statistics: The global statistical indexes also indicate a general improvement of models'
performance in ATMES‐II compared to the ETEX real time modeling exercise. For instance, only
eight models out of 49 (16%) had a bias higher than 0.4 ngm ‐3 (400 pg/m 3 ) in absolute value;
the number of models above the same threshold in ETEX real time was 24 out of the 28 (86%)
participants. Almost all models showed a satisfactory agreement with the measured values.
However, few models were distinguished by a particularly good (or bad) performance in all
respects. More than half of the models showed a relatively small error (NMSE), indicating a
limited spread of the predictions around the corresponding measurements. Again, while in the
ETEX real‐time exercise only four models had an NMSE less than 100, 42 models were below
this threshold in ATMES‐II. Improvements compared to ETEX could also be seen in the number
of predicted and observed pairs within a factor of 2 (FA2) and 5 (FA5) of each other; whereas in
ATMES‐II half of the models had FA5>45%, in ETEX no model reached that value. There was no
negative Pearson correlation coefficient, with the best models showing values slightly less than
0.7.
Conclusions: The three main original objectives of ETEX as follows:
• to test the capability of institutes involved in emergency response to produce predictions
of the cloud evolution in real‐time;
• to evaluate the validity of their predictions; and
• to assemble a database that allows the evaluation of long‐range atmospheric dispersion
models.
11