Environmental Engineering Reference
In-Depth Information
For further investigation we calculated also other indicators, namely: FMS
(figure of merit in space), FA2, FA5 (factors of 2 and 5), FOEX (factor of excess)
and the bias.
Analysis of FMS, FA2 and FA5 shows that the multi-model median slightly
over-performed the other ensemble sets. However FOEX is much better for 75th-
percentiles, which is the consequence of the already mentioned under-prediction.
On the other hand all the biases are positive and the lowest for the medians and
averages.
The most important part of the analysis was the verification of the results on
the maps. In particular we analyzed space overlap for time integrated concentration
at the end of simulation time against measurement for the threshold 1 × 10
−10
g/m
3
. First of all it can be seen that the simulations based on older meteorological
datasets performed worse than the ones driven by newer weather forecasts.
The multi-model median has higher score for space overlap by few percents than
the best EPS-based ensemble medians and it seems that the shape of the plume
generated by multi-model median is closer to measurement than for EPS-based
ensemble median.
One should keep in mind however, that multi-model simulations were driven
by re-analyzed meteorological fields. Thus the conclusion can be also like that: the
atmospheric dispersion system based on the EPS meteorological data has capabilities
similar to the models utilizing meteorological data from the re-analysis.
3. Conclusions
Summarizing the analysis we can conclude that:
−
In principle there is an overall equivalence in the prediction capacities of
the two techniques.
−
The multi-model median in a number of aspects over-performed other
representations of the ensembles.
−
Taking into account that in this analysis we considered multi-model
simulations based on re-analyzed meteorological fields while the EPS-
based ones used normal forecasts, the systems using EPS also demonstrated
strong capabilities.
−
In general the median can be considered as a rational choice to represent
the ensemble, although the average particularly for EPS-based ensemble
could be equally good.
Taking into account these conclusions an interesting suggestion could be to
combine these two approaches i.e. to create multi-model ensembles based on EPS
meteorological data. The advantage of this type of methodology is such that one
can easy obtain uncertainty information (like variance) from EPS-based simulations
for each model and then try to make an optimal combination of models results
using these uncertainties.
