Environmental Engineering Reference
In-Depth Information
3. Results and Discussion
The iteration converged with typically 15-30 gradient evaluations. Examples of
the estimated emission corrections are shown in Fig. 1. On certain areas, the
corrections were consistent throughout the experiment, as reflected by the average
emission correction i n Fig. 1 . Especially, the concentrations at stations near Mt.
Etna were strongly overestimated in the reference run, which resulted in up to
100% reduction in some grid cells. Total emission of SO 2 from the area near the
volcano was reduced by approximately 50-75% depending on the day. This suggests
that the parametrisation of the volcano was, at least, not representative of the time
period under consideration.
54N
54N
52N
3
2
1.75
1.5
1.25
1.05
0.95
0.75
0.5
0.25
0
52N
3
2
1.75
1.5
1.25
1.05
0.95
0.75
0.5
0.25
0
50N
50N
48N
48N
46N
46N
44N
44N
42N
42N
40N
40N
38N
38N
36N
36N
6E 8E 10E 12E 14E 16E 18E 20E 22E 24E
6E 8E 10E 12E 14E 16E 18E 20E 22E 24E
Fig. 1. Emission correction factors. Left: corrections for day 1. Right: average correction over
the period of 14 days
3.1. Effect on forecasts
For the stations in the control set, the mean RMS forecast error for each day of the
assimilation experiment is presented in Fig. 2 . Compared to the background run, a
substantial improvement is observed on average.
However, most of the improvement is concentrated on a small set of stations.
This is illustrated by right panel of Fig. 2, which shows the median with 13th and
86th percentiles of the RMS errors. Here, the difference between the reference and
assimilation runs is greatest in the upper percentile. It can therefore be concluded
that for the present case, the main benefit of data assimilation was in improvement
of the worst-case performance of the model.
A possible explanation for the uneven improvement across the stations is in the
different contributions by local and remote sources. Obviously, only the emissions
affecting the stations during the 24 h assimilation window can be adjusted by the
assimilation. Hence, the possible discrepancies in remote transport only contribute
to the optimised initial state, which in turn is likely to have only transient impact.
 
Search WWH ::




Custom Search