16 
Verification of 6 У г - months’ operational forecasts 
5 Verification of 6 1 /2-months’ operational forecasts 
A comparison is made of DMO, combination forecasts (Kombi) and Surge, as well as Mix, an arithmetic 
mean of these three predictions. Wherever a man-machine mix of the BSH water level forecast service 
was available, it has been included in the verification. Four daily forecasts since the beginning of 
operational use, at 05, 11, 17 and 23 UTC in the format shown below (Tab. 8), were included in the 
evaluation. The forecasts made at other times hardly differ from those above because new DMO or GFS 
input is only available every 6 hours. 
Tab. 8: Wind surge observations (fourth line) and forecasts (lines 5-9) for Cuxhaven harbour [cm] (left side) together with 
corresponding error squares [cm 2 ] (right side). Forecasts are given with lead times 1 to 33 hours (second line) for three high 
and low waters (third line) after 20 August 2009, 23 UTC. 
1 
Cux : 
Issue: 
20090820 
2 3 z 
2D-Modell: 18z 
GFS: 18 z 
Aberdeen 
18 z 
Borkum: 
2 2 z 
2 
LeadT 
1 
8 
13 
20 
26 
33 
1 
8 
13 
20 
26 
33 
3 
HHMM 
0025 
0714 
1236 
1946 
0112 
0801 
0025 
0714 
1236 
1946 
0112 
0801 
4 
Obs 
7 
12 
10 
23 
12 
28 
5 
МММ 
1 
7 
15 
15 
7 
175 
36 
25 
25 
64 
25 
6 
DMO 
3 
5 
17 
13 
9 
10 
223 
16 
49 
49 
100 
9 
324 
7 
Mix 
9 
6 
14 
18 
10 
10 
76 
4 
36 
16 
25 
4 
324 
8 
Kombi 
13 
7 
9 
23 
10 
10 
66 
36 
25 
1 
0 
4 
324 
9 
Surge 
12 
7 
16 
17 
10 
10 
126 
25 
25 
36 
36 
4 
324 
The error squares of the predictions to the left are shown on the right side. The added values are in the 
middle, not taking into account the last column on the right side because no МММ was made for 
T+33h. 
A complete evaluation of all prediction times from 14 July 2009 to 30 January 2010 results in the follow 
ing RV reductions: DMO 0% (reference prediction); Mix 39%, Kombi 42%, Surge 39% (Tab. 9). 
The verification period started with 1 1 /г summer months of low dynamics and good results - though 
below expectations - followed by a long autumn season with mostly variable weather and westerly 
flow. DMO was extraordinarily accurate in this period, and MOS over-corrected its initialisation errors 
on average, which led to an RV(*,DMO) far below expectations. Only Mix was capable of reducing more 
than 10% error variance in autumn. The situation changed dramatically at the end of the verification 
period , when an unusually severe cold spell with prevailing E-SE winds set in. The model had huge 
initialisation errors which were easily corrected by MOS, resulting in a reduction of variance by more 
than % during the last three weeks of the verification period. 
Table 9: Sum of Squared Errors (SSE) and corresponding RV values in different time periods. 
RMSE/cm RV(*,DMO) 
n 
DMO 
Mix 
Kombi 
Surge 
Mix 
Kombi 
Surge 
14Jul0 9 
- 01Sep09 
981 
10.2 
8.5 
8.0 
8.9 
30% 
38% 
23% 
02Sep09 
- 08Dez09 
1960 
11.5 
10.7 
11.2 
11.2 
14% 
6% 
6% 
09Dez09 
- 08Janl0 
586 
14.4 
10.2 
10.4 
10.3 
49% 
48% 
48% 
09Janl0 
- 30Janl0 
429 
24.5 
14.6 
11.8 
12.0 
64% 
77% 
76% 
14Jul0 9 
- 30Janl0 
3956 
13.7 
10.6 
10.4 
10.6 
39% 
42% 
39%