Development results 
15 
The main variance reducing components are: 
1. Classification, at 28%. 
2. Pers2D correction, with an average 26%, but decreasing strongly with increasing length of the 
prediction period, from 45% at T+3 hours to 15% at T+33 hours. 
3. The 1 -predictor MOS equation. It eliminates the systematic error of DMO, which is about +4 cm, 
and optimises the standard deviation of DMO through a coefficient not equal to 1. The coefficient 
in this equation decreases from 1.00 at Fp = 1 hour to 0.88 at Fp = 33 hours and, from about T+10 
hours, it achieves 2-digit reductions in the percentage of variance with this incipient convergence 
towards the expected climatological value alone. Averaged over phase 2, this is as much as 12%. 
DMO thus plays an outstanding role as a predictor. Unlike other MOS systems used, e.g., to forecast 
air temperatures, where the quality of MOS predictions mostly deteriorates only negligibly in the 
absence of DMO, and where DMO predictors rarely account for more than 50% of the total reduction 
of the error variance, the BSH MOS equations without DMO on average only would have the lower 
quality of pure DMO, with about 60% RV less. 
Nevertheless, this is an excellent example of a symbiosis of physics and statistics.