Geoscience Reference
In-Depth Information
25.6.2.1
Bias
The absolute bias difference between the DA and the CTRL experiment is defined
as j Bias j DA j Bias j CTRL . The 30-day's mean will be investigated first. For the 24-h
forecasts, the absolute bias difference in 2-m temperature forecast (Fig. 25.14 a)
shows that the bias is reduced in DA over most of the Southwest Asia region. The
bias in Iran, Afghanistan and Pakistan is on average
1:8 ı C less than the CTRL
forecasts, with the largest impact occurring on the south or southwest slope of the
Afghanistan Hindu Kush mountain areas (see Fig. 25.1 ).
The absolute bias difference in 10-m wind speed for 24-h forecasts (Fig. 25.14 b)
reveals that the largest impacts to DA are over the Arabian Sea, Persian Gulf and
the border between Pakistan and Afghanistan, where there are minimal high terrain
effects. Whereas the impact of the satellite data assimilation on 2-m temperature
is observed near the mountain areas, while the impact on the 10-m wind speed is
apparentin places far away from these mountain regions, and especially over water
areas. However, the evidence shows that the bias increased in many areas including
southeast Iran, northwest India and the other areas.
Compared to the 24-h forecasts in the CTRL experiment, the precipitation
forecast bias with DA (Fig. 25.14 c) decreased slightly over the Mediterranean Sea,
Black Sea coast, Saudi Arabian desert, and the Iranian Zagros mountain areas.
However, the absolute bias of the precipitation forecast for the DA experiment
increased over the Himalaya mountain areas.
For the upper levels, the absolute bias difference in 500 hPa temperature, geopo-
tential height and wind field forecasts are presented in Fig. 25.15 . The radiance data
assimilation reduces the forecast bias of the geopotential height (Fig. 25.15 b) and
wind field (Fig. 25.15 c, d) over most of Southwest Asia areas. These results show
the impact of satellite radiance data assimilation on the upper level geopotential
height and wind field forecasts are not associated with the configuration of terrain.
Meanwhile, the 500 hPa temperature forecasts are modulated by the radiance
assimilation differently (Fig. 25.15 a). Here, the satellite data assimilation does not
improve the temperature forecasts over the central Southwest Asia areas including
Saudi Arabia, Iranian Zagros mountains and Afghanistan Hindu Kush mountains.
0:3
-
25.6.2.2
Mean-Square-Error Skill Scores
Murphy ( 1988 ) found forecasting skill scores are generally defined as measures of
the relative accuracy of two forecasts, where one of the two forecasts is defined
as a “reference system”. For the following experiments, the CTRL forecasts are
considered as the reference system. Based on the mean-square-error, the skill score
(SS) can be expressed as follows:
SS
.d;r;a/ D 1 Œ
MSE
.d;a/=
MSE
.r;a/
(25.6)
Search WWH ::




Custom Search