Geoscience Reference
In-Depth Information
Even though the magnitude of the squared SD error (Fig. 25.2 c) in the highest ter-
rain region (E: west Himalayas mountains) is near equivalent to that of the forecast
bias, it is very small in the other areas. However, the biases and corresponding MSEs
are comparable in magnitude over most of the other mountain areas (Fig. 25.2 b).
The fraction of squared biases to the MSEs (Fig. 25.2 d) is greater than 50 % in
most of the areas, which showed clearly that a large contribution to the total model
forecast errors in these regions are derived from a systematic model error. The result
indicates an apparent model deficiency in the description of surface temperature in
high terrain areas.
To illustrate the above point, squared biases, MSEs, and squared SD error in the
whole SWA region are depicted in Fig. 25.3 . For the 24-h forecasts, the total model
forecast errors are dominated by the model systematic errors (Fig. 25.3 a-c). The
fraction of squared biases to the MSEs (Fig. 25.3 d) exceeds 50 %; the distribution
of total model forecast errors is also dependent on the topography of the model
domain (Fig. 25.3 a, b vs. Fig. 25.1 ). The 48-h forecast errors are a little higher than
the 24-h forecast errors (Fig. 25.3 e-h).
25.5.1.2
Precipitation
In contrast, the precipitation MSEs in the 24-h forecasts are dominated by squared
SD error (Fig. 25.4 ) over all nine selected sub-regions. The biases are not correlated
to the height of terrain. The maximum of the squared bias (Fig. 25.4 a) over the
highest terrain region is much smaller than the squared SD error. The fraction of
squared biases to the MSE (Fig. 25.4 d) is far less than 50 % in all selected sub-
regions, which showed clearly that a larger contribution to the total model forecast
error is from a nonsystematic model error. These results indicate an apparent model
problem in the description of the initial conditions or the model resolution. The 48-h
forecast errors are much higher than the 24-h forecast errors in most of areas.
For the whole study domain, the MSEs in the 24-h forecasts are obviously
dominated by the model nonsystematic errors (Fig. 25.5 a-c). The fraction of
squared biases to the MSEs (Fig. 25.5 d) is under 50 % except for some Himalaya
mountain areas. The distribution of total model forecast errors has nothing to do
with the structure of higher terrain. The areas of 48-h forecast errors greater than
20
mm 2 clearly extend over wider areas in the model domain (Fig. 25.5 e-h).
25.5.1.3
Wind Speed at 10-M
Similar to precipitation, the MSEs in 10-m wind speed are largely associated with
the nonsystematic errors in most of the sub-regions (Fig. 25.6 a-c). The largest
model bias occurs over northwestern Iran (B) (Fig. 25.6 a). The biases over the west
Himalaya mountain region (E), east Saudi Arabia (G) and west India (I) are almost
zero. The fractions of squared biases to the MSEs (Fig. 25.6 d) are under 50 % over
all selected areas.
Search WWH ::




Custom Search