Strategy for Regional Seasonal Forecasts
193
Fig. 10.6 shows a comparison ofresults from the four models. It appears that the
48 km Eta-model performed best compared with the observations (top). The superior performance ofthe 48 km Eta model is particularly evident in the precipitation
forecasts (not shown). A one-year investigation was conducted on the precipitation
forecasts, using the so-called "threat score", in which the rates of precipitation are
divided into 8 categories. The results show that the 48 km Eta-model was the best
in aU categories.
10.5 Long-range prediction
10.5.1 Current state of monthly forecasts in general
Here we utilize monthly forecasts of the Japan Meteorological Agency (courtesy
of Drs. K. Takano and T. Tokioka). Their results show that, for forecasts longer
than about five days, ensemble forecasts are needed and are indeed indispensable.
For forecasts up to one month, it is important to select the most appropriate and
most efficient ensemble of the initial conditions (about 10 members according to
the Japanese Meteorological Agency). The forecast skill depends on the quality of
members in the ensemble initial condition. To achieve maximum skill, it is necessary to utilize advanced knowledge ofthe linear dynamics (Singular Vector Analysis) as to the rate of growth of disturbances.
In contrast, 3 months forecasts may not need this advanced knowledge, because
linear dynamics will affect only the first month. Thus, a high level ofknowledge of
the applied mathematics relevant to this issue may not be required for seasonal
forecasts, but multiple forecasts (an ensemble) is absolutely needed (Brankovic and
Palmer 1997).
10.5.2 Teleconnections from the tropics to the Mediterranean Sea
Although Europe is somewhat remote from the ENSO region, there is stiH some
ENSO influence on Europe. Ward (1996) showed observational evidence for significant correlations of Sahel rainfaU and meteorological variables over the eastem
Mediterranean Sea with the Indian monsoon rainfall. This teleconnection can be
explained by linear dynamics (Rodwell and Hoskins 1996). It is hoped that Atlantic
SST anomalies also influence climate of the Mediterranean Sea in a systematic
way.
10.6 Regionallong-range prediction
10.6.1 Limited domain 30 day predictions over complex terrain
Giorgi and Bates (1989) made 30 day simulations, for Jan 1979 over the westem
US. Their model was not nested in another, but the initial conditions and lateral
boundary conditions were given by the ECMWF large scale analysis of the obser-
193
Fig. 10.6 shows a comparison ofresults from the four models. It appears that the
48 km Eta-model performed best compared with the observations (top). The superior performance ofthe 48 km Eta model is particularly evident in the precipitation
forecasts (not shown). A one-year investigation was conducted on the precipitation
forecasts, using the so-called "threat score", in which the rates of precipitation are
divided into 8 categories. The results show that the 48 km Eta-model was the best
in aU categories.
10.5 Long-range prediction
10.5.1 Current state of monthly forecasts in general
Here we utilize monthly forecasts of the Japan Meteorological Agency (courtesy
of Drs. K. Takano and T. Tokioka). Their results show that, for forecasts longer
than about five days, ensemble forecasts are needed and are indeed indispensable.
For forecasts up to one month, it is important to select the most appropriate and
most efficient ensemble of the initial conditions (about 10 members according to
the Japanese Meteorological Agency). The forecast skill depends on the quality of
members in the ensemble initial condition. To achieve maximum skill, it is necessary to utilize advanced knowledge ofthe linear dynamics (Singular Vector Analysis) as to the rate of growth of disturbances.
In contrast, 3 months forecasts may not need this advanced knowledge, because
linear dynamics will affect only the first month. Thus, a high level ofknowledge of
the applied mathematics relevant to this issue may not be required for seasonal
forecasts, but multiple forecasts (an ensemble) is absolutely needed (Brankovic and
Palmer 1997).
10.5.2 Teleconnections from the tropics to the Mediterranean Sea
Although Europe is somewhat remote from the ENSO region, there is stiH some
ENSO influence on Europe. Ward (1996) showed observational evidence for significant correlations of Sahel rainfaU and meteorological variables over the eastem
Mediterranean Sea with the Indian monsoon rainfall. This teleconnection can be
explained by linear dynamics (Rodwell and Hoskins 1996). It is hoped that Atlantic
SST anomalies also influence climate of the Mediterranean Sea in a systematic
way.
10.6 Regionallong-range prediction
10.6.1 Limited domain 30 day predictions over complex terrain
Giorgi and Bates (1989) made 30 day simulations, for Jan 1979 over the westem
US. Their model was not nested in another, but the initial conditions and lateral
boundary conditions were given by the ECMWF large scale analysis of the obser-