Strategy for Regional Seasonal Forecasts
195
driver model, i.e., the GCM. In this procedure, the regional model restarts every 48
hours with the GCM output as its new initial conditions, and the results at the
beginning of each integration period are largely influenced by the spin-up of the
regional model.
In order to show the merit ofthe nested model runs., the observations, the nested
model simulations, and the original GCM simulations (top, ndddle, bottom) for the
JJA precipitation are compared (Fig. 10.8).
This procedure is quite different from the method of Giorgi and Bates (1989).
However, it is not certain whether this approach is better or not (see the final comment in section 10.7). Perhaps the most important point in their study is that a sufficiently 1arge area is adopted. It is interesting to note, however, that, a1though the
eastem Pacific was not included in the regional models domain, the impact from
this region is transmitted through the lateral boundary conditions.
In order to investigate the interannual variability of the monsoon, JJA of 1987
and JJA 1988 were compared from integration initialized in the mid-April. 3 runs
were done for each year, starting at 14, 15 and 16 April of each year. The nested
model did a much better job of simulating the interannual variability in the monsoon between 1987 and 1988 than did the GCM alone.
10.7 Conclusions and commcnts
For Question 1, a larger region than the Mediterranean is required, such as 25°S70 0 N, 120 0 W-00-1300E. The effect ofthe Indian monsoon as well as the influence
of the Atlantic Ocean and ENSO should then be included through the lateral
boundary conditions.
Comments What is the best approach to the ne sting? Is it that of Ji-Vemekar
(1997) or that of Giorgi-Bates (1991)? According to Fennessy (COLA-personal
communication), updating the interior of the regional model every 2 days is not
necessary with the latest versions of the Eta model. In other words, the initial condition is specified only at t=O.
Recent results relevant to the limited domain seasonal forecasts have come from
BALTEX (Bengtsson and Wang 1998). A moisture budget was calculated using a
regional model forced by obserations ofthe in/out-fluxes ofmoisture along the lateral boundaries over the Baltic Sea. The experiment was set up by using continuous
boundary values from Deutsche Wetter Dienst 6 hourly operational analysis. The
model was run for 3 months (August 1 - October 31,1995) without restart. Resolution was 24 levels and 116° lat./lon .. The simulated precipitation agrees well with
the observed rainfall in that region. Thus there are many examples that the regional
seasonal forecasts are possible.
However, as a final remark , it must be kept in mind that, a spectral model may
be another choice (see Juang et al. 1994; 1997).
For Question 2, a smaller region can be taken, for research, such as Equator60
o
N, 10 0 W-600E. However, this region is so small that realistic forecasts cannot
195
driver model, i.e., the GCM. In this procedure, the regional model restarts every 48
hours with the GCM output as its new initial conditions, and the results at the
beginning of each integration period are largely influenced by the spin-up of the
regional model.
In order to show the merit ofthe nested model runs., the observations, the nested
model simulations, and the original GCM simulations (top, ndddle, bottom) for the
JJA precipitation are compared (Fig. 10.8).
This procedure is quite different from the method of Giorgi and Bates (1989).
However, it is not certain whether this approach is better or not (see the final comment in section 10.7). Perhaps the most important point in their study is that a sufficiently 1arge area is adopted. It is interesting to note, however, that, a1though the
eastem Pacific was not included in the regional models domain, the impact from
this region is transmitted through the lateral boundary conditions.
In order to investigate the interannual variability of the monsoon, JJA of 1987
and JJA 1988 were compared from integration initialized in the mid-April. 3 runs
were done for each year, starting at 14, 15 and 16 April of each year. The nested
model did a much better job of simulating the interannual variability in the monsoon between 1987 and 1988 than did the GCM alone.
10.7 Conclusions and commcnts
For Question 1, a larger region than the Mediterranean is required, such as 25°S70 0 N, 120 0 W-00-1300E. The effect ofthe Indian monsoon as well as the influence
of the Atlantic Ocean and ENSO should then be included through the lateral
boundary conditions.
Comments What is the best approach to the ne sting? Is it that of Ji-Vemekar
(1997) or that of Giorgi-Bates (1991)? According to Fennessy (COLA-personal
communication), updating the interior of the regional model every 2 days is not
necessary with the latest versions of the Eta model. In other words, the initial condition is specified only at t=O.
Recent results relevant to the limited domain seasonal forecasts have come from
BALTEX (Bengtsson and Wang 1998). A moisture budget was calculated using a
regional model forced by obserations ofthe in/out-fluxes ofmoisture along the lateral boundaries over the Baltic Sea. The experiment was set up by using continuous
boundary values from Deutsche Wetter Dienst 6 hourly operational analysis. The
model was run for 3 months (August 1 - October 31,1995) without restart. Resolution was 24 levels and 116° lat./lon .. The simulated precipitation agrees well with
the observed rainfall in that region. Thus there are many examples that the regional
seasonal forecasts are possible.
However, as a final remark , it must be kept in mind that, a spectral model may
be another choice (see Juang et al. 1994; 1997).
For Question 2, a smaller region can be taken, for research, such as Equator60
o
N, 10 0 W-600E. However, this region is so small that realistic forecasts cannot