Low (web) Quality - BALTEX

151
and 30km). Domains D1 (90 and 60-km) and D7 (90-km)
were considered for only coarser grid distance due to
computational constraints. Domains D4 and D5 which have
same longitudinal extents but different latitudinal extents
were re-run to verify and validate results at finer grid
distance. As mentioned above, the larger scales simulated in
model dynamics will depend not only on the size of the
domain but also on the geographical coverage. The seven
domains were thus chosen to provide an ensemble of largescale
dynamics and terrain effects as affected by different
domains. For each of the domains mentioned above, the
model was integrated for 5 days starting from five different
initial conditions 6-hours apart beginning from 00UTC, 24 th
July 2005. Although the event was highly localized in both
space and time, we have considered a spatial window of
2 o x2 o centered on the general event location and a time
window of 24-hour (06UTC, 26 th July to 06UTC, 27 th July
2005) for diagnosis and analysis.
D6
D1
D4
D2
Figure 1. Experimental domains.
D5
D7
D3
5. Benchmark Simulation
Due to large number of simulations involved , high
resolution simulations (10-km) were carried out for only two
domains (D4 and D5) which are comparable in size but vary
in geographical coverage. These simulations and their
comparison with observations are then used as benchmark
for the acceptability of simulations with coarser resolutions.
We then, compared the spatial distribution of 24-hour
accumulated simulated rainfall for these domains at four
horizontal resolutions (10, 30, 60 and 90-km) and found that
a localized intense rainfall event around Mumbai was seen
for all the resolutions and for both the domains. In
particular, both high (10-km) and coarse (30-km) resolution
simulations could capture the distribution reasonably well
especially in terms intensity when compared with satellite
data.
6. Results and Discussion
Comparison of spatial distributions of 24-hour (event
window) accumulated, simulated ensemble rainfall based on
five initial conditions for five domains of 30-km resolution
with satellite data has shown considerable inter-domain
variations in the simulated distribution of rainfall.
Simulations with resolutions of 90 and 60 km also have
shown significant variations in the distribution and intensity
of the simulated rainfall with the size and geographical
coverage of the domain. This inter-domain variability in the
distribution of rainfall is profound at higher grid spacing. It
was also found that the domain D5 (which included
equatorial belt) simulated the observed distribution better
than all the other domains, and for all the three resolutions.
The effects of changes due to longitudinal or latitudinal
extents or both on the 24-hour accumulated area averaged
rainfall (R av ) and the maximum rainfall (R max ) were
analyzed. The results are from ensemble average
simulations with the five leads (initial conditions)
described above. It was shown that changes in the domain
size results in significant changes in R av and R max, it was
true for domains of different resolutions as well. This
variability of R av and R max due to change in domain size
and coverage in terms of Standard Deviation (SD) as
percentage of mean across different domains was in the
range of 30 to 40 % and 34 to 40 % respectively. It was
seen that effects of change in the domain size in general
are larger for lower resolutions. This implies the need for
an optimum model configuration with appropriate domain
size or ensemble of domains and resolution to resolve both
ends of the spectrum of scales. We next examined the
relative role of resolution on R av and R max. Dispersion of
forecasts due to changes in initial conditions provides both
a measure of reliability of the forecasts and a measure of
variation in mesoscale forecasts due to changes in the
large scale conditions (initial fields). In particular, we
expect the response to different initial conditions to be a
strong function of the domain size and geographical
coverage as the large scale fields evolve differently over
different domains. Summary of the relative variability or
sensitivity of R av and R max in terms of SD (as % mean)
with respect to different domains, resolutions and IC’s are
summarized in Table.1. As another measure of relative
sensitivity of the simulations to domain size we have
considered standard deviation (as % of mean) in terms of
24-hour accumulated total rain over event location (R T )
for different domains (with varying number of domains
depending on the resolution). Time evolution of this
standard deviation indicated that variation due to change
of domain is typically 40 % and can be as much as 70 to
80%; in comparison a change of resolution doesn’t change
this dispersion that significantly except at isolated hours
and not more than by a factor of 2.
Table 1. Summary of variability of R av and R max in terms
of SD as % of mean for different domains, resolution and
Initial Conditions.
Rain
References
Model
Domain
Resolution
Initial
Condition
R av 30-40 10-30 11-50
R max 34-40 21-43 6-40
Dudhia, J., PSU/NCAR Mesoscale Modeling System
Tutorial Class Notes and User’s Guide; MM5 Modeling
System Version 3, 2004
Giorgi, F., and L.O. Mearns, Introduction to special
section: Regional climate modeling revisited. J.Geophys.
Res., Vol.104, No. 6, pp. 6335-6352, 1999
Pielke, R.A., Mesoscale Meteorological Modeling.
Academic Press, pp. 673, 2002
Warner, T.T., Peterson, R.A. and Treadon R.E., Tutorial
on lateral boundary condition as a basic and potentially
serious limitation to regional numerical weather
prediction. Bull. of the Amer.Meteor.Soc., No.78, pp.
2599-2617, 1997