The operational RSM with a 10-km resolution
(Fig. 1)
is evaluated by forecasters on a daily basis and in a few case
studies of heavy-rainfall events (Wang et al. 1998). In this work,
the non-hydrostatic mesoscale spectral model (MSM) (Juang 2000)
has been coupled with the modified Oregon State University land
surface model (LSM) originally developed by Pan and Mahrt (1987)
(Chen et al. 1996; Chen and Dudhia 2001)
for three regions of the state of Hawaii:
the Molokai-Maui-Hawaii domain at a 3-km resolution, the Oahu
domain at a 1.5-km resolution, and the Kauai-Niihau domain at
a 1.5-km resolution (Zhang et al. 2000; Chen et al. 2002) (
Fig. 2,
Fig. 3, and
Fig. 4).
Since the summer of 2001, we have been conducting the daily
high-resolution (1.5 km) experimental forecasts for Oahu using the
coupled MSM/LSM. The model forecasts are available at our website
and are evaluated by forecasters in real time. In early 2003,
we started to conduct the daily experimental forecasts for the
Molokai-Maui-Hawaii domain at a 3-km resolution using the coupled
MSM/LSM upon the completion of the 1.5-km Oahu forecast.
We have assessed the impact of improved representation of the
terrain and surface boundary conditions on simulating orographic
and local effects under summer trade-wind conditions, localized
rainfall and orographically amplified high winds based on a few
cases (Zhang et al. 2000; Chen et al. 2002). Evaluations of
the RSM/MSM coupled with the LSM are being made for a one month
period during May-June 2002 using our archive from daily
experimental forecasts (Chen et al. 2002).
(1) Trade-wind weather
The f97 version of the RSM/MSM model contains only one soil type,
sandy-clay loam, and one vegetation type, broadleave trees, with
a constant (70%) vegetation fraction at every grid point. Our
analyses show that this version of RSM/MSM model simulated the
orographic effects reasonably well, including upstream decelerating
flow, splitting flow along the windward coasts, orographically
enhanced strong winds within the channels between the islands and
within the saddles between the mountains, and leeside vortices
(Zhang et al. 2000). However, the amplitude of the simulated
diurnal surface temperature range is smaller than observed; and
the simulated diurnal winds are also too weak as compared with
observations. For example, during the period of May 20 through June
20 2002, at Kalaeloa Airport, Oahu (PHJR,
Fig. 3),
both the 10-km RSM and 1.5-km MSM models predicted lower temperature
in the afternoon hours and higher temperature in the early morning
(Figs. 5a, b)
(UTC = HST + 10 h).
In contrast, the 1400 HST and 0200 HST surface temperature was
predicted reasonably well by the 1.5-km coupled MSM/LSM model
(Figs. 5a, b).
The initial soil moisture for the coupled MSM/LSM model is taken
from the 24-h forecast of the previous day. Without the LSM, the
1.5-km Oahu MSM model also over-predicts the amplification of
wind speed on the lee side slopes. With improved boundary conditions,
this problem was corrected in the coupled MSM/LSM model
(Fig. 5c).
This is because the roughness length used in MSM is too small in
the urban areas. We specified rather realistic roughness length
in the coupled MSM/LSM.
During 24-27 May 2002, the trade winds were weaker than normal
with a pronounced onshore wind component in the afternoon hours
(Fig. 5d).
The afternoon sea breezes during this period were better simulated
by the coupled MSM/LSM as compared with the MSM simulations
(Fig. 6).
The surface air pressure was well predicted by all three models
(not shown). For two other Oahu surface stations (Honolulu
International Airport and Kaneohe Marine Corps Air Station), the
1.5-km Oahu MSM/LSM also outperformed both the 1.5-km MSM and
the 10-km RSM models (not shown). The 1.5-km Oahu MSM/LSM forecasts
at these two sites also agree well with observations. Nevertheless,
the 1.5-km MSM/LSM over-predicts trade-wind rainfall on the
windward slopes of the Koolau mountains with very little rainfall
over the ridge axis (not shown) in contrast to observations and
climatology (Giambelluca et al. 1986) that show the maximum rainfall
occurs near or slightly downstream of the ridge. This problem
is under investigation.
Based on observations and model forecasts from the RSM, MSM and MSM/LSM,
statistics of surface variables are computed at the three surface sites:
PHNL (Honolulu International Airport), PHJR (Kalaeloa Airport) and PHNG
(Kaneohe Marine Corps Base Hawaii) (
Fig. 1 and
Fig. 3)
for 12-h, 24-h and 36-h forecasts.
They are summarized in
Table 1,
Table 2,
Table 3 and
Table 4
which contain (1) the 32-day
mean, (2) the correlation coefficients of model simulations with observations,
and (3) forecasting errors for 2-m temperature, 2-m dew point temperature,
10-m wind speed and 10-m wind direction, respectively. The forecasting
error, fe, is defined as
where A refers to any of the four surface variables, n is the number of days
of observations (n = 32), subscripts o and m respectively denote observations
and model simulations, and superscript i ranges from 1 to 32.
For 2-m temperature (
Table 1 )
and 2-m dew point temperature (
Table 2 ),
the
MSM/LSM forecasts consistently show the highest correlation coefficient with
observations, with fewer forecasting erros than the RSM and MSM forecasts.
The 32-day mean in the MSM/LSM forecasts is also the closest to the observed
mean. Cold biases are evident in the 24-h RSM and MSM forecasts. In addition,
the forecasted 2-m dew point temperatures from RSM and MSM are negatively
correlated with observations at PHJR for the 24-h forecast.
The overestimation of 10-m wind speed by the RSM and MSM is reflected in
the 32-day means at all three surface sites (
Table 3 ).
It is more severe in
the RSM simulations than in the MSM simulations. The correlation coefficients
of the RSM, MSM and MSM/LSM forecasted surface wind speed with observations
are only slightly different at each surface site. However, the MSM/LSM
simulations persistently provide the fewest forecasting errors. Amongst the
three surface sites, correlation coefficients are higher at PHNL and PHNG
than at PHJR.
Large differences in the statistics of the 10-m wind direction are noted in
Table 4.
The RSM 12-h and 36-h forecasts show a negative correlation with
observations at all three surface sites. The correlation coefficient of the
MSM/LSM 24-h forecasts reaches as high as 0.92 at PHNL and 0.88 at PHJR, but
is -0.09 at PHNG. It appears that there are large uncertainties in wind
direction forecasts.
(b) High impact weather
We have conducted numerical simulations for several high-wind
and heavy-rainfall events throughout the entire state of Hawaii.
High-wind events frequently occur after the passage of a mid-
latitude cold front over the central North Pacific with a
pronounced high-pressure cell north of the island chain. Our case
studies provide convincing evidence that high-resolution numerical
models with improved surface conditions are required to
successfully simulate these events (Chen et al. 1999; Zhang et al.
2000). We are summarizing our results for possible publication. A
high-wind case and a heavy-rainfall case are briefly presented
as examples.
(a) A high-wind event
During 14-15 February 2001, a high-pressure cell behind a
midlatitude cold front that moved north of the island chain
generated strong winds that knocked down trees and cut power
over all the major islands of Hawaii. The National Weather
Service issued a high-wind advisory for all islands on the
afternoon of 14 February. High-wind warnings were issued for
oceanic channels between major islands. The Alenuihaha Channel
between Maui and Big Island was particularly windy. From the
newspaper reports, most of the power failures and damages
occurred on the lee side. Oahu residents reported power failures
throughout the day. The largest area affected was in Kalihi
near downtown Honolulu where 6,500 customers were left in dark
after 3:55 pm. The operational 10-km RSM forecast did show strong
winds along the northern and southern corners of the
Big Island and the Alenuihaha Channel
(Fig. 7a).
Nevertheless, the winds from the 10-km RSM for the island of Oahu
were rather uniform
(Fig. 7a).
Similar to other high-wind cases, our nested high-resolution
(1.5-km) Oahu MSM/LSM experimental forecasts called for the
amplification of trade-wind flow on the lee side of the Koolau
and Waianae mountain ranges
(Figs. 7b, c)
in agreement with the damage reports and surface wind observations.
Winds are relatively weak on the windward slopes. In addition to
high winds, strong trade flow also brought in trade-wind showers
especially on the windward side, which drifted toward the lee side.
The 10-km RSM run predicted no rainfall for Oahu (not shown).
However, appreciable rainfall was predicted along the windward side
of our nested Oahu domain
(Fig. 7d)
in better agreement with the observed local weather. It is apparent
that our nested MSM/LSM Oahu domain provided better model guidance
than the 10-km operational RSM run because the terrain and orographic
effects are better resolved by the MSM/LSM. Furthermore, without
the LSM, the MSM over-predicts the wind speed along the leeside slopes
as compared with the surface wind reports (not shown).
(b) A heavy-rainfall event
The recent Hilo flood during 1-2 November 2000 is the most
devastating rain storm to hit the area in two decades. The
synoptic-scale pattern was well predicted by the AVN run but
the excessive localized rainfall, >700 mm in 24 h
(Fig. 8a),
was completely missed. In contrast, our 3-km MSM/LSM simulated
the localized rainfall reasonably well
(Fig. 8b).
The prevailing winds were from east-southeast (not shown) with
orographically enhanced precipitation on the windward slopes.
The simulated 24-h rainfall accumulation has maxima exceeding 550 mm
in the Hilo area and Volcanoes Park area in reasonable good
agreement with the observed rainfall. The orographic enhancement
was only partly resolved by the 10-km RSM with a maximum rainfall
~ 80 mm along the Hilo coast
(Fig. 8c).
Furthermore, without the land surface model the simulated rainfall
is about 20% less than the MSM/LSM 24-h rainfall prediction with
a 24-h maximum ~450 mm
(Fig. 8d).
For this case, in addition to favorable large-scale conditions,
orographic effects are very important for the development of
excessive localized heavy rainshowers. The local circulations
related to atmosphere/land interactions also affect the rainfall
rate but are only of secondary importance.