2011 International Conference on Alternative Energy in Developing Countries and Emerging Economies
- 324 -
for sea surface temperature [7]. The model was run for
seven days plus one day for spin-up, repeated over an
entire year to create an annual series of hourly output.
Land cover data sets from the Global Land Cover 2000
Database [8] and the Land Development Department of
Thailand (LDD) were combined and converted to 24
USGS categories used in default MM5 inputs. The
amount of urban land surface area in the final domain
was under-represented in the default data, with just 5% of
grid cells classified as urban compared to 45% in the
modified data. The single urban category was also split
into three subcategories to account for varying building
morphology across the city. These subcategories (Fig. 2)
are identical in all aspects except aerodynamic surface
roughness, with 0.5 m for low-density and height urban
areas, 1.5 m for medium- to high-density and height
urban areas, and 3.0 m for high-rise areas. These values
were chosen from the undertaking of visual surveys,
examination of satellite imagery, and comparison with
aerodynamic surface roughness from other studies [9,
10].
Fig. 2. Urban subcategories designed for
this study, and location of the PCD Tower.
Default values for fraction of vegetation cover over the
land surface in MM5 are of low resolution and thus vary
little over the final domain. These were therefore
replaced with data originating from MODIS satellite-
derived normalized difference vegetation index data,
converted to vegetation fraction and averaged to give
monthly values, to more accurately represent vegetation
fraction.
A MODIS satellite-derived albedo product was also
incorporated, but due to a high proportion of missing data
in rainy months, tabulated rather than gridded values
were used. Seasonal averages were calculated for each
land cover category and inserted into the land surface
parameter table in MM5. Modified values over
categories containing cropland, pasture, grassland and
shrubland are lower and have less seasonal variation than
default MM5 values. This is to be expected as Bangkok
is a tropical area with a significant percentage of land
area covered with dense vegetation that varies relatively
little over the seasons. In forested and urban areas, which
both have similar vegetation cover worldwide, there was
little difference in albedo.
To test the response of the model to land surface
inputs, the model was run over two weeks in each of
January and July, both with default land surface inputs
and with the modified inputs described above. Hourly
near-surface temperature and wind speed at 100 m agl
were extracted from model output to the location of the
Bangkok Pollution Control Department (PCD)
meteorological tower (location shown in Fig. 2). These
were compared with hourly data from observations
acquired from the PCD tower, and mean bias
MB
calculated for each using
N
OP
MB
N
i
i
i
¦
)
(
,
(1)
where
P
i
and
O
i
are the predicted (modeled) and observed
values respectively at hour
i
, and
N
is the total number of
hourly prediction-observation pairs. Comparison of
default and modified inputs in Fig. 3 and Fig. 4 shows
improved temperature results with modified inputs,
particularly in the dry season (January). Although mean
bias for wind speed has little change, we used the
modified inputs for the annual simulation as they better
represent the land surface of Bangkok.
Fig. 3. Mean bias of near-surface temperature at the PCD
tower, for default and modified land surface model inputs.
Fig. 4. Mean bias of wind speed at 100 m agl at the PCD
tower, for default and modified land surface inputs.
