Environmental Engineering Reference
In-Depth Information
2. Methodology
Two multi-dimensional interpolating formulations are considered: the first one is
based on the well-known inverse-squared distance weighting (ISD) interpolator
while the second one is implemented using a radial basis functions (RBF) neural
network (Nabney, 2001). In both formulations, the metamodel can be viewed as a
multi-dimensional interpolation scheme that given a state of “input” conditions
x
= (
u
in1,2
,
α
in1,2
,
Ε
in1,2
) provides the “output” state
y
= (
u
out1,2
,
α
out1,2
,
Ε
out1,2
)
consisting of the wind velocity (
u
), wind direction (
a
) and TKE (
E
) vertical
profiles, averaged over the two inflow and outflow lateral boundaries of a given
microscale computational domain. For each metamodel, a calibration set of input
and output states (
x
k
,
y
k
),
k
= 1..
N
cal
has to be determined by explicitly simulating
N
cal
microscale cases during the initialisation stage.
The calibration set is constructed starting from a small number of urban areas,
with linear dimensions of the order of a few hundred meters, which are selected as
representative of the urban geometry and subsequently simulated with MIMO for
specific sampling periods. Area-averaged inflow and outflow profiles are extracted
from the calculated fields for each period and used as the actual calibration points
for each metamodel class. Following appropriate discretisation and classification
of the entire urban area, metamodel instances are associated to the rest of the
urban cells of the mesoscale grid, after rotations and vertical scaling are applied to
account for differences in street geometry, orientation and average building height
in each cell. During the coupled operation of the model, each metamodel instance
receives input from the inflow boundaries of the corresponding mesoscale cell and
in turn produces the extra forcing terms, approximating the effect of local urban
structures. These are introduced back into the mesoscale domain by means of
Newtonian relaxation terms.
3. Applications: The Athens Case
The performance of the coupled MEMO-MIMO model system was assessed using
a 5-day case between 8 May and 12 May 2002 for the greater Athens area. A set
of four calibration cases, subsequently labelled (1)-(4), was constructed based on
microscale simulations of two representative urban areas, located near the shore of
Piraeus and over Patision and Fokionos Negri streets, respectively. High-resolution
microscale computational grids were constructed from three-dimensional Geo-
graphical Information System (GIS) data of the building and road structure. An
example of calculated two-dimensional flow field at a horizontal level of 15 m
over the sample area of Patision, is shown in
Fig. 1a.
The differences between the wind speed and TKE fields calculated by the
coupled system and the corresponding fields of the standard MEMO simulation
are illustrated in
Fig. 1b and c.
A significant reduction of the predicted wind speed
