117
point averaged forcing. Hence, one of the elements that determine surface heterogeneity,
I.e. climate spatial variability, is lost. Pitman et al. (1992) illustrated well how this
could lead to large errors. They considered a GeM cell of a few hundred km size and
performed two simulations with a stand alone version of BATS. In the first, BATS was
driven by grid-average forcing and in the second, precipit.ation was allowed to occur only
over a fraction Ie of the grid point with an intensity equal to P / Ie where P is the grid
box average precipitation. The latter configuration would be more realist.ic especially in
summer conditions, when precipitation is a highly localized process. Using BATS, Pitman
et al. (1992) showed that the surface water budget changed from evaporation-dominated
in the first experiment to runoff-dominated in the second, mostly because greater runoff
is produced at the higher precipitation rates of the second experiment. Although this
result is partially dependent on the treatment of runoff in BATS (see section 3.2.4), it
emphasizes the need to include climate redistribution, especially precipitation, in surface
process models.
An extension of the mosaic approach, which also includes climate redistribution within
the tiles, has been proposed by Leung and Ghan (1995). In their model, tiles are not based
on vegetation or surface type, but on elevation classes, and different climate forcings are
calculated for each class. In particular, the climate forcing is based on the motion of
a parcel along the sloping terrain, so that different classes have different atmospheric
temperature and orographically-induced precipitation forcing. The main drawback of the
model of Leung and Ghan (1995) is that each elevation class behaves in the same way,
regardless of it's location with respect to mountain systems (upwind or downwind), and
different vegetation types within an elevation class are not recognized.
A mosaic approach which in principle allows to overcome the difficulties of the models
above is that of Seth et al. (1994), defined the "vector" approach. In the vector model
of Seth et al. (1994) an AM grid-box is divided into a regularly spaced subgrid of N 2
elements. Each element of the sub-grid is assigned an individual surface type, elevation
and climate forcing and interacts with the atmosphere independently of the others. This
model thus allows explicit spatial redistribution of vegetation, elevation and climate forcing specific to the location of the sub-grid element. It's main drawback is that an ESEM
needs to be called for each of the N 2 grid points, which can be computationally rather
expensive. In the work of Seth et al. (1994) Ii vectorized version of BATS was developed
so that use of "vector"-BATS (or VBATS) for, say, a subgrid of 64 grid points, was not
much more expensive than use of the original BATS. However, this approach can also be
especially suitable for parallel computing architectures.
Some results from the work of Seth et al. (1994) can serve to illustrate the direct
effects of the inclusion of inter-patch heterogeneity. Three areas of :~OOx300 km 2 size
(the equivalent grid point spacing of a typical GeM) were considered in the eastern
point averaged forcing. Hence, one of the elements that determine surface heterogeneity,
I.e. climate spatial variability, is lost. Pitman et al. (1992) illustrated well how this
could lead to large errors. They considered a GeM cell of a few hundred km size and
performed two simulations with a stand alone version of BATS. In the first, BATS was
driven by grid-average forcing and in the second, precipit.ation was allowed to occur only
over a fraction Ie of the grid point with an intensity equal to P / Ie where P is the grid
box average precipitation. The latter configuration would be more realist.ic especially in
summer conditions, when precipitation is a highly localized process. Using BATS, Pitman
et al. (1992) showed that the surface water budget changed from evaporation-dominated
in the first experiment to runoff-dominated in the second, mostly because greater runoff
is produced at the higher precipitation rates of the second experiment. Although this
result is partially dependent on the treatment of runoff in BATS (see section 3.2.4), it
emphasizes the need to include climate redistribution, especially precipitation, in surface
process models.
An extension of the mosaic approach, which also includes climate redistribution within
the tiles, has been proposed by Leung and Ghan (1995). In their model, tiles are not based
on vegetation or surface type, but on elevation classes, and different climate forcings are
calculated for each class. In particular, the climate forcing is based on the motion of
a parcel along the sloping terrain, so that different classes have different atmospheric
temperature and orographically-induced precipitation forcing. The main drawback of the
model of Leung and Ghan (1995) is that each elevation class behaves in the same way,
regardless of it's location with respect to mountain systems (upwind or downwind), and
different vegetation types within an elevation class are not recognized.
A mosaic approach which in principle allows to overcome the difficulties of the models
above is that of Seth et al. (1994), defined the "vector" approach. In the vector model
of Seth et al. (1994) an AM grid-box is divided into a regularly spaced subgrid of N 2
elements. Each element of the sub-grid is assigned an individual surface type, elevation
and climate forcing and interacts with the atmosphere independently of the others. This
model thus allows explicit spatial redistribution of vegetation, elevation and climate forcing specific to the location of the sub-grid element. It's main drawback is that an ESEM
needs to be called for each of the N 2 grid points, which can be computationally rather
expensive. In the work of Seth et al. (1994) Ii vectorized version of BATS was developed
so that use of "vector"-BATS (or VBATS) for, say, a subgrid of 64 grid points, was not
much more expensive than use of the original BATS. However, this approach can also be
especially suitable for parallel computing architectures.
Some results from the work of Seth et al. (1994) can serve to illustrate the direct
effects of the inclusion of inter-patch heterogeneity. Three areas of :~OOx300 km 2 size
(the equivalent grid point spacing of a typical GeM) were considered in the eastern