A
second approach to account for anisotropic variation in the data would be to
develop a directionally selective smoothing filter for the elevation layer.
The investigation of the effects of neighbourhood size on correlation, between
precipitation and orographic elevation, revealed a positive relationship between
these two variables. In effect, increasing the neighbourhood size incorporates
a greater and greater degree of information about the elevation of the surrounding
area into the elevation of the immediately local area (the grid cell). However,
from the previous investigation of the nature of anisotropic trends within the
study area, it would be reasonable to assume that the inclusion of elevation
information from east or west of a particular point would contribute an insignificant
amount of additional information compared with that gained from neighbourhood
elevation information added from north and south of the point. This is to say
that the improvement in performance by using a 10 km neighbourhood low pass
filter is likely predominantly attributable to the inclusion of 10 km of elevation
information along the north-south axis, with little or no improvement attributable
to the inclusion of 10 km of information along the east-west axis.
To accommodate for this, a new filter would be designed. It would operate in
exactly the same as the low pass filter except instead of using a square-shaped
moving window, it would utilize a rectangular moving window. If the elongated
axis of such a window was aligned with the north-south axis, then it would smooth
the elevation layer along this axis with little smoothing along the east-west
axis. This would likely be further improved upon by rotating the window such
that its long axis corresponds with the axis of the greatest elevational trend.
Conceptually, this would not be difficult to achieve though it may be difficult
to implement given that standard software does not allow this functionality,
requiring it to be custom programmed. It would also be interesting to investigate
even one step further to determine whether inclusion of southern elevation information
is significantly meaningful compared with information about the generally increasing
elevation to the north, in the current study area. From the basic model of the
orographic effect in the Lower Mainland and the arguments discussed above, it
appears that it would be worthwhile to investigate whether or not the improvements
corresponding to increased filter size are primarily a function of the incorporation
of aggregated elevation information only from the north of the point of interest.
This follows a relatively complex line of thought, but could hold some value
in understanding precipitation modeling in the Lower Mainland.