each element of the array the relative cover of
ten cover types was measured. These were
combined to form three main types: heather
species, grass species and bare soil. The cover
percentages of these types along the elements
of the array form a pattern. In the image
exists a similar pattern in pixel values, induced
by the field pattern (figure 1).
ground element
Figure 1: Grass cover percentage along an array
of ground elements, and pixel values on corres
ponding positions in the image.
Correspondence between field pattern and
image pattern is supposed to reach a maxim
um when those pixels are selected that co
incide in space with the ground elements. The
residual variance from the regression of
ground cover percentages on pixel values can
be used as a measure of this correspondence.
Optimization of the image localization with
respect to the field data was done by selecting
the array of pixels that yielded the least pos
sible residual variance. In figures 2 and 3, the
hypothetical effect of shifting and rotating the
array of pixels on the residual variance is
depicted. When pixel and ground array do not
overlap, the residual variance is supposed to
fluctuate at random on a high level.
Figure 4 shows actual contours of this vari
ance when an array of pixels is shifted in the
image. In this way, positioning the field data
in the image can be done with subpixcl ac
curacy.
In most cases, ground elements will not
align with pixels. Pixel arrays were therefore
resampled from the image, allowing them to
take every desirable position and orientation.
Resampling was done by sampling a number
of points (625 in this case) within the projec
tion of the new pixel in the image, and by
taking the average of the sampled values as
the new pixel value.
A line intercept method (Cialcs 1979) was
used in the field to measure the cover of
heather, grass and bare soil in a series of
ground elements. Along three parallel lines,
spaced 12.5 m apart, the lengths of the cover-
types intersecting with (he line were recorded.
These lengths were summed over a distance of
25 m, the size of a ground clement. The sum-
total of lengths for a cover type, relative to
x-position
Figure 2: Expected behaviour of residual varian
ce as a function of the position of pixel and
ground element arrays relative to each other.
When they fully overlap, residual variance rea
ches a minimum.
Angle between pixel-array and ground elenent array
Figiire 3: Expected behaviour of residual varian
ce as a function of the angle between the array
of pixels and the array of ground elements.
Residual variance reaches a minimum when the
arrays arc parallel.