|
|
|
|
|
|---|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Orthophotographs (1:25000) were provided
by the Department of Land Administration (DOLA).
Georeferencing of orthophotographs
was achieved by entering UTM coordinates for
the edges of the images. Initial trial scans of the orthophotographs proved
unsatisfactory due to errors incurred during the scanning process. The orthophoto
images were rescanned until an error of < 4 mm in 1200 mm (0.3%) was achieved
in the squareness of a test image provided by DOLA to the scanning agency (CAD/CAM
Centre Technology Park). Poor contrast across the orthophotographs made selection
of common GCP’s in some areas difficult and required
further adjustment of contrast to aid in identifying the selected points.
Georeferencing of air photos was achieved
by matching ground control points on scanned
orthophotographs with their equivalent on the scanned air photos. The air photograph
image was then resampled using nearest neighbour methods with an output pixel
size of 3 m. RMS errors for all georeferenced
resampling were restricted to < 0.6 of pixel
size, with the intention of maintaining a spatial
error of < 1.02 m. The output pixel size
of 3 m was determined to be the optimum given the overall image size and system
space limitations. A pixel size of 1 m would
have resulted in a composite image of approx 3 Gb
exceeding the available system. The resultant air photograph mosaic
image size at 3 m pixel resolution was approx
200 Mb. Each air photograph image of 20 Mb
was subset to 2 x approx 4 Mb file sizes to
provide sufficient overlap in the final mosaic
while reducing the overall size of files to handle.
The final mosaic was produced using two options,
firstly a feather overlap option which reduces contrast variation between sub-images
of the mosaic while allowing some transparency
between sub-images (Figure A1). This has the effect
of blurring the overlapping components of each image and reduces the spatial
accuracy of the overall image in areas of overlap. The image was given the
file name 1994fs.img (Table 2) and had a file size of
approx 200 Mb. Department of Transport requested
a smaller file size for the resultant images so a smaller image was resampled
using a cubic convolution method
to a 15 m pixel size, this being chosen to
provide a final image file size of approx 8 Mb..
This image was given the file name 1994fsrcc.img (Table 2).
Secondly, a maximum overlap reduction of 99.9% was applied in both X and Y axes
which has the effect of butt-joining the portions of images at the centres of
their overlap, thus visually eliminating the areas of overlap and maintaining
the maximum spatial accuracy of the
image. This image was given the file name 1994mors.img (Table
2). Again a smaller file size was requested and the cubic convolution resampled
image with a 15 m pixel resolution was named
1994morsrcc.img (Table 2). The naming conventions of
the image files are presented in Table 2 which gives a brief description of
the file status and the process.
|
|
|
|
|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
The final mosaic images were compiled from approximately 80 sub-images shown in Appendix 2. All .img files including the large and resampled mosaics and postscript and TIFF files of the mosaics, were TAR copied to DAT tape in a compressed write mode. A flow diagram of the entire process is summarised in Appendix 3. A table showing file handling procedures is contained in Appendix 4.
Air photograph interpretation was carried out for each of the dates on the air photographs using the keys contained in Appendix 5. Each photograph was interpreted independently by two interpreters using stereoscopes to allow magnification and observation of imagery in three dimensions. This enabled use of terrain shape to assist in the identification of samphire. Both sets of interpretations of samphire cover were then transferred to a clear transparency overlay on the orthophotographs. After the samphire overlays were completed for each date they were then digitised over the digital mosaic image using on-screen digitising techniques in Imagine software. In all cases the union of areas determined by the two interpreters were digitised, thus only errors of commission occur in the final agreement of the two interpretations.
The digital overlays were then embedded into a image raster layer and the image classified using an unsupervised classification routine into two classes, samphire and the rest. These classes were used to determine total areas of samphire. The common pixel size for classification purposes was 4.5 m. This was dictated by software limitations. The overlay images were then used to subset a common total scene and local scenes of interest for further determination of site specific details.
|
|
