Geoscience Reference
In-Depth Information
netic radiation at varying wavelengths, sensors that have poorly calibrated
multiple detectors within a band, sensor calibration error, signal digitization
error, and scene-specific error such as off-nadir viewing, irradiance variation,
and terrain topography (Richards 1986). Correction of band-to-band distor-
tion is performed using image histograms (shifted to the origin to remove
atmospheric scattering effects), whereas line striping effects are reduced by cal-
ibration of detectors or by matching detector statistics during computer pro-
cessing (Teillet 1986).
A final type of error may be caused by a time lag between ground truthing
and image collection. In this case, pixels may be noted as incorrect in the error
matrix (described later in this chapter) when they may be actually correct at
the time of image acquisition.
Vector images have been traditionally recorded and stored as maps. Maps
are subject to many errors. Some errors are introduced during the creation of
the map, such as the original line smoothing, which may not follow the true
isolines on the ground (Chrisman 1987). Other errors may be associated with
the physical medium used to store the map (e.g., paper stretch and distortion).
Maps may be represented in computer
GIS
s by a variation of the vector data
structure (Peuquet 1984) or converted to a raster data structure. In its simplest
form, the vector data structure has map lines approximated to a set of points
(nodes), which are linked by lines (or arcs). Vector data may be obtained by
digitization.
Digitization introduces a number of errors. Varying line thickness on the
original map requires automatically scanned vector lines to be thinned. Dur-
ing manual digitization the center of the map line must be followed carefully
if the map lines vary in thickness (Peuquet and Boyle 1984). This requires very
careful hand digitizing or high-accuracy automatic scanners. The number of
vertices (points) used to approximate a curve is also critical (Aldred 1972). Too
few vertices will result in the line appearing stepped, and too many vertices cre-
ate large data volumes. Thus, even with extreme care, error is introduced dur-
ing digitization.
As for raster images, the main method of correcting geometric error in vec-
tor images is by using ground control points from a cartographically correct
map to transform the vector image to a known projection.
Data layer error quantification
Methods for quantifying error in a raster data layer are based on the error
matrix (also called a contingency table or confusion matrix) concept, first
expounded for remotely sensed data in the 1970s (Hoffer 1975).
