satellite imagery usually contains

satellite imagery usually contains metadata giving the scene frame – the sensor
direction in relation to the earth at scan time – air photographs need to be
registered to known ground control points.
These ground control points were ‘known’ from terrestrial triangulation,
but could be in error. The introduction of Global Positioning System (GPS)
satellites has made it possible to correct the positions of existing networks of
ground control points. The availability of GPS receivers has also made it possible
for data capture in the field to include accurate positional information
in a known coordinate reference system. This is conditioned by the requirement
of direct line-of-sight to a sufficient number of satellites, not easy in
mountain valleys or in city streets bounded by high buildings. Despite this
limitation, around the world the introduction of earth observation satellites
and revised ground control points have together caused breaks of series in
published maps, to take advantage of the greater accuracy now available. This
means that many older maps cannot be matched to freshly acquired position
data without adjustment.
All of these sources of spatial data involve points, usually two real numbers
representing position in a known coordinate reference system. It is possible to
go beyond this simple basis by combining pairs of points to form line segments,
combining line segments to form polylines, networks or polygons, or regular
grid centres. Grids can be defined within a regular polygon, usually a rectangle,
with given resolution – the size of the grid cells. All these definitions imply
choices of what are known in geographical information systems (GIS) as data
models, and these choices have most often been made for pragmatic reasons.
All the choices also involve trade-offs between accuracy, feasibility, and cost.
Artificial objects are easiest to represent, like roads, bridges, buildings, or
similar structures. They are crisply defined, and are not subject to natural
change – unlike placing political borders along the centre lines or deepest
channels of meandering rivers. Shorelines are most often natural and cannot
be measured accurately without specifying measurement scale. Boundaries
between areas of differing natural land cover are frequently indeterminate,
with gradations from one land cover category to another. Say that we want
to examine the spatial distribution of a species by land cover category; our
data model of how to define the boundary between categories will affect the
outcome, possibly strongly. Something of the same affects remote sensing,
because the reported values of the observed pixels will hide sub-pixel variation.
It is unusual for spatial data to be defined in three dimensions, because of
the close links between cartography and data models for spatial data. When
there are multiple observations on the same attribute at varying heights or
depths, they are most often treated as separate layers. GIS-based data models
do not fit time series data well either, even though some environmental
monitoring data series are observed in three dimensions and time. Some GIS
software can handle voxels, the 3D equivalent of pixels – 2D raster cells –
but the third dimension in spatial data is not handled satisfactorily, as is
the case in computer-assisted design or medical imaging. On the other hand,