Modeling Species Distribution with GIS
419
whether an image pixel is correctly classified, a point within the polygon is verified against the ground truth information. A specific problem encountered
with vector images is ground truth samples that occur across boundary lines;
in this case the class with the largest area within the sample area may be
selected to represent the vector map image.
A method of assessing map accuracy based on line intersect sampling was
described by Skidmore and Turner (1992). Line intersect sampling is used to
estimate the length of cover class boundaries on a map 10 that coincide with the
true boundaries of the cover classes on the ground. A ratio of coincident
boundary to total boundary is proposed as a measure of map accuracy and this
ratio is called the boundary error. Although this technique was developed for
vector maps, it is equally applicable to raster maps.
Skidmore and Turner (1992) found that the true boundary lengths were
not significantly different from the estimated boundary lengths sampled using
line intersect sampling, with a′ = 0.05. The estimated boundary accuracy (64
percent) was extremely close to the true boundary accuracy (65.1 percent), and
there was no significant difference between the true and estimated boundary
accuracy.
Reliability of the output
Apart from assessing the accuracy of the original data sets used to produce GIS
models, the final products of the modeling effort must be validated.
As previously evidenced, the process of model building takes its move from
a number of data maps of the same region (e.g. elevation, soils, ground cover),
which are digitized and geographically rectified to a common projection in a
GIS. Such maps may be stored as a series of layers in a GIS. Each point within a
polygon, or each cell in a raster layer (where each raster cell is assigned one
value), takes the values of the layers directly above the point. The model is then
built defining specific questions that may be asked about specific points or
cells, such as the slope or aspect at the point. Eventually the biologist is interested in mapping areas that satisfy the known ecological requirements of a
species (e.g., slope greater than 30°, with erodible soils, occurring on a
southerly aspect). By overlaying the map layers, such simple queries may be
answered.
Nevertheless, the simple process of overlaying layers propagates the errors
of the original data and sometimes amplifies them. If the sources of errors are
identified and their accuracy is known, an estimate of the accuracy of the final
output can be achieved through error propagation analysis.
419
whether an image pixel is correctly classified, a point within the polygon is verified against the ground truth information. A specific problem encountered
with vector images is ground truth samples that occur across boundary lines;
in this case the class with the largest area within the sample area may be
selected to represent the vector map image.
A method of assessing map accuracy based on line intersect sampling was
described by Skidmore and Turner (1992). Line intersect sampling is used to
estimate the length of cover class boundaries on a map 10 that coincide with the
true boundaries of the cover classes on the ground. A ratio of coincident
boundary to total boundary is proposed as a measure of map accuracy and this
ratio is called the boundary error. Although this technique was developed for
vector maps, it is equally applicable to raster maps.
Skidmore and Turner (1992) found that the true boundary lengths were
not significantly different from the estimated boundary lengths sampled using
line intersect sampling, with a′ = 0.05. The estimated boundary accuracy (64
percent) was extremely close to the true boundary accuracy (65.1 percent), and
there was no significant difference between the true and estimated boundary
accuracy.
Reliability of the output
Apart from assessing the accuracy of the original data sets used to produce GIS
models, the final products of the modeling effort must be validated.
As previously evidenced, the process of model building takes its move from
a number of data maps of the same region (e.g. elevation, soils, ground cover),
which are digitized and geographically rectified to a common projection in a
GIS. Such maps may be stored as a series of layers in a GIS. Each point within a
polygon, or each cell in a raster layer (where each raster cell is assigned one
value), takes the values of the layers directly above the point. The model is then
built defining specific questions that may be asked about specific points or
cells, such as the slope or aspect at the point. Eventually the biologist is interested in mapping areas that satisfy the known ecological requirements of a
species (e.g., slope greater than 30°, with erodible soils, occurring on a
southerly aspect). By overlaying the map layers, such simple queries may be
answered.
Nevertheless, the simple process of overlaying layers propagates the errors
of the original data and sometimes amplifies them. If the sources of errors are
identified and their accuracy is known, an estimate of the accuracy of the final
output can be achieved through error propagation analysis.