An Essay on Uncertainty in Spatial Data Integration

Spatial data come in a variety of forms from oral or text descriptions and values, to any number of digital representations of the real world.  There are a plethora of techniques used to describe the world in space through raster and vector data alone.  The quality of such data however can be described by addressing four key factors.  Accuracy measures how close data can matches true values and descriptions.  Not all values can be exactly measured.  Due to human error or the detection/precision limits of equipment accuracy can easily be skewed.  Accuracy however, is largely associated with scale – decreasing scale increases accuracy.  Precision – reproducible performance qualities – measures how exactly data measured and stored.  In an assessment using high precision, several errors can be repeated with similar patterns of the same error, however low precision will yield little or no representation to the original error.  Error is a typical deviation or variation from reality.  Finally, uncertainty is the overall doubt or lack of confidence in data.  Though error and uncertainty are similar, the discrepancies are known and can possibly be avoided, where uncertainty deals with the absence of knowing the truth of a situation.  Moreover, spatial data quality is a measure of how well GIS data represent reality.

Due to the fact that a map is a model of reality and that model can never be a completely valid representation of reality, it is important to expect a certain level of uncertainty when working on a project as a quality control or assurance method.  Further, the differences in data types that are attempting to represent the same real-world features inherently exhibit a degree of variation.  For instance a raster image and a vector representation of the same coastline will align separately; by any number of spatial units.  Adjusting the scale at which the features are obtained may alleviate this misalignment.  Combining vector data digitized separately will rarely align perfectly.  Thus, combining two such coverages will result in sliver polygons, disconnected vertices, or polyline dangles.  It will be necessary to use overlay tools – such as a spatial join, union, etc. – to aggregate datasets while avoiding unwanted errors.

Storing information in metadata files is another powerful method that can be used to track errors that are present in GIS application.  Metadata is a means of recording information about a given dataset.  It can describe, among many other matters and parameters, which methods have been used to capture data in a coverage.  Providing such information can easily and efficiently inform others who use the dataset what he or she should anticipate as far as what deviation or error occur.  A lecture in class discussed how improvement in awareness of data quality is necessary, as data quality can have large impacts on geographic data analysis if it is not addressed properly; or even overlooked entirely.

Scale is another issue that may assist in the avoidance of uncertainty.  Determining an appropriate level at which any error that may occur in a dataset will be minimized is a sure way to improve the accuracy of a dataset.

Controlled uncertainty is another spatial data quality issue that works in inverse from the previously discussed methods.  Deliberate degradation of data is used as a mechanism for protecting the data at hand, whether a dataset covers records, or locations whose location or confidentiality may be guarded.  The Fisher reading summarizes well that analysis performed on a project without accommodating for uncertainty will have significantly degrade its validity, and therefore its usefulness is questionable.  By planning for a certain degree of error, it is possible to continue to create valid results.