Accuracy is defined as the closeness of observations to the truth. Within geographic information science, this definition does not necessarily cater for all situations associated with geographic information—hence the use of alternative terms, such as uncertainty, precision, and vagueness. The reason for the deficiency is that while the definition works well for features in the built environment, the natural environment presents considerable difficulties when we try to describe and model it. In addition, there are several different perspectives to accuracy that are important in GIS. These are positional, temporal, and attribute accuracy and the issues of logical consistency and completeness.
Positional accuracy, the accuracy of a feature’s database coordinates, can sometimes easily be confirmed. For example, how close a street pole’s database coordinates are to its real-world coordinates can be determined by using specialized field-surveying equipment to gain an answer correct to a few centimeters. This works well for features in the built environment that are represented by points in a database.
However, in other cases, such as determining the accuracy of the location of a lake boundary as it is represented in a database, it may be impossible to assess positional accuracy. In this instance, the boundary will rise and fall with the water level, so we do not necessarily know what “truth” we should be trying to test against. In addition, if we use field survey to determine the coordinates of points on the actual lake boundary, then we face the problem of trying to compare these points with the boundary in the database, which is recorded by a series of straight-line segments. Measuring the positional accuracy of other natural phenomena represented in digital form is similarly difficult. For instance, the location of a vegetation boundary may vary between experts, depending on the criteria (such as precision, minimum mapping unit, and classification) used to delineate different vegetation types. This may be coupled with the problem that a field survey cannot always be conducted to record the accuracy of a (conceptual) boundary that may not actually exist on the ground—although plantation and clear-cut forest boundaries would be an exception.
The lake edge example can also be used to introduce the concept of temporal accuracy, the accuracy of the temporal information held in a database. For example, if a lake polygon had the time stamp of the date of the aerial photography from which it was digitized, then that time stamp should not be in error. However, temporal accuracy should not be confused with the “database time,” the date the polygon was recorded in a database (which might be a considerable time after the aerial photography was flown), or with a currency” or “up-to-dateness,” a measure of how well the database reflects the real-world situation at the present or a particular time in the past.
Attribute accuracy, the accuracy of attributes listed for a database feature, can also be easily checked in some cases but not others. For example, the street address for a land parcel in a database can be quickly checked for correctness, but determining the accuracy of the land use description for the same parcel can be difficult. The parcel may contain a large building with underground car parking, retail shops, and residential apartments, yet the database records the land use only as “retail.” So the database is only partly accurate. In the natural environment, the accuracy of soil classifications might be checked by testing at-point sample sites, but soils are rarely pure in their classification, and the database description will be only partly correct.
Logical consistency is a form of accuracy used to describe the correctness of relationships between database features and those found in the real world. An example of this is an emergency dispatch application where it is critical that all roads connecting in the real world are actually connected in the database; otherwise, incorrect routing of emergency vehicles
Accuracy may also be documented in terms of completeness—that is, are all features in the real world shown in a database? Often, features are deliberately deleted from databases for the sake of simplicity, such as including only major walking tracks or vegetation polygons over a certain size. Omission is not always an indication of nonexistence. Thus, in some cases, accuracy is easy to determine according to our definition, but often there is conflict as a result of the way we model the world with geographic information and of the nature of the world itself.