The estimation of digitizing error and its propagation results in GIS and application to habitat mapping

In order to identify and quantify sources of digitizing error, and to understand how error is carried through a map overlay, three experiments were performed. First, four operators repeatedly digitized eight randomly distributed points eight times. Data were analyzed using parametric statistics and by an error models so that average point digitizing error, the operator error (random error and bias), and the machine error (random error) were estimated; the operators\u27 digitizing characteristics were statistically analyzed. In the second experiment I designed and created a standard coverage with the ARC/INFO GENERATE command. The coverage consists of a set of special geometric entities: a series of differently sized circles, differently shaped triangles and rectangles drawn by PC ARCPLOT. These figures were designed to test the effect of the original map polygons\u27 characteristics on both machine error and operator error, such as area and perimeter, figure shape and geometric entities combination, line curvature, number of vertices selected for representing a line, position of a geometric entity on digitizing board. Several operators repeatedly digitized the coverage six or more times. The operator error (area error and perimeter error) were obtained by subtracting the standard coverage from a digitized coverage. The machine error (area error and perimeter error) were obtained by subtracting the theoretical true coverage from the standard coverage. The results were analyzed statistically. The causation of errors and the operators\u27 digitizing characteristics were further discussed. The third experiment addressed how digitizing errors are propagated through map overlay. In this experiment the digitized coverages created in the second experiment by each operator were overlaid with the ARC/INFO UNION command. Area error, perimeter error, and the numbers of the spurious polygons were collected. The means, sums, maximum, minimum, and standard deviation of area error and perimeter error were obtained. The relationship between area error, perimeter error, and number of spurious polygons of the overlaid coverage were analyzed. This study: (1) focuses on position, as opposed to attribute, error; (2) examines errors in vector-based, not raster-based, GIS; and (3) examines errors caused during the digitizing process, and their propagation through map overlay. The digitizing method is point mode, not stream mode. Results were applied to error management and error reduction to: (1) create a theoretical model which can be used to check quality of the vector source coverages, and to lead users to correctly utilize the GIS data, to prevent them from making unnecessary mistakes; (2) identify some rules to properly use the ARC/INFO ELIMINATE command, and to set MMU (minimum mapping unit) for a particular project