Effect of manual digitizing error on the accuracy and precision of polygon area and line length

Manual digitizing has been recognized by investigators as a significant source of map error in GIS, but the error characteristics have not been well defined. This thesis presents a methodology for simulating manual digitizing error. Stream mode digitizing error was modeled using autoregressive moving average (ARMA) procedures, and point mode digitizing was stochastically simulated using an uniform random model. These models were developed based on quantification of digitizing error collected from several operators. The resulting models were used to evaluate the effect digitizing error had upon polygon size and total line length at varying map accuracy standards.

Digitizing error produced no bias in polygon area. The standard deviation of polygon area doubled as the accuracy standard bandwidth doubled, but the standard deviation was always less than 1.6 percent of total area for stream mode digitizing. Smaller polygons (less than 10 square map inches) had more bias and more variance relative to their size than larger polygons. A doubling of the accuracy standard bandwidth caused a quadrupling of line length bias and a doubling to tripling of the line length standard deviation. For stream mode digitizing, reasonable digitizing standards produced line length biases of less than 2 percent of total length and standard deviations of less than 1 percent of total length. Bias and standard deviation both increased with increasing line length (or number of points), but the bias and standard deviation as a percent of total line length remained constant as feature size changed.