Robustness in GIS algorithm implementation with application to line simplification

The sensitivity of computed results to implementations of algorithms in GIS is considered in this paper on the example of a precisely defined recursive variant of the Ramer-Douglas-Peucker line simplification algorithm (called the R-D-P algorithm). We establish a robust version of the R-D-P algorithm where the determination of the simplification is rounding error free if the data are already machine numbers. Under these assumptions, the results are reproducible which is not the case with other versions of the algorithm.