Characterizing the error distribution of lidar elevation data for North Carolina

Abstract

Spatial data quality is a paramount concern in all geographical information systems (GIS) applications. Existing standards and guidelines for spatial data commonly assume the positional error is normally distributed. While non-normal behaviour of the error in digital elevation data has been observed in previous research, current guidelines for digital elevation data still assume that the errors for observations in open terrain are normally distributed. This research employed an accuracy assessment dataset from a substantial lidar data collection effort, the North Carolina Floodplain Mapping Program. Strong evidence was found that the vertical error of lidar elevation data is not normally distributed and that both major and minor outliers are very common. Of the five land cover types considered, only the distribution for urban areas approximated a normal distribution, even though these observations were generally much less accurate than those for open terrain. No influence of slope on the occurrence of non-normal behaviour in the distributions was found. The RMSE_z (root mean square error) statistic used to characterize the fundamental accuracy of digital elevation data was found to be very sensitive to the occurrence of outliers, questioning its use in current guidelines.