Finding mountain summits in high-resolution LiDAR data

Two semi-automated processes are described for determining all of the mountain summits above a minimum topographic prominence threshold in a large set of high-resolution LiDAR elevation measurements. The first uses a parallel algorithm to find a set of candidates, and refines the candidates by comparing against results from running the same algorithm on a much lower-resolution digital elevation model. The second uses computerassisted manual review of LiDAR point clouds to improve an existing set of candidates derived from topographic maps. Both methods identify significant deficiencies in existing LiDAR datasets, most importantly, misclassification of return type in point clouds, leading to erroneous mountain elevation and prominence values calculated from the digital terrain models in these datasets. We find a root-mean-square vertical error on 5242 non-vegetated summits in the U.S. of 1.05m due to misclassification, compared with a published specification of less than 0.1m of absolute error on the same dataset.