ICESat waveform-based land-cover classification using a curve matching approach

Abstract

Light detection and ranging waveforms record the entire one-dimensional backscattered signal as a function of time within a footprint, which can potentially reflect the vertical structure information of the above-ground objects. This study aimed to explore the potential of the Geoscience Laser Altimeter System sensor on board the Ice, Cloud, and Land Elevation Satellite to perform land-cover classification by using only the profile curve of the full waveform. For this purpose, a curve matching method based on Kolmogorov–Smirnov (KS) distance was developed to measure the curve similarity between an unknown waveform and a reference waveform. A set of reference waveforms were first extracted from the training data set based on a principal component analysis (PCA). The unknown waveform was then compared with individual reference waveforms derived using KS distance and assigned to the class with the closest similarity. The results demonstrated that the KS distance-based land-cover classification using the waveform curve was able to achieve an overall accuracy of 87.2% and a kappa coefficient of 0.80. It outperformed the widely adopted rule-based method using Gaussian decomposition parameters by 3.5%. The research also indicated that the PCA- selected reference waveforms achieved substantially better results than randomly selected reference waveforms.

Acknowledgements

The authors would like to thank the National Snow and Ice Data Center for providing ICESat data.