Land Type Classification of ICESat-2 Global Geolocated Photon Data

Master thesis (2020)

Authors

J.M. Heywood Aerospace Engineering

Contributors

W. van der Wal Physical and Space Geodesy - CEG (supervisor 1)
R.C. Lindenbergh Optical and Laser Remote Sensing - CEG (supervisor 1)
Faculty
Aerospace Engineering
Copyright: © 2020 Joseph Heywood
Random Forest LIDAR ICESat-2 Laser Earth observation CORINE Land Type Classification
More Info
expand_more

Published Date

18-09-2020

Language

English

Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Faculty

Aerospace Engineering

Abstract

NASA's ICE, Cloud and Land Elevation Satellite-2 (ICESat-2) has been measuring the topography of the Earth's surface since its launch in September 2018. Equipped with a single instrument, namely, the Advanced Topographic Laser Altimeter System (ATLAS), ICESat-2 is able to acquire the along track vertical profiles of its laser footprints. While satellite based land classification has traditionally been performed with the use of multi-spectral data, which doesn't consider the vertical structure of the surface in question, the three-dimensional Light Detection and Ranging (LiDAR) product provided by ATLAS allows for the observation of the vertical structure of the illuminated surface. This provides information for the discrimination of surfaces that are only distinctly different from one another in this dimension, such as different types of vegetative species. Moreover, a greater understanding of how the signal interacts with different land types will benefit current and future users of the data. This study presents a first look at the potential of the base scientific data set provided by ATLAS, "ATL03", as a means of land type classification. Features extracted from ATL03 vertical profiles are used to classify multiple land types in The Netherlands, namely, "Artificial Surfaces", "Agricultural Areas", "Forest and Semi-Natural Areas", "Wetlands" and "Water Bodies". 100m grid cells were classified and validated with the CORINE land cover database. The overall classification accuracy was 71.2%, however, after a visual inspection of the misclassification errors it was found that that the actual accuracy was a minimum of 5.5% higher, that is, 76.7%. 51 features were created to discriminate between land classes and their importance per class was analysed. In general, simple statistical parameters, such as the standard deviation and percentile ranges worked well in distinguishing between classes. For the classes with a greater vertical range, such as "Artificial Surfaces", features that described the height and prominence of its scattering surfaces were most important.