Improving Positional Correspondence of GEDI Full Waveform data for Structural Mapping

NASA's Global Ecosystem Dynamics Investigation satellite mission(GEDI) measures the three-dimensional structure of the Earth using its high-resolution LIDAR. The laser system has three lasers divided into eight tracks, each taking a 25m footprint area. The level 1 data product of the GEDI contains geolocated full waveforms; i.e., the waveform gives a complete vertical profile of the return signal that bounces from the ground, providing a 15 cm vertical resolution. Given GEDI's relatively sparse coverage over the earth's surface, our goal is to develop machine learning models to estimate structural characteristics seamlessly using ubiquitously available multi-temporal satellite imagery as input to the models. Thus establishing positional correspondence between GEDI full-waveform and the spectral satellite images is quintessential. However, the positional accuracy of GEDI's waveform is around 10 m which calls for major improvement, given the dense, abstract, and complicated structure of a forest. We plan on using USGS's 3D Elevation Program(3DEP) LiDAR point cloud data which have a horizontal positional accuracy of a few cm, to create pseudo-waveforms. We will use these pseudo-waveforms to validate and establish the improved positional correspondence of the GEDI's full-waveform. We shall use KL divergence, which quantifies the similarity between two signals by assuming them as probability distribution curves, to compare the two waveforms and validate the full-waveform's position. Finally, we shall compare the full-waveform with other pseudo-waveforms in 10 m diameter of the given position to assess if the position needs modification.