2/3/2025 | 3:30 PM - 5:30 PM | Regency B

High-altitude earth observation with diffusion models for satellite LiDAR reconstruction

Author(s)

Andres Ramirez-Jaime | University of Delaware
Nestor Porras-Diaz | University of Delaware
Gonzalo Arce | University of Delaware
Mark Stephen | NASA Goddard Space Flight Center

Abstract

Satellite LiDARs play a critical role in high-altitude Earth observation, providing valuable insights into terrestrial and atmospheric processes. However, the limited resolution and sparsity of satellite LiDAR data pose significant challenges for effective analysis. In this work, we address these challenges by solving key inverse problems—including compressive sampling, super-resolution, and denoising—through the inversion of a probabilistic forward imaging model. By employing diffusion generative models as a prior, we constrain the solution space to the distribution of plausible high-resolution LiDAR data, enabling robust and accurate reconstructions. Central to our approach is the use of a novel representation, the HyperHeight Data Cube, which facilitates efficient modeling and reconstruction of satellite LiDAR data. This work not only enhances the quality of existing data but also paves the way for future satellite missions, advancing the state of remote sensing for Earth sciences.