EarthObsNet: A Comprehensive Benchmark Dataset for Data-Driven Earth Observation Image Synthesis

This is a Preprint and has not been peer reviewed. This is version 1 of this Preprint.

Add a Comment

You must log in to post a comment.


Comments

There are no comments or no comments have been made public for this article.

Downloads

Download Preprint

Authors

Zhouyayan Li, Muhammed Yusuf Sermet, Ibrahim Demir

Abstract

Remote Sensing imagery serves as an important data source for Earth surface monitoring and surface processes studies. It is highly likely that RS imagery will become more and more indispensable in the future due to its high scalability and compatibility with data-driven models and ever-evolving software and hardware that become increasingly good at processing large datasets. Although its promising future, the usage of Earth surface observation imagery, such as Landsat, Sentinel-2, and Sentinel-1 images, has been largely limited to retrospective studies, where those images serve mainly as documentations of past events. Recently, there are attempts to expand the current usage of RS Earth surface observation images to forward-looking applications to support decision-making and fast response against future natural hazards. Unlike many well-defined and well-studied topics such as change detection and semantic segmentation for which many benchmark datasets are openly available, so far, there are limited public datasets for Earth surface observation image synthesis tasks for fast prototyping and comparison. To close this gap, we introduced a comprehensive dataset containing previous Earth surface observations, precipitation, soil moisture, land cover, Height Above Nearest Drainage (HAND), DEM, and slope collected during the catastrophic 2019 Central US Flooding events that lasted more than two seasons in Mississippi and Missouri River tributaries. We also incorporated reference labels to allow further investigation of the usefulness of the synthesized images in downstream applications, such as flood inundation mapping. We hope to provide an essential dataset for Earth observation image synthesis studies, with the goal of attracting more attention and inspiring more efforts to broaden the usage of Earth surface observation images into forward-looking applications.

DOI

https://doi.org/10.31223/X5GD7J

Subjects

Engineering

Keywords

benchmark dataset, Earth's surface image synthesis, multi-purpose analysis, Deep learning, SAR, meteorological and geomorphic input

Dates

Published: 2024-03-03 22:40

Last Updated: 2024-03-04 03:39

License

CC BY Attribution 4.0 International

Additional Metadata

Conflict of interest statement:
None

Data Availability (Reason not available):
Data and source code is openly available. Please check the Dataset and Code Availability section in the manuscript.