In January 2021, Planet set out to lead the RapidAI4EO consortium to drive cutting-edge continuous land surveillance applications across Europe. This effort is supported by a competitive grant under the Horizon 2020 program to develop improved AI processes and provide critical training data for more frequent land use land cover updates. Earned. As a result of this program, today we are proud to release one of the largest (temporal and spatial) training datasets of satellite imagery to date, suitable for a wide variety of research applications in the area of machine learning. This dataset is accessible to the entire remote sensing community of Source Cooperative, Radiant Earth’s new cloud-based neutral data publishing utility (license terms apply).
This dataset covers 500,000 patch locations across Europe with a frequency of every 5 days over 2 years, and describes the representation and spatial distribution of the countries. The source of the EO data is Planet partner Vision Impulse. The company creates cloud-free Sentinel-2 image mosaics with a resolution of 10 meters each month, and Planet Fusion Monitoring provides he 3-meter images every five days. Our Fusion Monitoring product offers a combination of multiple sensor data types, all refined into a single uninterrupted data stream. Fusion consists of fusing high-frequency daily satellite data with publicly sourced satellite data to provide gapless insights that are ideal for time series analysis.
“Europe already has a world-class downstream EO services industry, thanks to programs like Copernicus and Horizon. We believe that we will be able to support the EU’s progress towards furthering the growth of the European EO ecosystem and promoting further growth in the European EO ecosystem,” said Massimiliano Vitale, Senior Vice President Operations EMEA at Planet.
Rich time series data are important for training models that identify changes in landscape types such as crops, forests, and urban areas. Some land cover changes, such as seasonal crop patterns, can only be identified by understanding how they change over time. While European land cover datasets have existed for some time, this high-frequency time series across all locations is a key innovation, providing more insight into the changing terrain of the European region than ever before. can be classified and evaluated. Although this dataset is designed for land use and land cover change analysis, its insights can be generalized to many research initiatives that benefit from dense time series, such as agricultural monitoring.
“Thanks to the RapidAI4EO consortium led by Planet, we are proud to host the largest open Earth observation training dataset ever,” said Jed Sandwall, Executive Director of Radiant Earth. “The ambitious scale of this project has helped accelerate the development of Source Cooperative, a new data publishing utility. Planet is setting a new standard for open Earth observation training datasets, and this dataset We hope that this will enable reproducible scientific research for years to come.”
Training data already enables the creation of AI-powered change detection models, which can help derive heat maps of changes and prioritize areas for map updates. With this high-frequency time series, we believe this data can open the door to a new family of high-fidelity machine learning models that can disentangle phenology from structural change and learn land cover dynamism. I’m here. The release of this new training dataset is an exciting step forward for understanding land use in Europe, especially for research purposes, and we look forward to the many benefits it will bring to the region.
