Rapid and continuous land monitoring for sustainable development
The advent of high-resolution, nearly-daily earth observation data is offering unprecedented possibilities for rapid and continuous land monitoring. This data requires effective processing, however, there has been little exploration of Deep Learning (DL) techniques to leverage the spatiotemporal dimension at scale. Training data remains rare relative to the spatiotemporal sampling which is necessary to adequately monitor natural and man-made phenomena.
The RAPIDAI4EO project will establish the foundations for the next generation of continuous land monitoring applications by:
- Creating the most complete and dense spatiotemporal training set, combining Sentinel-2 with high cadence, very high resolution, harmonized multispectral Planet imagery at 500,000 patch locations over Europe, and open sourcing these datasets for the benefit of the entire remote sensing community.
- Developing and benchmarking alternative ways of detecting and classifying change from very high cadence observations by training state-of-the-earth multiscale supervised and unsupervised DL classifiers on these unique data sources.
- Delivering high cadence high resolution change detection heatmaps for the entire European continent.
- Demonstrating a highly effective end-to-end process to update operational land cover products (e.g. CORINE land cover), with emphasis on improved understanding of land use, speeding up update cycles and reducing maintenance costs.
Measuring and understanding the human footprint on our planet
RAPIDAI4EO potentially constitutes a game changer in the ability to derive time-critical and location-specific insights into dynamic land surface processes. RAPIDAI4EO’s ambition is therefore to enable new and better ways of measuring and understanding the human footprint on our planet, which is a key challenge of the UN Sustainable Development Goals.
This project brings together industry leaders with a strong, demonstrated record of disruptive innovations and young innovators: Planet, Vision Impulse, VITO, IIASA and ONDA DIAS/Serco Italia.
Within this project VITO will:
– Develop and benchmark state of the art unsupervised and/or weakly supervised model(s) for generic land use land cover (LULC) change detection based on Sentinel-2 and PlanetScope imagery
– update operational LULC products for selected AOI(s) that have been automatically highlighted by the change detection models
– Demonstrate how the latent representations used by the unsupervised DL models can accelerate the production of the LULC products by eliminating the need to work with a full time series in model development.