Swedish Space Data Lab 2.0

Purpose and goal:

The goal was to continue from where the National Space Data Lab ended, focusing on AI models, annotated data, development of Edge Learning with space data.

To achieve this, we have created an annotated dataset that aids the development of AI models. Thanks to it, we were able to train a library of AI models designed to predict cloud optical thickness, with a focus on thin clouds, which are known to be the most difficult to detect.

The Edge Lab was instrumental to the development of a simulation tool to determine communication latency and training strategies for AI models on board of satellites. Expected results and effects:

The goals of the project have been met and the project has enabled new functionalities to retrieve satellite data and new tools and datasets to facilitate the knowledge extraction from datacubes.

The simulation code, the library of AI models to detect thin clouds and the annotated synthetic datasets represent a concrete step forward towards lowering the bar to the use of satellite data. They have ignited the interest of many organizations that we will gather in the continuation of this project to further develop such tools and create new ones equally craved for.

Approach and implementation:

Building a user-friendly graphical interface for the platform goes slowly but steadily forward and we were able to use it and demonstrate its importance especially in the last phase of the project. It took a relatively long time to determine which use case could have the broadest impact before converging on the thin cloud one, but the direction we took is undoubtedly correct.

Through the participation in and the organization of various events, seminars and hackathons, and the publication of several papers, we disseminated our results and planted the seed for future collaborations.