FDL 2019

 

2019 Challenge areas

The FDL yearly cycle starts with challenge definition. Early in the year, we bring together some of the brightest and best minds we can find, from space science, AI and technology, and on/off-Earth applications to explore our challenge areas. During the course of our day-long Big Think events in Europe and the US, we aim to identify some broad challenges, which the FDL research teams could tackle in the summer.

Through a process of iteration with a PI (principal investigator) leading each challenge, we refine and narrow those challenge areas until we have identified one, or several, tightly articulated questions to resolve.

FDL challenges must represent a clear and present scientific problem, for which there is available data, that could be significantly advanced by AI tools and techniques. It is these challenges that the research teams further narrow in the opening weeks of the FDL research sprint to refine their own particular concept approach. The broad challenge areas we start the year with move from provisional to confirmed as we understand how, and when, they meet these criteria. Our confirmed challenge areas for 2019 are:

Artwork39_9136da26-b38c-4de0-bb09-7de8768ba3c7.png

Atmospheric Phenomena and Climate Variability

How might AI improve climate models and enable better decision making for resilience planning?

CNN-based approaches are essential for exploiting the long history of Earth observations to evaluate natural variability and secular trends in cloud characteristics. Developing a semi-supervised analysis framework that builds a neural network model could exploit this data richness.

This procedure would start by training a CNN classifier on labeled and deterministically defined cloud classes, learning the partitioning of classes, and analysing variability within each separate class by using a variational auto encoder to find candidates for new cloud classes. If new class candidates appear, representative images would be synthesised for each new class and input as sources to the supervised CNN approach. This procedure would be iterated to reach a new partition configuration that accommodates the new classes. The framework could also be extended to integrate information from multiple satellites over decades.


unnamed-2.png

Disaster Prevention, Progress and Response

How might we utilise AI and Earth observation data to support improved decision making to protect the planet?

Working closely with our partners at UNICEF, can we investigate how AI can improve our capabilities to forecast and respond to floods using orbital imagery, coupled with ground observations and social data?


unnamed.png

GROUND STATION PASS OPTIMIZATION FOR CONSTELLATIONS
How might AI be used to further optimise spacecraft operations?

During routine operations of existing ESA missions, a large amount of time is invested in the scheduling and planning of ground station passes for satellites. This is mainly with some automation and manual work and involves trial and error. Can we create an optimised ground station schedule for spacecraft using AI?


Ideally, such a solution would reduce the amount of tracking hours while maximising the science return.




You can read about the US based NASA FDL 2019 challenges by following this link.