Learning
Bits

I was invited by Céline Portenier and Stefan Wunderle to present our work on photographic visualizations of weather forecasts in the Colloquium in Climatology, Climate Impact and Remote Sensing at the University of Berne. This was also an opportunity to expand the presentation and include more background information about GANs and our latest evaluation results. Many thanks for the great discussion and feedback!

The 2022 ECMWF Machine Learning Workshop provided a virtual forum for the discussion of machine learning applications for NWP and climate services. I presented our work on using GANs to create photographic visualizations of weather forecasts.

I am a member of the Digital Society working group on automated decision making systems (ADMS). We have published the first version of our position paper on how ADMS should be regulated to benefit humanity.

The scope of our proposal includes ADMS that make, or at least support, decisions in a fully automated way using technical systems. Our proposed legal framework is technology-neutral, instead it follows an approach based on harm and risk. Sanctions are only imposed retrospectively in the event of harm, whereas high-risk applications are subject to certain a priori restrictions. Anyone using an ADMS must assess and categorize its risk to the health, safety and fundamental rights of individuals and society.

The framework provides three categories for this purpose: "low risk", "high risk" and "unacceptable risk". The categories are based on the risk posed by the system to individuals, as well as to society as a whole. "Low risk" systems pose a low risk to individuals and none to society, while "unacceptable risk" systems pose an unacceptably high risk to individuals or society, and therefore must be banned. In between are the "high risk" systems. For these ADMS, we demand a far-reaching duty of transparency and due diligence, which should enable the public to assess their risks and benefits, and thus make informed decisions on their use.

KNMI held a one day symposium on various aspects of data science for meteorology and climatology. I was invited to present our approach to data-driven quality control of meteorological observations.

The MET Alliance is a group of "national aeronautical meteorological service providers from eight European States that cooperate on further improvement and rationalisation of the meteorological services for aviation". I was invited to give an overview of our past and current efforts on camera-based visibility estimation to the expert team on AUTO OBS.