“Most of the world is covered by water. A fisherman’s job is simple: Pick out the best parts” – Charles W. Waterman

… in the context of our current world of information and global disasters: Analysts agree that the needed information exists. The challenge is to find them with limited effort and time, and time is what counts especially when it comes to environmental disasters.

On Thursday, September 8, 2022, we had the opportunity to present and discuss our ideas in front of numerous representatives from the German water industry at the Early Birds event of the Competence Center for Digital Water Management (Kompetenzzentrum für Digitale Wasserwirtschaft). The competence center is a think-tank with the goal to bring meaningful innovations and concrete solutions forward.

The competence center asked us if we could use our early warning system to spot the first signs of fish kills simply by monitoring multilingual social media or newspapers. The reason for this is the recent events on the Oder River in August 2022, which were reported by Polish media a few days earlier than German media.

We accepted the challenge and started collecting data for a new AI model that identifies global fish kill events by analyzing news articles and social media. The idea is based on our Event Classification system in combination with the Prometheus AI system. 

Within a few days, we were able to create a data set for fish kills. This set includes about 1,000 multilingual samples for identified fish kills or reports that contain certain keywords but are not related to that category at all. A question that often comes up when we explain how we add new categories to our event classification:

Question: Is this data set size big enough for the training of a new AI model?

Answer: Definitely yes! As we constantly collect and annotate new data, we can also constantly create new improved AI models. So, it is a starting point.

The following screenshot shows a simplified result of the first AI model in production.

In sum, it took us only 3 days to extend our already existing early warning system with the new event-type fish kill. By now, we are constantly monitoring global events also with a focus on environmental disasters, such as wildfires, chemical accidents, and gas leaks, …