Abstract/Summary

The research presented in this paper aims to accelerate the natural science research process by partially automating the execution of experiments using AI-assisted Concept-Change Detection (C-CD), e.g., for monitoring systems and studying biodiversity and ecosystem functions. The purpose of C-CD is to detect concept changes, also known as concept drift, that may be relevant to the study or ecosystem state. For example, in intertidal marine ecosystems, the event of sudden flooding can lead to dramatic changes in biodiversity. It could also be of scientific interest to take sensor samples more frequently in the period leading up to such events. The paper proposes an architecture for C-CD to customize AI-based analysis of sensor data streams. Furthermore, the paper implements portions of the architecture and is applied on sensor data from the Spiekeroog Coastal Observatory (SCO) as a feasibility study. The study demonstrates C-CD's ability to detect anomalies that are either of scientific or technical interest to the operation and exploration activities of SCO.