Challenges and Opportunities of Standards for AI Methods for Ocean Applications

AI methods have established their role in different domains and is becoming pervasive also in environmental sciences like meteorology. Excellent examples of the efficiency of AI methods can be found in applications like pattern recognition in images or the analysis acoustic data. The advantage is that a large amount of data can be checked automatically and without human intervention. The situation is different with time series data, where the amount of data is not very large, and the interpretation must be combined with physical or biochemical models. What is still seen as an obstacle to a more thorough use of AI methods, especially machine learning, is the fact that ML algorithms are purely data-driven and only take modeling into account to a limited extent and are therefore not directly suitable for a validation system. Besides the selection of the right model for the individual science case there is also the issue of validating the resulting data and information. The role of standards in this context needs to be discussed and defined in order to pave the way for a wider adoption of AI methods.