Modelling Lake Glumsø with Q2 learning

In this paper, we describe an application of Q2 learning, a recently developed approach to machine learning in numerical domains, to the automated modelling of an aquatic ecosystem from measured data. We modelled the time behaviour of phytoplankton and zooplankton in Danish Lake Glumsø using data collected by S.E. Jørgensen. The novelty of Q2 learning is in its paying attention to the qualitative correctness of induced numerical models. We assessed the results by, first, performing a comparison of numerical accuracy between our approach and some state-of-the-art numerical machine learning algorithms applied to the Glumsø data, and second, we obtained expert evaluation of the induced models. The results show that Q2 approach is at least comparable to competing methods in terms of numerical accuracy and gives good insight into domain phenomena.