Artificial intelligence (AI) and machine learning (ML) have introduced new conceptual approaches to analyzing and interpreting complex data and show great promise for science. However, the needs of limnology require more than an application of existing ML techniques. Limnology, like most scientific disciplines, requires robust predictions, discovery of mechanism, and updates to existing conceptual models. Applying ML toward these needs requires overcoming a set of challenges: limnological data sets are relatively small and highly heterogeneous; ML requires specialized skills not included in most ecological training; and experts in AI will often conceive of a scientific problem differently than domain scientists. Confronting these challenges requires interdisciplinary collaboration and a flexible analytical framework. We present knowledge guided machine learning (KGML) as a paradigm for co-advancing the sciences of limnology and AI, providing both an analytical framework and a forum for interdisciplinary collaboration. KGML fuses the knowledge of limnology with contemporary deep learning algorithms and provides a framework for the formation of effective and productive interdisciplinary collaboration. KGML has improved the accuracy of predictions from hydrodynamic models, revealed missing processes in nutrient models, identified broad-scale changes in lake area, and is rapidly advancing our predictions and understanding of lake water quality. These preliminary KGML advances set the stage for a new generation of advancement for the science of limnology.

Authors:

Paul Hanson, University of Wisconsin - Madison, Center for Limnology  (pchanson@wisc.edu)

Cayelan Carey, Virginia Tech, Biological Sciences  (cayelan@vt.edu)

Arka Daw, Virginia Tech, Computer Science  (darka@vt.edu)

Hilary Dugan, University of Wisconsin - Madison, Center for Limnology  (hdugan@wisc.edu)

Xiaowei Jia, University of Pittsburgh, Computer Science  (XIAOWEI@pitt.edu)

Anuj Karpatne, Virginia Tech, Computer Science  (karpatne@vt.edu)

Ankush Khandelwal, University of Minnesota, Computer Science  (ankush.kwal@gmail.com)

Robert Ladwig, Aarhus University, Department of Ecoscience  (rladwig@ecos.au.dk)

Mary Lofton, Virginia Tech, Biological Sciences  (melofton@vt.edu)

Bennett McAfee, University of Wisconsin - Madison, Center for Limnology  (bmcafee@wisc.edu)

Abhilash Neog, Virginia Tech, Computer Science  (abhilash22@vt.edu)

Jordan Read, The Consortium of Universities for the Advancement of Hydrologic Science, Inc.  (jread@cuahsi.org)

Sophia Skoglund, University of Wisconsin - Madison, Center for Limnology  (skskoglund@wisc.edu)

Kathleen Weathers, Cary Institute of Ecosystem Studies  (weathersk@caryinstitute.org)

Vipin Kumar, University of Minnesota, Computer Science  (kumar001@umn.edu)