Eco-KGML Project Goals

Home - Project Goals - Workshops - The Team - Documents

Our goals are to develop a novel class of Ecology Knowledge Guided Machine Learning (Eco-KGML) models to explain dynamical lake-level behavior in large populations of lakes by learning hybrid compositions of process-based and machine learning modules that capture varying processes underlying lake water quality. Through model development and training across a range of scales, from focal lakes with high data volumes to U.S. lake populations with sparse data, we will investigate how exogenous drivers, watershed context, and water quality processes interact to determine macrosystem-scale water quality patterns. Our work will enable a new line of research in Eco-KGML models for lake water quality, where hybrid compositions of process-based and machine learning modules will not only improve our ability to predict water quality dynamics but also offer explainability of the underlying mechanisms behind lake processes and their interactions. Moreover, our research in Eco-KGML will have the capacity to learn and separate processes specific to a single lake from those that generalize across types of lakes according to ecological characteristics, e.g., lake area, trophic state, or hydrologic regime. This more flexible and comprehensive use of both knowledge and data will enable the study of scale-dependent relationships between lakes and their drivers while providing more robust predictions for lakes across those scales.