The prediction of mid-winter and spring breakups of ice cover on Canadian rivers using a hybrid ontology-based and machine learning model

Abstract

Mid-winter breakups (MWBs) are an increasingly common event on Canadian rivers resulting in the early breakup of river ice cover, with complex and numerous drivers. This study focuses on the development of an MWB Ontology which allows the key data, events, and relationships in an ice season to be defined and analyzed. The MWB Ontology is applied to a national case study of 54 rivers in Canada. Through assessment of the MWB Ontology with network analysis techniques, a hybrid modelling framework coupling the MWB Ontology with machine learning is developed. The hybrid models produce greatly reduced errors in their forecasts of the timing breakup events, with the best performance being a mean absolute error of 12.47 days for MWBs and 10.68 days for spring breakup. The results demonstrate the utility of the MWB Ontology as a tool for collating data and a means for analysis and forecasting of these events.