Apurba Das, Ananya Kowshal, Sujata Budhathoki and Karl-Erich Lindenschmidt (2022). Application of Machine Learning approaches in ice-jam flood forecasting and prediction. Proceedings of the GWF Annual Open Science Meeting, May 16-18, 2022.

Ice jam formation is a key concern for many rivers in cold regions. Ice jams can lead to destructive floods during the ice season and create major disturbances in the aquatic environment. In the past years, numerous approaches, including numerical, empirical and data-driven models, have been applied to understand and potentially help reduce the detrimental impacts of ice-jam formation and flooding on riverine communities. Although these approaches are capable of solving many ice-related problems, they still have some limitations. Recent advancements in machine learning offers many methodological opportunities to deal with a plethora of data. This study explores the use of modelling output (e.g. hydrological and hydraulic model results and global circulation model (GCM) output) in various machine learning algorithms to predict river-ice hydraulic processes, such as ice-jam formation and mid-winter breakup along the Saint John River, which is a transboundary and transborder river in North America. A hydrodynamic model (RIVICE) was applied to simulate hundreds of river ice scenarios and pre-breakup hydraulic conditions of the river. The simulated hydraulic conditions of the river were then used in machine learning classification algorithms to forecast the location of ice-jam formation during spring breakup. Since recent changes in climatic conditions make the river more vulnerable to ice-jam formation, such as mid-winter breakup, this study examined the impacts of future climate on mid-winter breakup. A hydrological model (MESH) was applied to derive the streamflow conditions of the river under future climatic conditions. The temperature and precipitation data were derived from GCM output for similar future climatic conditions. The data were then applied in a machine-learning classification algorithm to predict the probability of mid-winter breakup along the river. This research will help understand the hydro-climatic impacts on the ice-jam processes and offer new knowledge to manage these complex river ice processes.

AOSM2022 Core modelling group First Author: Apurba Das, University Saskatchewan Additional Authors: Ananya Kowshal, Sujata Budhathoki and Karl-Erich Lindenschmidt