Transferability of a neuro-fuzzy river ice jam flood forecasting model

Many northern communities are prone to spring ice jam floods but lack sufficient data to develop predictive warning models. The ability to transfer a model between river basins is highly desirable but has not previously been achieved due to the statistical, site specific nature of most river breakup models. Building on previous research, the ability to apply soft computing methods to model spring river ice breakup was further investigated with the goal of evaluating basin transferability and climate change scenarios using logic based fuzzy and neuro-fuzzy modeling techniques. The Athabasca River (prototype development site) and the Hay River (model transfer site) in northern Canada were chosen as test sites due to their propensity for ice jam related floods. The fuzzy logic based model was transferable between basins, in that extreme flood years could be distinguished from years when little flooding was reported; but the high quantitative accuracy of the neuro-fuzzy model was not reproduced at the second site. Climate change scenarios for the Athabasca River indicate a continuously decreasing risk of severe ice jams while the frequency in the Hay River basin is expected to first increase for a period before waning.