Predicting Ice Jams With Neural Networks

Abstract

One of the most difficult problems facing hydraulicians is the development of a method that predicts the formation of breakup ice jams. Because of the suddenness with which breakup jams and related flooding occur, prediction methods are desirable to provide early warning and allow rapid, effective ice jam mitigation. Breakup ice jam prediction models are presently limited due to the lack of an analytical description of the complex physical processes, and range from empirical single-variable threshold-type analyses to statistical methods such as logistic regression and discriminant function analysis. In this study, a neural network method is used to predict breakup ice jams at Oil City, PA. Discussion of how the neural network input vector was determined and the methods used to appropriately account for the relatively low occurrence of jams are addressed. The neural network prediction proved to be more accurate than other methods attempted at this site.