Forecast of water level and ice jam thickness using the back propagation neural network and support vector machine methods

Ice jams can sometimes occur in high latitude rivers during winter and the resulting water level rise may generate costly and dangerous flooding such as the recent ice jam flooding in the Nechako River in downtown Prince George in Canada. Thus, the forecast of water level and ice jam thickness is of great importance. This study compares three methods to simulate and forecast water level and ice jam thickness based on field observations of river ice jams in the Quyu Reach of the Yellow River in China. More specifically, simulation results generated by the traditional multi-variant regressional method are compared to those of the back propagation neural network and the support vector machine methods. The forecast of ice jam thickness and water level under ice jammed condition have been conducted in two different approaches, 1) simulation of water level and ice jam thickness in the second half of the period of measurement using models developed based on data gained during the first half of the period of measurement, 2) simulation of water level and ice jam thickness at the downstream cross sections using models developed based on data gained at the upstream cross sections. For this reason, as the results of simulation and field observations indicated, the back propagation neural network method and the support vector machine method are superior in terms of accuracy to the multi-variant regressional method.