Ice breakup forecast in the reach of the Yellow River: the support vector machines approach

Accurate lead-time forecast of ice breakup is one of the key aspects for ice flood prevention and reducing losses. In this paper, a new data-driven model based on the Statistical Learning Theory was employed for ice breakup prediction. The model, known as Support Vector Machine (SVM), follows the principle that aims at minimizing the structural risk rather than the empirical risk. In order to estimate the appropriate parameters of the SVM, Multiobjective Shuffled Complex Evolution Metropolis (MOSCEM-UA) algorithm is performed through exponential transformation. A case study was conducted in the reach of the Yellow River. Results from the proposed model showed a promising performance compared with that from artificial neural network, so the model can be considered as an alternative and practical tool for ice breakup forecast.