Ming-Chang Wu

Effective forecasting of hourly typhoon rainfall using support vector machines

[1] Typhoon rainfall is one of the most difficult elements of the hydrologic cycle to forecast because of the high variability in space and time and the complex physical process. To obtain more effective forecasts of hourly typhoon rainfall, novel models with better ability are desired. On the basis of support vector machines (SVMs), which are a novel kind of neural networks (NNs), effective hourly typhoon rainfall forecasting models are constructed. As compared with backpropagation networks (BPNs), which are the most frequently used conventional NNs, SVMs have three advantages: (1) SVMs have better generalization ability, (2) the architectures and the weights of the SVMs are guaranteed to be unique and globally optimal, and (3) SVM is trained much more rapidly. An application is conducted to clearly demonstrate these three advantages. The results indicate that the proposed SVM-based models are better performed, robust, and efficient than the existing BPN-based models. To further improve the long lead time forecasting, typhoon characteristics are added as key input to the proposed models. The comparison between SVM-based models with and without typhoon characteristics confirms the significant improvement in forecasting performance due to the addition of typhoon characteristics for long lead time forecasting. The proposed SVM-based models are recommended as an alternative to the existing models. The proposed modeling technique is also expected to be useful to support reservoir operation systems and flood, landslide, debris flow, and other disaster warning systems.

The effect of data quality on model performance with application to daily evaporation estimation

The model performance is usually influenced by the quality of the data used in model construction. If the model performance is less affected by data quality, the resulting estimates will be more reliable. In this paper, the variation in model performance due to different data quality is explored in a field-scale application. Hence, two models, the proposed support vector machine (SVM) based model and the Stephen and Stewart (SS) model, are employed for daily estimation of evaporation at an experiment station. Five scenarios corresponding to different data qualities are designed to evaluate the effect of data quality on model performance. Additionally, the most effective meteorological variables influencing evaporation are obtained by a systematic input determination process. These most effective meteorological variables are used as inputs to the SVM-based model. The results show that the model performance decreases as the data quality decreases (i.e. the percentage of missing data increases). However, the estimation accuracy of SVM-based models is still better than that of the SS model. Moreover, the variation of model performance of the SVM-based model is smaller than that of the SS model. That is, the negative impact of different data quality is effectively decreased by using the SVM-based model instead of the SS model.