An ensemble of classification method for anomalous propagation echo detection with clustering based subset selection method

There are several types of non-precipitation echoes which appear in radar images and disrupt weather forecasting process. An anomalous propagation echo is one of unwanted observation result and has fairly similar characteristics compared with precipitation echoes like rain echo or snow echo. It is occurred by either super-refraction or ducting of radar beam because of temperature and humidity, or other complicated atmospheric conditions. Considering that the anomalous propagation echo makes the weather forecasting process hard, it should be removed from radar data. There are several ongoing researches about distinguishing the anomalous propagation echo from observed radar data. In this paper, we suggest an ensemble classification method based on artificial neural network and partition method using clustering. The method allows to implement an efficient classification method when a feature space has complicated distributions by separating input data into atomic and non-atomic data using a clustering method. Then each derived cluster will get its own base classifier using an artificial neural network which has a good performance in classification and applied various practical fields. By comparing the suggested method to a traditional artificial neural network classifier, it is confirmed that the suggested method can derive better performance than the traditional one.