Wavelet-based automated localization and classification of peaks in streamflow data series

This paper combines discrete wavelet transform (DWT) with artificial intelligence algorithm in order to develop a new unsupervised method for fast detecting, localizing, and classifying flood events in real-world stage-discharge data time series. Localization is performed through a simple hill-climbing search algorithm initialized by the position of the highest DWT coefficients. The proposed method does not require any a priori information such as catchment characteristics or alert flood thresholds.

► We developed a new unsupervised method for fast localizing and classifying flood events.
► Discrete wavelet transform and hill-climbing search algorithm are used.
► Method does not require any a priori information such as catchment characteristics.
► The algorithm has been tested by applying it to a real-world discharge dataset.