Hydrologic regionalization using wavelet-based multiscale entropy method

- A novel multiscale wavelet entropy method is proposed for catchment regionalization.
- The method is applied to streamflow data from 530 stations in the United States.
- 14 clusters of catchments, with distinct wavelet multiscale entropy, are identified.
- The regionalization results are tested using four standard homogeneity tests.

SummaryCatchment regionalization is an important step in estimating hydrologic parameters of ungaged basins. This paper proposes a multiscale entropy method using wavelet transform and k-means based hybrid approach for clustering of hydrologic catchments. Multi-resolution wavelet transform of a time series reveals structure, which is often obscured in streamflow records, by permitting gross and fine features of a signal to be separated. Wavelet-based Multiscale Entropy (WME) is a measure of randomness of the given time series at different timescales. In this study, streamflow records observed during 1951-2002 at 530 selected catchments throughout the United States are used to test the proposed regionalization framework. Further, based on the pattern of entropy across multiple scales, each cluster is given an entropy signature that provides an approximation of the entropy pattern of the streamflow data in each cluster. The tests for homogeneity reveals that the proposed approach works very well in regionalization.