Beyond Counting Zeroes: Using Entropy-Based Hydrologic Signatures and Classification for Streamflow Intermittency Assessment

When field observations are lacking or are difficult to obtain, streamflow intermittency is usually assessed from the frequency of zeroes in streamflow time series, using some (often arbitrary) definition of the zero-count threshold signaling an intermittent flow regime. This research proposes alternative and robust indicators for differentiating between perennial and intermittent flow regimes from discharge time series. We perform inductive hydrologic classification of 136 rivers and streams in Greece and Cyprus at daily and monthly scales, using selected hydrologic signatures. PCA is used to reduce the dimensionality of the input data, followed by classification using hierarchical clustering to group rivers into classes of similar hydrological characteristics. Two superclusters are defined, interpreted as corresponding to perennial and intermittent river regimes respectively. A subset of monthly hydrological signatures explaining most of the variance in the flow record is used with univariate clustering to cluster the data into two groups with maximum homogeneity. The resulting cluster membership is validated by assessing similarity with the clusters from the multivariate hydrologic classifications. Seven signatures are identified (entropy of flow volumes, entropy of flow distribution, CV of mean annual flow, index of flow variability, variability of mean monthly flows, average maximum monthly flow, and 10th percentile of monthly flows) as the most useful in differentiating between perennial and intermittent river regimes; cluster separation is also assessed visually on scatterplots between signature pairs. Overall, the entropy-based signatures outperform the others. Finally, threshold values of the proposed hydrologic signatures are identified for streamflow intermittency assessment.