Drought Characteristics Derived Based on the Standardized Streamflow Index: A Large Sample Comparison for Parametric and Nonparametric Methods

Abstract

The streamflow drought hazard can be characterized in a variety of ways, including using different indices. Traditionally, percentile‐based indices, such as Q95 (the flow exceeded 95% of time), have been used by the hydrological community. Recently, the use of anomaly indices such as the Standardized Streamflow Index (SSI), a probability index‐based approach adopted from the climatological community, has increased in popularity. The SSI can be calculated based on various (non)parametric methods. Up to now, there is no consensus which method to use. This study aims to raise awareness how the inherent sensitivity of the SSI to the used method influences derived drought characteristics. We compared SSI time series computed with seven different probability distributions and two fitting methods as well as with different nonparametric methods for 369 rivers across Europe. Results showed that SSI time series and associated drought characteristics are indeed sensitive to the method of choice. A resampling experiment demonstrated the sensitivity of the parametric SSI to properties of both the low and high end of the sample. Such sensitivities might hinder a fair comparison of drought in space and time and highlight the need for a clear recommendation which method to use. We could recommend overall suitable methods, for example, from the parametric approaches, the Tweedie distribution has several advantageous properties such as a low rejection rate (2%) and a lower bound at zero. However, the most suitable method depends on the used evaluation criteria. Rather, we stress that shown approach‐specific sensitivities and uncertainties should be carefully considered.