Abstract
In the literature, numerous papers report comparative analyses of drought indices. In these types of studies, the similarity between drought indices is usually evaluated using the Pearson correlation coefficient, r, calculated between corresponding severity time series. However, it is well known that the correlation does not describe the strength of agreement between two variables. Two drought indices can exhibit a high degree of correlation but can, at the same time, disagree substantially, for example, if one index is consistently higher than the other. From an operational point of view, two indices can be considered in agreement when they indicate the same severity category for a given period (e.g. moderate drought). In this work, we compared six meteorological drought indices based on both correlation analysis and Cohen's Kappa test. This test is typically used in medical or social sciences to obtain a quantitative assessment of the degree of agreement between different methods or analysts. The indices considered are five timescale-dependent indices, i.e. the Percent of Normal Index, the Deciles Index, the Percentile Index, the Rainfall Anomaly Index, and the Standardised Precipitation Index, computed at the 1-, 3-, and 6-month timescales, and the Effective Drought Index, a relatively new index, which has a self-defined timescale. The indices were calculated for 15 stations in the Abruzzo region (central Italy) during 1951–2018. We found that the strength of agreement depends on both the criteria of drought severity classification and the different indices' calculation method. The Cohen's Kappa test indicates a prevailing moderate or fair agreement among the indices considered, despite the generally very high correlation between the corresponding severity times series. The results demonstrate that the Cohen's Kappa test is more effective than the correlation analysis in discriminating the actual strength of agreement/disagreement between drought indices.
Highlights
It is difficult to give a univocal definition of drought, this can be very generically defined as a natural hazard that arises from a considerable deficiency in precipitation for prolonged periods (Folger 2017)
In the presentation and discussion of the results, the indices that require a timescale specification will be indicated with the notation Xk, where X is the generic acronym of the index and k is the timescale (e.g. SPI1 indicates the Standardised Precipitation Index (SPI) computed at the 1-month time scale)
The agreement in the drought severity classification obtained from some popular meteorological drought indices (Percentage of Normal Percent of Normal Index (PN), Deciles Index Decile Index (DI), Percentile Index PI, Rainfall Anomaly Index RAI, Standardised Precipitation Index SPI, and Effective Drought Index EDI) has been evaluated for a typical Mediterranean region, based on a 68-year time series
Summary
It is difficult to give a univocal definition of drought, this can be very generically defined as a natural hazard that arises from a considerable deficiency in precipitation for prolonged periods (Folger 2017). Drought characterisation in a given area is very useful, both for defining early-warning systems (Kogan 2000), and for water resources planning and management (Zargar et al 2011). For these purposes, drought indices are typically used: these are measures deriving from the elaboration of suitable variables or data such as precipitation, evapotranspiration, or satellite images, which allow us to express the drought characteristics (e.g. severity) in a more effective way than the original information (Zargar et al 2011). This way of proceeding has the advantage of allowing a rapid understanding of the type of ongoing event and of making its description independent of the scale of values assumed by a particular index
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
