Sensitivity analysis of standardization procedures in drought indices to varied input data selections

- Sensitivities of SPDI and SC-PDSI to different datasets are compared.
- SPDI is less sensitive to variations in data samples than SC-PDSI.
- Scale parameter is the most sensitive factor in GEV for SPDI.
- Duration factors are main causes for the high sensitivity of SC-PDSI.

SummaryReasonable input data selection is of great significance for accurate computation of drought indices. In this study, a comprehensive comparison is conducted on the sensitivity of two commonly used standardization procedures (SP) in drought indices to datasets, namely the probability distribution based SP and the self-calibrating Palmer SP. The standardized Palmer drought index (SPDI) and the self-calibrating Palmer drought severity index (SC-PDSI) are selected as representatives of the two SPs, respectively. Using meteorological observations (1961-2012) in the Yellow River basin, 23 sub-datasets with a length of 30 years are firstly generated with the moving window method. Then we use the whole time series and 23 sub-datasets to compute two indices separately, and compare their spatiotemporal differences, as well as performances in capturing drought areas. Finally, a systematic investigation in term of changing climatic conditions and varied parameters in each SP is conducted. Results show that SPDI is less sensitive to data selection than SC-PDSI. SPDI series derived from different datasets are highly correlated, and consistent in drought area characterization. Sensitivity analysis shows that among the three parameters in the generalized extreme value (GEV) distribution, SPDI is most sensitive to changes in the scale parameter, followed by location and shape parameters. For SC-PDSI, its inconsistent behaviors among different datasets are primarily induced by the self-calibrated duration factors (p and q). In addition, it is found that the introduction of the self-calibrating procedure for duration factors further aggravates the dependence of drought index on input datasets compared with original empirical algorithm that Palmer uses, making SC-PDSI more sensitive to variations in data sample. This study clearly demonstrate the impacts of dataset selection on sensitivity of drought index computation, which has significant implications for proper usage of drought indices and related assessments, and potentially provide some valuable references for future researches on drought indices improvements.