Modeling a Drought Index Using a Nonparametric Approach

Alternatively, to other studies that used parametric distributions (e.g. Gamma) in the estimation of the Standardized Precipitation Index (SPI), this study aims to apply a nonparametric method based on Kernel Density Estimator (KDE) for calculating the SPI. Results of the proposed method were compared with the ones from the most widely used parametric distribution, using a long dataset of monthly precipitation of four meteorological stations in Iran (including Bushehr, Mashhad, Tehran and Esfahan) over a period of 107 water years (1895-2002). The capability of KDE-based SPI was compared with the Gamma-based SPI at four-time scales of 3, 6, 9 and 12 months. The frequencies of the drought classes of SPI were calculated and compared with corresponding expected frequencies. The results revealed that the KDE is more consistent with the expected values of the SPI drought/wet classes frequencies (especially in the extreme classes) at all stations as well as at the four-time scales, compared to the Gamma distribution. The greatest deviation from the expected frequencies for KDE and Gamma distribution were about 10% and 150%, respectively. This study proposes a new analytical approach in modeling SPI that provides more accurate results pertaining frequency of occurrences of extreme drought events. The output of the study can be used in many fields (e.g. tourism, agriculture, insurance, etc.) that are influenced by severe droughts.