ACCEPTING THE STANDARDIZED PRECIPITATION INDEX: A CALCULATION ALGORITHM¹

TLDR

The Palmer Drought Severity Index has been used for decades, but the Standardized Precipitation Index was developed to better represent abnormal wetness and dryness, necessitating a standardized method for comparison. The study aims to develop a standard calculation method for the SPI so that it can be accepted as an alternative to the Palmer indices. The authors compare SPI values derived from different probability distributions, analyze how these models affect dry event characteristics, and provide Fortran 77 source code for the calculation. They conclude that the Pearson Type III distribution is the best universal model, that SPI reliability depends on sample size, and that time scales longer than 24 months may be unreliable.

Abstract

ABSTRACT: The Palmer Drought Severity Index (PDSI) has been calculated for about 30 years as a means of providing a single measure of meteorological drought severity. It was intended to retrospectively look at wet and dry conditions using water balance techniques. The Standardized Precipitation Index (SPI) is a probability index that was developed to give a better representation of abnormal wetness and dryness than the Palmer indices. Before the user community will accept the SPI as an alternative to the Palmer indices, a standard method must be developed for computing the index. Standardization is necessary so that all users of the index will have a common basis for both spatial and temporal comparison of index values. If different probability distributions and models are used to describe an observed series of precipitation, then different SPI values may be obtained. This article describes the effect on the SPI values computed from different probability models as well as the effects on dry event characteristics. It is concluded that the Pearson Type III distribution is the “best” universal model, and that the reliability of the SPI is sample size dependent. It is also concluded that because of data limitations, SPIs with time scales longer than 24 months may be unreliable. An internet link is provided that will allow users to access Fortran 77 source code for calculating the SPI.