The Palmer Drought Severity Index: Limitations and Assumptions

View More View Less
  • 1 U.S. Geological Survey, Reston, VA 22092
© Get Permissions
Full access

Abstract

The structure of the Palmer Drought Severity Index (PDSI), which is perhaps the most widely used regional index of drought, is examined. The PDSI addresses two of the most elusive properties of droughts: their intensity and their beginning and ending times. Unfortunately, the index uses rather arbitrary rules in quantifying these properties. In addition, the methodology used to standardize the values of the PDSI for different locations and months is based on very limited comparisons and is only weakly justified on physical or statistical grounds. Under certain conditions, the PDSI values are very sensitive to the criteria for ending an “established” drought and precipitation during a month can have a very large effect on the PDSI values for several previous months.

The distribution of the PDSI conditioned on the value for the previous month may often be bimodal. Thus, conventional time series models may be quite limited in their ability to capture the stochastic properties of the index.

Abstract

The structure of the Palmer Drought Severity Index (PDSI), which is perhaps the most widely used regional index of drought, is examined. The PDSI addresses two of the most elusive properties of droughts: their intensity and their beginning and ending times. Unfortunately, the index uses rather arbitrary rules in quantifying these properties. In addition, the methodology used to standardize the values of the PDSI for different locations and months is based on very limited comparisons and is only weakly justified on physical or statistical grounds. Under certain conditions, the PDSI values are very sensitive to the criteria for ending an “established” drought and precipitation during a month can have a very large effect on the PDSI values for several previous months.

The distribution of the PDSI conditioned on the value for the previous month may often be bimodal. Thus, conventional time series models may be quite limited in their ability to capture the stochastic properties of the index.

Save