National Overview - March 2013
Changes in the Historical Palmer Drought Indices

« National Overview - March 2013


Changes in the Historical Palmer Drought Indices

The National Climatic Data Center has corrected a minor "bug" in the computer code used to compute the family of Palmer Drought Indices. The error was discovered by Jacobi, Perrone, Duncan, and Hornberger. This error is limited to one line of code (see ftp://ftp1.ncdc.noaa.gov/pub/data/software/palmer/pdi.f) that handles soil moisture holding capacity and as best NCDC can determine, this bug was introduced when Palmer's original research was first coded into Fortran IV and has been present ever since. The results of this fix for the Palmer Modified Drought Index (PMDI) are described below. Similar graphs and charts are provided for Palmer Hydrological (PHDI) and Palmer Drought Severity Index (PDSI), and complete period of record data sets are available at (http://www1.ncdc.noaa.gov/pub/data/cirs/). After February 2013, legacy datasets will no longer be updated.

The correction produced small changes in the PMDI (used operationally to compute the percent area of the U.S. affected by drought), the Palmer Hydrological Drought Index (PHDI), and the Palmer Drought Severity Index (PDSI). More information about these indices can be found at http://www.ncdc.noaa.gov/temp-and-precip/drought/historical-palmers.php.

The correction in the PMDI computation resulted in PMDI values nearly identical to the original computation. The correlation of the old data to new is 0.9994 (0.999 R2, see Table 1). The range of differences in the computations (New - Old) was -1.12 to +0.93 and the standard error was 0.217. With the exception of March 1941, the change in computational method in all other months and climate divisions (CD) (466,464 individual month and climate division combinations) resulted in a change of less than +/- 1.0 (Table 2). Eighty-seven percent of all these month/CD combinations resulted in changes smaller than +/- 0.1. This result was generally irrespective of month or of the drought intensity. Graphs are provided that show the changes that resulted when drought and wet spells of varying intensity occurred. A ranked cumulative distribution of the differences is also available.

The correction also resulted in changes to the historical rankings of the percent area of the U.S. experiencing drought (see Table 3). In the top 10, the recent drought of 2012 (July) moved from 9th place (Old) to 8th place (New), and July 1954 (Old 10th place) was supplanted by August 2012 (New 10th place). A complete ranking of the data can be accomplished using the data in the above link. The changes in the percent area are also shown in Figures 3a and 3b for moderate to extreme drought and moderate to extreme wetness and in 3c and 3d for severe to extreme drought and severe to extreme wetness.

There was little decadal change with the correction. In general, there was a slight increase in the percent area in drought as well as in the percent area experiencing wet spell conditions in all decades (Table 4, percent area moderately wet/dry and severely wet/dry graphs). On average, the 1910s and 1940s had a slightly bigger increase in drought area than the other decades. This result shows up in maps of the changes as well (see map viewer). Generally, looking at the most expansive drought years and months dry areas got a little drier, but most of the country saw very little change. Likewise, maps of the top 5 month and years with the single largest change in a climate division are shown (see map viewer). Overall, these months show little change over the country and no systematic issues.

We are very appreciative of work of Jacobi, Perrone, Duncan, and Hornberger in finding this long-standing bug in our code.



Table 1. Summary Statistics for Comparison on New vs. Old Palmer Runs. "Production" refers to the old operational code, and "New" refers to the modified code that has been implemented.
PMDIPHDIPDSI
Standard Deviation Production2.3622.4992.392
Standard Deviation New2.3802.5182.408
Range of Difference-1.12 to 0.93-3.71 -4.39-8.80 to 8.49
Standard Error0.0810.1540.217
Correlation0.9994 0.9981 0.9956
Covariance5.626.285.69