Research Dataset: Quantifying the Importance of Selected Drought Indicators for the U.S. Drought Monitor by Region and Season

In response to the need to better understand the effectiveness of various drought indicators for different regions and seasons, a team of scientists at the NASA Goddard Space Flight Center conducted research to quantify the relative importance—or “drought explainability”—of specific indicators with respect to the U.S. Drought Monitor (USDM). This research was published in the Journal of Hydrometeorology and was funded by NOAA’s National Integrated Drought Information System (NIDIS). 

In the study, the researchers looked at 113 indicators of drought, including in situ (observation-based), modeled, and remote-sensed data across a variety of timescales (from 1 week to 72 months).

The researchers used an information theoretic measure, mutual information, to compare these 113 indicators to U.S. Drought Monitor data from 2006-2019. They broke these results down by geographic region and season (spring, summer, fall, winter, and all seasons). The researchers calculated the mutual information for each of the 113 indicators individually, as well as for several groups of indicators (e.g., all remote-sensing indicators combined). The results show which drought indicators are most representative of the U.S. Drought Monitor’s depiction of drought for a given region and season. 

To do this, the research team first spatially interpolated and aggregated all the data to a climate division resolution. They then temporally interpolated and aggregated the indicators to the weekly resolution of the U.S. Drought Monitor maps, and finally calculated historical percentiles for each indicator. They used these percentiles to calculate the indicators’ mutual information with the U.S. Drought Monitor for six regions across the continental U.S., which mostly align with NOAA's Regional Climate Centers: High Plains, Midwest, Northeast, South, Southeast, and West. 

This approach looks at information entropy, or the randomness or uncertainty in a given variable. Mutual information is the degree to which one random data stream “explains” another, or the degree of common information between the two streams (e.g., the amount of information obtained about the U.S. Drought Monitor by observing a particular indicator). The researchers then normalized these values, so they could compare results across different locations. The resulting normalized mutual information is called fractional information. The fractional information results are available for download as CSV files below.

Fractional information = Mutual information (of each indicator with the USDM) / USDM information entropy

The researchers additionally calculated the fractional information for NOAA climate divisions (nClimDiv) and at a 1/24 degree (~4.6-km) resolution (nClimGrid). This involved sampling pixel by pixel, including neighboring pixels over a window to facilitate sample size requirements given missing data. These results, which are available as netCDF files below, are experimental and should be used for research purposes only. 

The input data used for this analysis are also available, including the data for each indicator and the percentiles for each indicator, calculated at the climate division level.

For more information or for any questions about the data, please contact the research team at the NASA Goddard Space Flight Center.