Linear Temperature Trends for the "lower 48" United States


Attached are five files containing estimated linear trends for temperatures in the lower 48 states for 359 stations. The methodology describing the computation of the trends can be found in Temperature Trends in the United States by Robert Lund, Lynne Seymour, and Karen Kafadar, Environmetrics 2001, to appear. [abstract from a preliminary report]

"We hope that the trends can serve as a starting point for future research, as well as serve as an interesting example for smoothing algorithms." -- Robert Lund

The resulting temperature trend maps

Description of the underlying temperature data.

The data in this study are monthly averages of daily temperatures. Daily temperatures were computed by averaging the maximum and minimum temperatures for that day. These monthly averages were downloaded from the United States Historical Climatology Network (USHCN) The USHCN consists of 1221 stations from the 48 contiguous United States. The subset of this data which was adjusted for the bias due to time of observation was used; hence, all measurements are consistent with a midnight-to-midnight measuring scale.

Many of the 1221 stations were discarded due to missing data or length of coverage. Stations whose data records were less than 95 percent complete, or that had six or more consecutive points missing during the period 1922-1996, inclusive, were discarded. Stations that began recording after 1922 and stations that stopped recording before 1996 were also discarded. The starting date of some of the stations was advanced to ensure that the above data quality restrictions were met by all stations. Fourier interpolation was used to estimate any remaining missing values.

A site change (changepoint) is defined herein to be a change of station instrument, shelter, or location. Station records where the site moves twice within three years were discarded because the number of observations (1 or 2) was deemed too small to accurately estimate some of the monthly means in the Section 3 model. The dates of the site changes were rounded to the nearest month.

After the above preprocessing of the data, 359 stations remained. The last data point for each of these stations was recorded in December 1996. The date of the first observation varies with station, but always occurs during January. The average starting year is 1902, the earliest starting year is 1820 (Minneapolis MN), and the latest starting year is 1922 (6 stations). Every station in the study has at least 75 years of data.

Trend Data Format

Essentially, the data are supplied as a table. Each row pertains to a particular station, each "column" is as described below. "Column" is in quotes because the first column is a fixed length character string that can contain whitespace ...

"column" Variable Description
1 Station State&location of the temperature recording site. For example, AL GREENSBORO refers to Greensboro, Alabama.
2 lat the latitude of the station
3 long the longitude of the station
4 k the number of changepoints the station experienced during its period of record. A changepoint is a change of station location, temperature recording instrument, or shelter. The changepoint times are noted in the original data.
5 trend the estimated linear trend for the station in degrees Celsius per Century.
6 stnerr the uncertainty margin (standard error) of the trend estimate (col 5)
7 z-score the trend divided by the standard error -- [col5 divided by col6]
8 p the autoregressive model order of the errors terms fitted in the regression model for the station
9 q the moving-average model order of the errors terms fitted in the regression model for the station
10 lag1corr the autocorrelation at lag one (one month apart) of the autoregressive moving-average time series model describing the error terms in the regression model.