Gridded U. S. Daily Temperature (maxima and minia) ------- -- -- ----- ----------- These data sets are described more thoroughly in NCEP/CPC Atlas No. 6 which is titled "A Gridded Data base of Daily Temperature Maxima and Minima for the Conterminous United States: 1948-1993" By John E. Janowiak, Gerald D. Bell and Muthuvel Cheliah of the Climate Prediction Center/NCEP/NWS/NOAA. Copies of this atlas can be obtained by e-mailing jjanowiak@ncep.noaa.gov. A separate atlas for daily gridded snowfall is under preparation; however the data format is the same as for the temperature data. These data are available from the NCEP public data server (ftp.ncep.noaa.gov). To get the data simply ftp to: ftp.ncep.noaa.gov (anonymous login) and then cd pub/precip/daily_grids. Then cd to either "snow" "tmax" or "tmin" and get the files for each year that is desired. a. General Description Daily observations of maximum and minimum temperature were extracted from the "Cooperative Summary of the Day" collection that is produced by the National Climatic Data Center (NCDC). That product is a compilation of data from more than 6000 recording stations in the United States. This count reflects all stations that reported at least once during the 1948-1993 period. Of these stations, approximately 3150 have reported at least 75% of the time in this 46 year period. At present, the gridded data set extends to 1993. It will be updated periodically as new information become available from NCDC. b. Quality Control The quality control effort was a combination of objective and subjective procedures as described below. The underlying philosophy of our quality control procedure was to modify the data as little as possible and only in cases where we felt confident that suspicious values were wrong. No attempt was made to correct suspicious reports; if a value was deemed wrong, the value was changed to the standard missing indicator ("-9999."). The data were first checked for unrealistic values and such values were removed from the dataset. Then data quality checks were tailored to specific data types. We ensured that the maximum temperature was never lower than the reported minimum temperature (and vice versa for minimum temperature). Furthermore, reports were flagged if the reported maximum temperature exceeded 46oC (115oF) or was below -29oC (-20oF). We then subjectively viewed these flagged reports and made decisions on the validity of the suspicious reports by considering the location of the station, time of year, and surrounding reports. We also incorporated the state record temperature reports into our decision making. If at all plausible, reports were left untouched. If not, they were set to "missing". The same procedure was followed for minimum temperature, except that reports were flagged if the reported minimum temperature exceeded 38oC (100oF) or was below -35oC (-58oF). At least one type of error that remains in the data set is phase error. By this we mean reports that appear to be reasonable in magnitude but which consistently occur a day (perhaps more) later or before other nearby stations over the course of days or even years. Such errors are extremely difficult to detect and this difficulty is compounded by the large volume of data that needs to be checked for such errors. c. Data Reorganization and Gridding An objective analysis scheme (Cressman 1959) was used to interpolate the station data to a 0.5 x 0.5 latitude/longitude grid. For each variable, a 141 x 71 array of gridded values was produced for each of the 16802 days during the 1948-1993 period. Note that the station density is sparse in regions of the West, particularly in Nevada, thus a 0.5 x 0.5 latitude/longitude is tenuous in such regions. As with any objective analysis procedure, the data are smoothed to some degree and the analysis bleeds into regions where no observations exist, such as over water in coastal regions. Therefore, a land-sea mask was used to mask out values over water surfaces in all maps in this atlas. A source of erroneous temporal variability looms if data are analyzed without regard to station reporting frequency. To illustrate this potential problem, suppose a gridbox that is home to a handful of stations, one of which routinely reports significantly lower minimum temperatures than the others (for example, a valley station in the mountainous western U.S.), so much so that it influences the gridbox mean. If that station fails to report one day (or year), the time series of mean temperature for that gridbox would abruptly jump for no physical reason but simply due to variable data sampling. One remedy for this situation is to use station reporting frequency to determine whether or not to include a given station. Based on our experience, however, we feel that such artificial variability, as described above, is not significant for studies on regional and continental spatial-scales, especially since the data have been analyzed using a Cressman technique which incorporates information from stations outside of a particular gridbox. d. Data Organization The data are stored in "little endian" (PC) binary as generated on a RedHat Linux operating system. Thus, users of HP & SGI (for example) workstations need to swap bytes in order to use these data. For such users, a "C" program called "swap.c" has been placed under the "daily_grids" directory. Simply compile the program using a "C" compiler. To swap the bytes for a filename "xxxx" simply type: swap xxxx xxxx.ieee The contents of file "xxxx.ieee" will then be readable on HP, SGI, etc. workstations. Each file contains the data analyses for a single year and is composed of 365 (or 366 for leap year) direct access records. Each record is a daily analysis (record 1 is for January 1, etc.) of length 10011 4-byte words (141 x 71 Fortran array). Each record is arranged such that the first 141 values are valid for latitude 20N, the 2nd 141 values are for 20.5N, etc.. For each of these "rows", the first value is for 130W, the 2nd for 129.5W, etc.. For GrADS users, each directory contains the appropriate ".ctl" file.