Recent Unusual Mean Winter Thomas R. Karl1, Robert E. Livezey2 Temperatures Across the Contiguous United States Abstract United States, even in an unusually cold (or warm) winter, can include a month of relatively mild (or cold) weather or A long-time series (1895-1984) of mean areally averaged winter contain small portions of the country that have relatively temperatures in the contiguous United States depicts an unprece- mild (or cold) weather throughout the winter. dented spell of abnormal winters beginning with the winter of 1975-76. Three winters during the eight-year period, 1975-76 through 1982-83, are defined as much warmer than normal (abnor- mal), and the three consecutive winters, 1976-77 through 1978-79, 2. Data much colder than normal (abnormal). Abnormal is defined here by the least abnormal of these six winters based on their normalized departures from the mean. When combined, these two abnormal The data set used to obtain areally weighted average winter categories have an expected frequency close to 21%. Assuming that temperature departures was originally used and described by the past 89 winters (1895-1984) are a large enough sample to esti- Diaz and Quayle (1978). The data, which begin in 1931, con- mate the true interannual temperature variability between winters, sist of monthly averages of temperatures for each of 344 state we find, using Monte Carlo simulations, that the return period of a series of six winters out of eight being either much above or much climatic divisions (CDs) in the contiguous United States, below normal is more than 1000 years. This event exceeds the calcu- and, prior to 1931, monthly averages of statewide averages of lated return period of the three consecutive much colder than normal temperature derived by simply averaging the data from all winters (1976-77 through 1978-79) all falling into a much below stations within a state. A national average is derived by normal category, i.e., one that is expected to contain approximately 10% of the data. The more moderate winters of 1981-82 and 1983-84 areally averaging the CDs to calculate state averages, and can also be considered abnormal by relaxing the limits necessary for then areally averaging the states to form a national average an abnormal classification, but this gives a return period of 467 years temperature. for the spell of eight abnormal winters in the nine consecutive win- Diaz and Quayle (1978) adjusted the data set prior to 1931 ters 1975-76 through 1983-84. by using the differences between the statewide average data and the statewide CD averages over an 18-year overlap pe- riod (1931-1948). They indicated that during that overlap 1. Introduction period, the station data were an excellent approximation to the CD data. Recently, those data were further scrutinized In recent years, there have been a considerable number of (Karl et al, 1983), and additional adjustments to the data press reports regarding unusual winter weather across the were necessary prior to 1931 for 11 western states (from United States, and a series of technical articles have ad- Montana to New Mexico and westward to the Pacific Ocean) dressed this topic (Diaz and Quayle, 1978; Diaz and Quayle, due to changes in station distribution. Karl et al (1983) de- 1980; Changnon, 1979; Cayan etal., 1982). In response to re- scribe these adjustments in some detail. cent requests for climatological information regarding win- After 1930, the CD averages are obtained from all stations ter weather in the United States over the decade 1973-74 within the CD reporting temperature and precipitation. through 1982-83, some interesting information has emerged; Since the cooperative weather observing stations make up specifically, the unusually large interannual fluctuations of the vast majority of these stations, it is noteworthy that be- the mean areal average winter temperatures across the ginning in the 1930s, there has been an increasing number of United States during the winters 1975-76 through 1982-83. cooperative weather observers reporting the maximum and These data are derived from cooperative and First Order minimum temperatures based on the 24 hours ending in the weather reporting stations in the contiguous United States. morning instead of the evening. Schaal and Dale (1977) The potential bias in this data set is discussed, as well as some noted that this practice produced a "cold" bias in the CD av- of the more subtle statistical problems encountered in the es- erages for stations in Indiana. timation of the return period for this group and other groups Based on the change in observation times, as derived from of abnormal winters. It should be recognized at the outset original manuscript records and as published in Climatological that an areal average mean winter temperature across the Data (U.S. Dept. of Commerce, 1980a), there were approxi- mately one-third more a.m. observers in 1980 than in 1931. This figure was derived from the records for seven states: Cali- fornia, Colorado, Illinois, Indiana, New York, North Caro- lina, and Washington. If these states are representative of the 1 National Climatic Data Center, Asheville, NC 28801. 2 Climate Analysis Center, Washington, DC 20233. entire United States, then an estimate of the cold bias in the time series due to observing schedules can be made using © 1984 American Meteorological Society data derived from Mitchell (1958), Baker (1975), Schaal and 1302 Vol. 65, No. 12, December 1984 Unauthenticated | Downloaded 10/06/21 07:15 AM UTC Bulletin American Meteorological Society 1303 TABLE 1. Average wintertime adjustments for a change of observation time from 1700 LST to 0800 LST. Observation Changes from to City, State Adjustment (°C) (LST) (LST) Source Washington, D.C. -1.0 1700 0800 Blackburn (1983) St. Paul, MN -1.0 1700 0800 Baker (1975) Indianapolis, IN -0.9 1700 0800 Schaal & Dale (1977) Austin, TX* -1.4 1730 0730 Mitchell (1958) Bismarck, ND* -1.2 1730 0730 Mitchell (1958) Columbus, OH* -1.3 1730 0730 Mitchell (1958) Denver, CO* -0.6 1730 0730 Mitchell (1958) Fresno, CA* -0.2 1730 0730 Mitchell (1958) Philadelphia, PA* -0.8 1730 0730 Mitchell (1958) Spokane, WA* -0.4 1730 0730 Mitchell (1958) Tampa, FL* -0.4 1730 0730 Mitchell (1958) Average -0.836 * January values are used to estimate wintertime adjustment. Dale (1977), and Blackburn (1983). Each of these authors be obtained. In 1980, 10% of the stations used in the CD av- calculated adjustments of various locations in the United erages were associated with urban areas with populations in States for changes in observation schedules. Those locations excess of 50 000. Assuming there has been no systematic in- include: 1) Washington, D.C., 2) Indianapolis, Ind., 3) St. crease or decrease in the percentage of stations monitoring Paul, Minn., 4) Austin, Tex., 5) Bismarck, N.D., 6) Colum- temperatures in urban areas of this size [this assumption ap- bus, Ohio, 7) Denver, Colo., 8) Philadelphia, Pa., 9) Spo- pears reasonable based on station distribution maps as plot- kane, Wash., 10) Tampa, Fla., and 11) Fresno, Calif. The ted in Decadal Census of Weather Stations (U.S. Depart- preferred a.m. and p.m. observation times are 0800 LSTand ment of Commerce, 1958)], then the overall urban influence 1700 LST respectively, except in the Central Time Zone, on the time series of winter temperatures would be 0.1 X where 0700 LST and 1700 LST prevail, as derived from Cli- 0.23°C or 5.57 X 10~4°C yr-1. This "warm" bias would apply matological Data (U.S. Department of Commerce, 1980a) for to the entire period of wintertime data for 1895-1984. We the seven states previously listed. When averaged across the would conclude that during the more recent 50 years of rec- 11 cities listed in Table 1, the average change in mean winter ord, the bias due to observation changes is estimated to be an temperature attributable to a switch from 1700 LST to 0800 order of magnitude larger than the bias due to urban effects. LST is approximately —0.8°C. This value can be used as a On the other hand, for some states the systematic time of ob- rough estimate of the wintertime temperature bias in 1980 servation changes and their associated biases can be miti- compared to 1931 if all stations had converted to a.m. from gated by the opening, closing, and relocations of stations, as p.m. observing schedules. Since only 33% of the observers demonstrated in Appendix A. converted to morning observations, the appropriate estimate In light of this discussion, the time series in Fig. 1 are pre- of the bias for the winter of 1980-81 would be -0.28°C. This sented for the contiguous 48 states. In this figure, one time is equivalent to —5.58 X 10"3oC yr"1 over the 50-year period series has been corrected for biases due to observation times since 1931. and urbanization while the other has no corrections. In our It would be remiss to consider the observation time bias subsequent analyses, both corrected and uncorrected time and omit the systematic "warm" bias that may have occurred series are used for comparative purposes. Small trends in the since the turn of the century due to urban influences. Numer- time series should be viewed cautiously for all of the reasons ous studies have documented these effects (Landsberg, discussed in this section. Emphasis should be placed on the 1981). Mitchell (1953) has separated out and quantified the large year-to-year climate fluctuations. urban influence on the mean seasonal and annual tempera- tures for six approximately equal areas in the contiguous United States by using data from 77 cities for the years 1900 to 1940.