• Barnston, A. G., 1994: Linear statistical short-term climate predictive skill in the Northern Hemisphere. J. Climate,7, 1513–1564.

  • Chagnon, S. A., 1999: Impacts of 1997–98 El Niño-generated weather in the United States. Bull. Amer. Meteor. Soc.,80, 1819–1827.

  • Deser, C., and J. M. Wallace, 1987: El Niño events and their relationship to the Southern Oscillation: 1925–86. J. Geophys. Res.,92, 14 189–14 196.

  • ——, and ——, 1990: Large-scale atmospheric circulation features of warm and cold episodes in the tropical Pacific. J. Climate,3, 1254–1281.

  • Gershunov, A., and T. P. Barnett, 1998: ENSO influence on intraseasonal extreme rainfall and temperature frequencies in the contiguous United States: Observations and model results. J. Climate,11, 1575–1586.

  • Halpert, M. S., and C. F. Ropelewski, 1992: Surface temperature patterns associated with the Southern Oscillation. J. Climate,5, 577–593.

  • Hastenrath, S., 1991: Climate Dynamics of the Tropics. Kluwer, 488 + xx pp.

  • Hoerling, M. P., and A. Kumar, 1997: Origins of extreme climate states during the 1982–83 ENSO winter. J. Climate,10, 2859–2870.

  • ——, ——, and M. Zhong, 1997: El Niño, La Niña, and the nonlinearity of their teleconnections. J. Climate,10, 1769–1786.

  • Hughes, P., 1982: Weather, climate and the economy. Weatherwise,35, 60–63.

  • IPCC, 1996: Observed climate variability and change. Climate Change 1995. The Science of Climate Change, J. T. Houghton et al., Eds., Cambridge University Press, 133–192.

  • Karl, T. R., R. W. Knight, D. R. Easterling, and R. G. Quayle, 1996:Indices of climate change for the United States. Bull. Amer. Meteor. Soc.,77, 279–292.

  • Kiladis, G. N., and H. F. Diaz, 1989: Global climate anomalies associated with extremes in the Southern Oscillation. J. Climate,2, 1069–1090.

  • Livezey, R. E., and W. Y. Chen, 1983: Statistical field significance and its determination by Monte Carlo techniques. Mon. Wea. Rev.,111, 46–59.

  • ——, M. Masutani, A. Leetmaa, H. Rui, M. Ji, and A. Kumar, 1997:Teleconnective response of the Pacific–North American region atmosphere to large central equatorial Pacific SST anomalies. J. Climate,10, 1787–1820.

  • Mearns, L. O., R. W. Katz, and S. H. Schneider, 1984: Extreme high-temperature events: Changes in their probabilities with changes in mean temperatures. J. Climate Appl. Meteor.,23, 1601–1613.

  • NCDC, 1994: Time bias corrected divisional temperature–precipitation–drought index. Documentation for Dataset TD-9640, NCDC, 12 pp. [Available from Data Base Management Branch, NCDC, Federal Building, 37 Battery Park Ave., Asheville, NC 28801-2733.].

  • Palmer, T. N., and J. A. Owen, 1986: A possible relationship between some “severe” winters in North America and enhanced convective activity over the tropical West Pacific. Mon. Wea. Rev.,114, 648–651.

  • Piechota, T. A., and J. A. Dracup, 1996: Drought and regional hydrological variation in the United States: Associations with El Niño–Southern Oscillation. Water Resour. Res.,32, 1359–1373.

  • Riebsame, W. E., S. A. Changon Jr., and T. R. Karl, 1991: Drought and Natural Resource Management in the United States. Impacts and Implications of the 1987–89 Drought. Westview Press, 174 + xvi pp.

  • Ropelewski, C. F., and M. S. Halpert, 1986: North American precipitation and temperature patterns associated with the El Niño/Southern Oscillation (ENSO). Mon. Wea. Rev.,114, 2352–2362.

  • ——, and P. D. Jones, 1987: An extension of the Tahiti–Darwin Southern Oscillation index. Mon. Wea. Rev.,115, 2161–2165.

  • Spiegel, M. R., 1975: Probability and Statistics. Schaum’s Outline Series, McGraw-Hill, 372 pp.

  • Trenberth, K. E., and D. S. Shea, 1987: On the evolution of the Southern Oscillation. Mon. Wea. Rev.,115, 3078–3096.

  • ——, and G. W. Branstator, 1992: Issues in establishing causes of the 1988 drought over North America. J. Climate,5, 159–172.

  • ——, and J. W. Hurrell, 1994: Decadal atmosphere–ocean variations in the Pacific. Climate Dyn.,9, 303–319.

  • ——, G. W. Branstator, and P. A. Arkin, 1988: Origins of the 1988 North American drought. Science,242, 1640–1645.

  • Troup, A. J., 1965: The Southern Oscillation. Quart. J. Roy. Meteor. Soc.,91, 490–506.

  • van Loon, H., and R. A. Madden, 1981: The Southern Oscillation. Part I: Global associations with pressure and temperature in northern winter. Mon. Wea. Rev.,109, 1150–1162.

  • Walker, G. T., and E. W. Bliss, 1932: World Weather V. Mem. Roy. Meteor. Soc..,4, 53–84.

  • Wolter, K., and M. S. Timlin, 1998: Measuring the strength of ENSO events—How does 1997/98 rank? Weather,53, 315–324.

  • View in gallery

    Idealized (a) normal distribution and (b) shifted normal distribution. An upward shift in the mean of 0.5 σ translates into a doubled risk of warm extremes (shaded above 1 σ), compared to the unshifted case, and a halved risk of cold extremes (shaded below −1 σ).

  • View in gallery

    Significant t-test scores of the SON SOI (leading by one season) in relation to the (a) 20 warmest and (b) 20 coldest winters at each climate division within the Dec–Feb 1895/96–1994/95 period. Cross-hatching at light and heavy densities denotes the 95% and 99% significance levels, respectively, for positive SOI (El Niño conditions); light and heavy gray shading refers to the 95% and 99% significance levels for negative SOI (La Niña).

  • View in gallery

    As in Fig. 2, except with the DJF SOI leading by one season in relation to the (a) 20 warmest and (b) 20 coldest springs at each climate division.

  • View in gallery

    As in Fig. 2, except for t-test scores of the JJA SOI leading by one season in relation to the 20 coldest autumns at each climate division.

  • View in gallery

    Significantly increased risks of climate division winter temperature extremes for the SON SOI leading by one season during (a) El Niño and (b) La Niña conditions within the DJF 1895/96–1994/95 period. Relative risks at or above the 175% and 200% levels of either very cold or very warm extremes are shown here, corresponding to 93.6% and 98.2% significance levels.

  • View in gallery

    As in Fig. 5, except for significantly increased risks of U.S. spring temperature extremes in relation to preceding DJF ENSO conditions.

  • View in gallery

    As in Fig. 5, except for significantly increased risks of U.S. summer temperature extremes in relation to preceding MAM SOI La Niña conditions.

  • View in gallery

    As in Fig. 5, except for significantly increased risks of U.S. autumn temperature extremes in relation to preceding JJA SOI El Niño conditions.

  • View in gallery

    As in Fig. 5, except for significantly suppressed risks of U.S. winter temperature extremes in relation to the preceding SON SOI during (a) El Niño and (b) La Niña conditions. Risk reductions by 75% and 100% for either very cold or very warm extremes are shown here and represent the 95% and 99.3% significance levels.

  • View in gallery

    As in Fig. 5, except for significantly suppressed risks of U.S. spring temperature extremes in relation to the preceding DJF SOI during (a) El Niño and (b) La Niña conditions.

  • View in gallery

    Conditional probabilities of extreme quintile temperature categories in winter for Louisiana (average of all climate divisions) in relation to (a) El Niño and (b) La Niña during the period 1896–1995, for three-season to zero lead times. Quintile values refer to 20 out of 100 possible values. The unconditional probability of occurence (20%) is indicated by a solid line. Significant positive or negative departures from the expected probability (at the 93.6% level or higher) are denoted by an asterisk.

  • View in gallery

    As in Fig. 11, except for Texas spring temperatures.

  • View in gallery

    As in Fig. 11, except for Washington spring temperatures.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 26 26 4
PDF Downloads 15 15 2

Short-Term Climate Extremes over the Continental United States and ENSO. Part I: Seasonal Temperatures

View More View Less
  • 1 NOAA–CIRES Climate Diagnostics Center, Boulder, Colorado
© Get Permissions
Full access

Abstract

This study documents statistical relationships between the El Niño–Southern Oscillation (ENSO) phenomenon and extreme seasonal temperature anomalies over the continental United States. Relationships are examined for El Niño and La Niña conditions for each of the four standard seasons. Two complementary approaches are used. In the first approach, seasonal temperature anomalies are ranked from coldest to warmest over a 100-yr climate division dataset. Mean Southern Oscillation index (SOI) values are then computed for times preceding or concurrent with extreme seasonal temperature anomalies to define regions where relationships between the SOI and seasonal temperature extremes are statistically significant. In the second approach, seasonal extremes in the SOI, which are generally related to El Niño or La Niña, are first identified, and then the numbers of extreme temperature seasons occurring in association with these events are determined. Comparison of the observed number of extreme seasons with the climatologically expected values provides quantitative estimates of how ENSO alters the conditional probability, or risk, of large seasonal temperature anomalies in a given region.

The results show that the greatest geographical coverage of statistically significant relationships between ENSO and seasonal temperature extremes occurs in winter and spring, especially with the SOI leading by one season. Certain well-recognized relationships for seasonal temperature anomalies are also confirmed for extreme seasons, such as the association of El Niño conditions with very warm winters over the Pacific Northwest and very cold winters along the Gulf Coast. Other less-discussed relationships also appear, including possible nonlinearities in relationships between El Niño and La Niña events and extremes in autumn temperatures. Some relationships show evidence of secular changes, especially in summer.

In some regions and times of year, El Niño and La Niña conditions substantially alter the probabilities of very warm or very cold seasons. For example, over Texas, El Niño conditions in winter almost triple the risk that the subsequent spring will be very cold, while significantly reducing the risk of a very warm spring. In the same region, wintertime La Niña conditions double the risk that the following spring will be very warm, while significantly reducing the likelihood of a very cold spring. Therefore, given the proper ENSO phase, skillful forecasts of regional risks of seasonal temperature extremes appear feasible.

Corresponding author address: Dr. Klaus Wolter, NOAA–CIRES Climate Diagnostics Center, R/E/CD1, 325 Broadway, Boulder, CO 80303-3328.

Email: kew@cdc.noaa.gov

Abstract

This study documents statistical relationships between the El Niño–Southern Oscillation (ENSO) phenomenon and extreme seasonal temperature anomalies over the continental United States. Relationships are examined for El Niño and La Niña conditions for each of the four standard seasons. Two complementary approaches are used. In the first approach, seasonal temperature anomalies are ranked from coldest to warmest over a 100-yr climate division dataset. Mean Southern Oscillation index (SOI) values are then computed for times preceding or concurrent with extreme seasonal temperature anomalies to define regions where relationships between the SOI and seasonal temperature extremes are statistically significant. In the second approach, seasonal extremes in the SOI, which are generally related to El Niño or La Niña, are first identified, and then the numbers of extreme temperature seasons occurring in association with these events are determined. Comparison of the observed number of extreme seasons with the climatologically expected values provides quantitative estimates of how ENSO alters the conditional probability, or risk, of large seasonal temperature anomalies in a given region.

The results show that the greatest geographical coverage of statistically significant relationships between ENSO and seasonal temperature extremes occurs in winter and spring, especially with the SOI leading by one season. Certain well-recognized relationships for seasonal temperature anomalies are also confirmed for extreme seasons, such as the association of El Niño conditions with very warm winters over the Pacific Northwest and very cold winters along the Gulf Coast. Other less-discussed relationships also appear, including possible nonlinearities in relationships between El Niño and La Niña events and extremes in autumn temperatures. Some relationships show evidence of secular changes, especially in summer.

In some regions and times of year, El Niño and La Niña conditions substantially alter the probabilities of very warm or very cold seasons. For example, over Texas, El Niño conditions in winter almost triple the risk that the subsequent spring will be very cold, while significantly reducing the risk of a very warm spring. In the same region, wintertime La Niña conditions double the risk that the following spring will be very warm, while significantly reducing the likelihood of a very cold spring. Therefore, given the proper ENSO phase, skillful forecasts of regional risks of seasonal temperature extremes appear feasible.

Corresponding author address: Dr. Klaus Wolter, NOAA–CIRES Climate Diagnostics Center, R/E/CD1, 325 Broadway, Boulder, CO 80303-3328.

Email: kew@cdc.noaa.gov

1. Introduction

Extreme short-term climate anomalies have enormous consequences for the United States. For example, direct losses attributed to persistent very cold weather over the central and eastern United States and drought in the West during the 1976/77 winter season were estimated at nearly $40 billion (Hughes 1982), while the 1988 drought and associated heat waves over the Midwest caused losses estimated at $39 billion and contributed to significantly increased death rates from heat stress (Riebsame et al. 1991; Trenberth and Branstator 1992). Effects of extreme climate events may also be beneficial;for example, the abnormally warm winter of 1997/98 over much of the United States contributed to large reductions in heating costs and to a record home construction season (Changnon 1999). It is clear that improved predictions of such climate events provide the potential for major economic benefits; however, given the chaotic nature of midlatitude flows, deterministic predictability at seasonal timescales is unlikely. Nevertheless, useful probability forecasts may still be possible, if the relative likelihood of such events is substantially altered by slowly varying and potentially predictable components of the climate system, such as tropical sea surface temperature anomalies associated with El Niño.

In recent years, scientists and the general public have increasingly invoked El Niño, or more generally the El Niño–Southern Oscillation (ENSO), as a cause for a broad array of extreme climate (and other) events. Over certain regions, particularly parts of the Tropics, evidence for such a relationship is overwhelming (Hastenrath 1991). Over much of the United States however, the extent of ENSO relationships to extreme climate events is less well established. There is substantial evidence for links in specific cases (e.g., Palmer and Owen 1986; Trenberth et al. 1988; Hoerling and Kumar 1997) and in higher-frequency events for selected regions (e.g., Piechota and Dracup 1996; Gershunov and Barnett 1998). However, many, if not most, seasonal climate extremes in the United States occur in years without El Niño or La Niña conditions, and therefore the extent to which ENSO influences the likelihood of such extreme events is uncertain.

The relationships between ENSO and seasonal mean anomalies in U.S. temperatures and precipitation have been documented in several studies (e.g., Ropelewski and Halpert 1986, hereafter RH; Halpert and Ropelewski 1992, hereafter HR; Kiladis and Diaz 1989, hereafter KD; and, most recently, Livezey et al. 1997). Kiladis and Diaz consider essentially linear relationships by subtracting the average temperature (and precipitation) anomalies during El Niño years from the corresponding fields during La Niña years. Halpert and Ropelewski allow for sign-dependent nonlinearity by examining separately “high” and “low” Southern Oscillation index (SOI)–temperature anomalies, while Livezey et al. (1997) distinguish between warm and cold central Pacific SST anomalies.

The present work differs from these previous studies by focusing on relationships between ENSO and shifts in the extremes, or tails, of the seasonal climate distributions rather than on changes in mean values. Although the mean ENSO-related climate anomalies identified by RH and KD appear relatively small (typically around one-half standard deviation in sensitive regions), even such small shifts may substantially alter the frequency of occurrence of extreme events, as defined from the unshifted distribution. For example, as illustrated in Fig. 1, given a normal distribution with fixed standard deviation σ, an increase in the mean of 0.5 σ doubles the frequency of positive anomalies exceeding 1 σ from the original distribution and triples the frequency of positive anomalies exceeding 2 σ. The large sensitivity of extreme events to relatively small changes in the mean has been discussed by Mearns et al. (1984) in the context of global climate change. Clearly, changes in higher moments can also alter the frequency of occurrence of extreme events, and in general the climate forecast problem involves estimating how the complete distribution is altered subject to particular initial and boundary conditions.

Ropelewski and Halpert (1986) and KD, as well as most studies in this area, use what is essentially a prospective analysis technique based on prior identification of ENSO events. That is, given ENSO conditions (say El Niño or La Niña events defined by a certain criterion), concurrent or subsequent climate anomalies are then identified. This approach is useful for predictive purposes, if ENSO is the predictor and the climate anomalies are the desired predictand. Such an approach does not, however, directly address how often extreme climate events occur in non-ENSO years or whether extreme events in a region occur only or primarily under specific ENSO conditions.

The present study applies two different approaches to studying relationships between ENSO and U.S. climate extremes. The first approach is essentially retrospective. That is, based on prior identification of the climate events of interest (e.g., the 20 coldest winters in Iowa in a 100-yr period), an ENSO index will be examined to determine whether a systematic ENSO signal preceded or was concurrent with these events. The second approach is prospective and based on an ENSO index, as in earlier studies, but rather than focusing on mean climate shifts, the emphasis instead is on determining how ENSO alters the conditional probability, or risk, of events in the tails (extremes) of the climate distribution.

Formally, the distinction between retrospective and prospective approaches to studying ENSO–extreme-event relationships can be understood as the difference between conditional probability distributions, P(X|y) and P(Y|x). Here, the former represents the conditional probability distribution of a measure of ENSO X under extreme event conditions y, while the latter represents the conditional distribution of a climate variable Y under ENSO conditions x. As examples of the two approaches, the former might be a distribution of sea surface temperatures associated with very cold winters in Iowa, while the latter could represent the distribution of winter temperatures in the Midwest under El Niño conditions. In general, the two approaches are not the same but rather are complementary.

The remainder of the paper is structured as follows. Section 2 describes the primary datasets, section 3 discusses analysis methods, section 4 presents retrospective analyses of ENSO–seasonal temperature extreme relationships, section 5 shows seasonal risk analyses, and section 6 summarizes the main results.

2. Data

Surface temperatures were obtained from the U.S. monthly climate division data for the 100-yr period 1896–1995 (NCDC 1994). In order to be able assess the robustness of the analyzed relationships, we will compare results for the full record against results for the first and second half of the record. Early in the period, observational data are likely inadequate to support analyses at spatial resolutions implied in the divisional data; however, it was felt valuable to include this earlier time to analyze the active ENSO period around the turn of the century. Because of this resolution issue, results should be considered only on the spatial scale of states or larger regions rather than of individual climate divisions.

The SOI, here defined by the standardized sea level pressure difference between Tahiti and Darwin (Troup 1965), is used as a measure of ENSO. The SOI has been employed frequently in earlier ENSO analyses, especially since its extension back to the last century (Ropelewski and Jones 1987). It is well known that low and high phases of the SOI do not always coincide with El Niño and La Niña events, respectively, especially if the oceanic events are defined strictly by coastal Peruvian SSTs (Deser and Wallace 1987, 1990); however, the relationship is sufficiently strong that in the following we will often refer to very low SOI periods as El Niño and very high SOI periods as La Niña, particularly to help distinguish these events from very warm and very cold seasons in the United States.

Many of the analyses described subsequently have also been conducted with other ENSO indexes (e.g., Wolter and Timlin 1998); however, the present study will focus on the SOI in order to assess century-long relationships with U.S. temperatures and to facilitate comparisons with earlier results (e.g., HR; KD).

3. Analysis methods

For each climate division and standard season, seasonal temperature anomalies were rank ordered from largest negative to largest positive values in the respective data periods (50 or 100 years). For the 100-yr dataset, the 20 lowest and highest ranked seasons (e.g., the 20 coldest and 20 warmest winters) were identified, corresponding to the lowest and highest quintiles of the seasonal temperature distributions. Similarly, for the 50-yr data subsets (1896–1945 and 1946–95), seasonal temperature anomalies were ranked for each climate division and the 10 coldest and warmest seasons selected, again corresponding to the lowest and highest quintiles of these distributions. Note that this classification corresponds with conventional “much below normal” and“much above normal” categories. This threshold for identifying extreme seasons is broad but provides a reasonable sample size in these relatively short datasets. Additional analyses with tighter thresholds suggest that the basic results discussed below are robust, but the term“extreme” should be interpreted subject to these considerations.

Seasonal values of the SOI were derived from monthly SOI time series for the 100-yr period as described in appendix A. To identify systematic temporal relationships between ENSO and extreme seasons, corresponding SOI values were composited from three to zero seasons preceding the extreme seasons. Analyses were constructed separately for the warmest and coldest seasons to aid in identifying possible nonlinearities in the occurrence of extreme seasons related to ENSO (e.g., Hoerling et al. 1997). Statistical significance was assessed by a two-sided t test with a null hypothesis that the mean SOI associated with the extreme seasons was zero. The potential interdependence of results from different climate divisions further requires the testing of field significance (Livezey and Chen 1983). Appendix B provides details of the method used for assessing field significance.

For the risk analyses of section 5, the SOI values for each season were first ranked from most negative to most positive values. The 20 (10) lowest and highest SOI values were used to identify El Niño and La Niña conditions, respectively, in each calendar season of the 100- (50-) yr datasets. ENSO cases identified by this procedure may differ from other definitions; however, years with at least two ENSO “seasons” correspond with El Niño or La Niña years in HR and KD. Table 1 summarizes the classifications for the SOI in the 100- (50-) yr datasets. El Niño and La Niña seasons in the first and last 50-yr subperiods are largely consistent with the full 100-yr classification. However, during spring only 6 out of 20 La Niña cases have occurred since 1945, while 12 out of 20 El Niño cases have occurred during the same period.

Table 1 suggests three epochs of ENSO activity: a very active period from 1896 through 1926 (for both El Niño and La Niña); then suppressed activity through the 1960s; followed by more active behavior, especially more El Niños, through 1995 (and beyond). The first epoch featured 20% more ENSO seasons than the long-term average, the middle period over 25% less, and the most recent period 20% more. Although other ENSO indexes might alter this assessment slightly, the tendencies toward lower ENSO activity through the middle part of this century and more frequent El Niño events recently are well established (e.g., Trenberth and Shea 1987; Trenberth and Hurrell 1994).

Another consideration is possible longer-term changes in U.S. temperatures. Seasonally analyzed trends in this century (Karl et al. 1996; IPCC 1996) indicate significant warming in parts of the northern United States during winter and spring, especially in the most recent 50 yr, and a more varied pattern of cooling and warming in summer and autumn. Possible relationships between these longer-term trends and seasonal temperature extremes will be addressed in the discussion.

To assess statistical significance in the risk analyses, probability distributions for the numbers of events expected through random sampling without replacement were calculated from the hypergeometric distribution (Spiegel 1975). Table 2a shows the probability of selecting n extreme cases in a sample of size s = 20 from a population of N = 100 seasons when the individual event probability is p = 0.2. For this full set, n ⩽ 1 extreme events would be expected in a random sample of 20 seasons 5% of the time, suggesting a possible significant suppression of such events, while n ≥ 7 would be expected by chance just over 6% of the time, and n ≥ 8 about 2% of the time. Table 2b documents the analogous distribution for s = 10 and N = 50, with event probability p = 0.2. For this shorter set, obtaining n = 0 (no extreme events) out of a sample of 10 seasons would be expected by chance 8% of the time, while n ≥ 4 would be expected by chance 10% of the time, and n ≥ 5 just under 2% of the time.

It is also useful to consider how given conditions change the risks of extreme events relative to the (unconditional) expected values. For this purpose, a relative risk r can be defined as the ratio of the observed number of events to this expected value. For the 100-yr dataset, the expected number of events for s = 20 is E(n) = 4. In this case, eight events would correspond to r = 2, or a doubled risk. Another interpretation is that if 12 out of 20 seasons (say, associated with El Niño) experienced climate events—for example, much below normal spring temperatures in Texas—then the estimated conditional seasonal risk would be 60%, compared to a normal risk of 20%.

4. Retrospective analyses

This section examines results of retrospective analyses between seasonal temperature extremes and ENSO. For times prior to or concurrent with extreme temperature seasons at each climate division, the composite mean SOI is calculated and the statistical significance evaluated by the methods described in section 3.

Figure 2a displays, for the 20 warmest winters (DJF) of the record (1896–1995) at each climate division, where the composite mean SOI in the previous autumn (SON) was significantly different from zero as estimated from a two-sided t test. Figure 2b shows the corresponding analysis for the 20 coldest winters. In these figures, cross-hatching indicates those climate divisions where the mean SOI is significantly less than zero (usually related to El Niño conditions), and gray shading indicates climate divisions where the mean SOI is significantly greater than zero (usually related to La Niña conditions).

Figure 2a shows that, along parts of the West Coast, western Rockies, and upper Midwest, very warm winters are preceded by negative SOI values the previous fall, suggesting a relationship to El Niño. Very warm winters over the southeastern United States from Texas to Florida are related to positive SOI values the previous fall, suggesting a relationship to La Niña. The areal coverage of the continental United States where the mean SOI values is significantly different from zero at the 5% level is 21.5%, which is more than four times as high as expected by chance and is field significant at the 1% level.

In contrast, for very cold winters (Fig. 2b), only very limited regions of the country show systematic relationships to nonzero SOI values the previous autumn, and the areal coverage is not field significant at the 5% level. Those areas where very cold winters appear linked to nonzero SOI values include eastern Montana and the western Dakotas (positive SOI), and the Gulf Coast region (negative SOI). Seen in conjunction, only small portions of the Gulf Coast and the northwestern United States show the symmetric relationships between very warm and very cold winters and low and high SOI stages that might be inferred from earlier work (e.g., Walker and Bliss 1932; van Loon and Madden 1981; RH; KD;HR).

Table 3 shows, for all four seasons, the percentage areal coverage of the United States at which t-test values of the SOI are significantly different from zero at the 5% level when the SOI leads by zero to three seasons. Table 3a indicates that the stronger relationship of warm winters rather than cold winters to ENSO conditions prevails for both the first and second 50-yr subperiods, as well as for most lead times from zero to three seasons, with largest areal coverage being for very warm winters at zero lead time (map not shown).

Figure 3 presents similar analyses to Fig. 2 for very warm and very cold springs (MAM) in relation to the SOI of the previous winter (DJF). The relationships between the SOI and extreme springs are stronger than in winter, particularly with the SOI leading by one to three seasons (Table 3b). Very warm springs in the Pacific Northwest occur predominantly with negative SOI values (Fig. 3a), consistent with winter El Niño conditions, while very cold springs in these regions tend to be preceded by winters with significant positive SOI values, consistent with La Niña conditions (Fig. 3b). Very warm springs over most of Texas and the southern plains are preceded by positive SOI values in winter, while very cold springs are preceded by negative SOI values. The overall patterns in Fig. 3 suggest a greater degree of symmetry in spring between ENSO and seasonal temperature extremes than for other seasons.

Similar analyses have been conducted for the summer (JJA) and fall (SON) seasons as well, but their relationships appear less stable. For example, in the most recent 50-yr period, very warm summers in the northeastern quadrant of the country are associated with El Niño conditions the previous autumn (not shown). Barnston (1994) noted a similar relationship in his canonical correlations linking recent U.S. summer temperature variability to a global tropical warming mode that may be initiated by El Niño. However, the early 50-yr period showed instead a tendency for very cold summers west of the Mississippi under the same conditions (not shown), while the full 100-yr analysis provides little evidence for either relationship (Table 3c). Whether this change in spatial relationships represents genuine climate change between the earlier and later periods or simply reflects sampling variability cannot be determined at this time. Since the significant t tests refer to modest composite mean SOI values for both periods, these results are more vulnerable to sampling variability than the higher mean SOI values found for the other seasons. In autumn, field significant results are reached only at a one-season lead (Table 3d), with the central United States being very cold from the upper Midwest to Texas following El Niño conditions the previous summer (Fig. 4).

Table 3 suggests that, for most times of year, there is a stronger relationship in the United States between ENSO and very warm seasons than between ENSO and very cold seasons. Associations of seasonal temperature extremes with El Niño or La Niña are strongest for winter and spring. Early versus recent half-century results also appear more stable for these seasons than summer and autumn. The largest areal coverage, highest t-test scores, and most linear relationships of U.S. seasonal temperature extremes to ENSO are in spring, with field significant associations at leads of one through three seasons. However, the overall areal coverage where the SOI differs significantly from zero is relatively modest at all times of year, never more than 30%. This indicates that, over much of the country, extremes in seasonal temperatures occur over a broad range of ENSO conditions. Specific ENSO conditions therefore are not a necessary prerequisite for seasonal temperature extremes in most regions and seasons. However, this does not rule out the possibility that a given ENSO phase may still alter the conditional probability of occurrence, or risk, of extreme events, as discussed in the next section.

5. Risk analyses

This section provides quantitative estimates of how given ENSO conditions influence the risk of very warm or very cold seasons in the continental United States. Analyses were performed relative to times when the SOI was either in the lowest quintile (El Niño) or highest quintile (La Niña) of its seasonal values over the respective time period (as shown in Table 1).

Risk analyses were carried out for both the full 100-yr period and separately for the first and last 50-yr periods following the methods described in section 3. A subset of these analyses is presented here. Zero-lag SOI temperature (and precipitation) risk maps for all running three-month seasons derived from the 100-yr climate division dataset are available online at http://www.cdc.noaa.gov/Climaterisks/. The following risk maps show areas where the risk ratio r is significantly greater than or less than one, corresponding to either significantly increased or reduced risk relative to unconditional (climatological) values. For the 100-yr dataset, increased risk levels of r ≥ 1.75 and r ≥ 2.0 are presented in section 5a, which are statistically significant at the 6.4% and 1.8% levels (Table 2). Reduced risk levels of r ⩽ 0.25 and r = 0 are featured in the risk maps of section 5b, corresponding to 5% and 0.7% significance levels (Table 2). Although possible in principle, significantly increased (or decreased) risks of both cold and warm seasonal extremes are very rare in the 100-yr analyses, amounting to less than 1% of the total domain for most cases computed here.

a. Areas of increased risks

Figure 5a shows, based on the full 100-yr record, areas of significantly increased risk of extreme cold or warm winters (DJF) when El Niño conditions exist the previous fall (SON). Fall El Niño conditions significantly increase the probability that the subsequent winter will be much warmer than normal over much of the West Coast and from the northern Rockies to the upper Midwest. Many of these areas show at least a doubled relative risk (r ≥ 2.0). In addition, risks of much below normal temperatures increase over most of southern Texas and the western Gulf. For fall La Niña conditions, the general features are broadly similar but with signs reversed (Fig. 5b). Under these conditions, the risk of very cold winters increases along the northwest coast and especially over the northern plains eastward to Lake Superior, while the risk of very warm winters increases over southern Texas and the Southeast.

Comparisons of the above results with the winter t-test analyses (Fig. 2) show broad similarities but also some interesting differences, with areas of significant risk increases seemingly covering more area than significant t-test results (see also Table 4). However, the significance thresholds for the risk increases (one sided, 6.4%) are less stringent than the t-test levels (double sided, 5%), and the finite sample size does not allow for standard significance levels (Table 2). We performed a separate analysis with relaxed t-test thresholds at the levels used for the risk analyses (not shown here). Areal coverage of the one-sided t-tests at the 6.4% level was comparable to the increased risk coverage at all seasons and lags. The relative absence of significant t-test results for cold winters (Fig. 2b) suggested that cold winters may occur over a wide range of ENSO conditions. The prospective risk analyses (Fig. 5) indicate, however, that if ENSO events are in place the previous fall, then the risk of much below normal winter temperatures increases significantly in certain regions, such as the western Gulf Coast, with El Niño, and the northern plains, with La Niña.

Figure 6 documents similar analyses of the risks of extreme spring temperatures when El Niño (Fig. 6a) or La Niña conditions (Fig. 6b) occur the previous winter. Wintertime El Niño conditions significantly increase the risk of much above normal temperatures the following spring along most of the West Coast and over the Pacific Northwest, and of much below normal springtime temperatures over a large area from eastern New Mexico and Texas eastward across much of the Southeast (Fig. 6a). Following wintertime La Niña conditions, parts of the Southwest and southern plains have significantly increased risks of very warm springs, while the risk of much below normal springtime temperatures increases mainly over parts of the Pacific Northwest, Great Lakes, and New England (Fig. 6b).

Spring risk patterns generally correspond well with the spring t-test patterns (Fig. 3) and show the greatest overall consistency of any season. For the 100-yr record, the areal coverage for significantly enhanced risks is consistently much higher than expected by chance at all lead times from zero to three seasons (Table 4), possibly related to the well-known persistence of ENSO events from boreal summer through winter. Areal coverage of significant risks is typically larger for preexisting El Niño conditions than for La Niña conditions (Table 4b), suggesting a somewhat more systematic effect of El Niño than of La Niña on spring temperature extremes.

Figure 7 shows extreme summer temperature risk enhancements associated with preceding spring La Niña conditions. Risks of much below normal summer temperatures increase significantly over much of the Midwest, particularly the central plains. Average spatial coverage of significant summer risk enhancements is low (Table 4c), particularly for El Niño events (maps not shown), with only a modest peak at a one-season lead. Taken together with the earlier results, it appears that over most of the United States extreme summer seasonal temperatures are possible under many different ENSO conditions, but that the risk of an abnormally cool summer increases over the central United States following the La Niña phase in spring.

A rather surprising result occurs for the fall season, which shows increased risks of much below normal temperatures over most of the northwestern half of the country following El Niño conditions the previous summer (Fig. 8). There is some correspondence of this pattern with the earlier t-test results, but the risk analyses show substantially larger areal coverage (cf. Fig. 4 and Tables 3 and 4), even with relaxed t-test thresholds (not shown). Interestingly, there is no corresponding strong relationship between very warm autumn conditions in the same region and La Niña the previous summer (Table 4, map not shown). Separate analyses for the first and second 50-yr periods support the above relationship for summer El Niño conditions and the absence of a corresponding La Niña signal (not shown). This lack of symmetry suggests potentially significant nonlinearities in autumn ENSO–U.S. climate relationships, as discussed by Hoerling et al. (1997), and may explain the absence of significant relationships in prior empirical studies. However, recent monthly U.S. temperature analyses by Livezey et al. (1997) link warm September central Pacific SST to cold September temperatures in the northwestern United States, thereby lending partial support to our results.

Table 4 summarizes, for all four seasons, the percentage areal coverage of the continental United States with statistically significant risk increases for seasonal temperature extremes under either El Niño or La Niña conditions for zero to three seasons in advance. Coverage at or beyond the 6% significance level exceeds 30% of the country for several seasons and lead times under El Niño conditions but only once under La Niña conditions. El Niño therefore appears to exert a larger and more consistent effect than La Niña on U.S. temperature risks. Overall, areal coverage is roughly the same for significant risk increases as for significant t-test scores, once taking into account differences in significance levels (not shown here). From the retrospective analyses, it appears that, in most regions, seasonal temperature extremes occur over a broad range of ENSO conditions. However, once an ENSO event is established, the prospective risk analyses indicate that it may significantly change the likelihood of a subsequent extreme event over a substantial part of the country.

Consistent with the retrospective results, spring shows the largest areal coverage of statistically significant risk increases, many of the highest regional risks, and the most symmetric relationships between U.S. seasonal temperature extremes and ENSO. Summer features the least evidence of ENSO influences on seasonal risks (Table 4c). Autumn risk results (Table 4d) are consistent with, but show larger areal coverage than, the retrospective analyses of section 4. Comparing risk and t-test results for the two 50-yr periods, risk analyses appear more stable than t-test results (not shown here) and less sensitive to long-term secular variability.

b. Areas of reduced risks

The primary focus so far has been to identify areas at increased risk of experiencing extreme seasons. However, identifying areas of significantly reduced risks of extreme events is also of considerable potential value. Figure 9 shows an analysis of areas of reduced risk of winter temperature extremes, given either El Niño or La Niña conditions in the previous autumn.

For El Niño conditions in fall, the risk that the subsequent winter will have much below normal temperatures decreases significantly for parts of the northern tier of states from the Pacific Northwest eastward to northern New England, while the risk of an extremely warm winter decreases significantly over much of the Southeast (Fig. 9a). This pattern appears somewhat shifted from the pattern of increased risks (with sign reversed) in Fig. 5a. In some regions, there are very large shifts in the comparative risks of very warm versus very cold winters (see also section 5c). Given La Niña conditions in fall, there are significantly reduced risks of very warm winters over most of the northern plains and of very cold winters over the Gulf Coast region (Fig. 9b). This distribution of reduced risks leads to large shifts in comparative risks of very warm versus very cold winters in the Dakotas and Minnesota (cf. Fig. 5b).

Figure 10 documents areas of reduced risk of spring temperature extremes, given either El Niño or La Niña conditions in the previous winter. For El Niño winters, the risk that the subsequent spring temperatures will be much below normal decreases significantly from the Pacific Northwest eastward to the western Great Lakes, while the risk of an extremely warm spring decreases significantly over the southern plains and much of theSoutheast (Fig 10a). Following La Niña winters, there is a significant reduction in the likelihood of very cold springs over portions of Texas and the adjacent Southwest (Fig. 10b). These patterns roughly mirror (with sign reversed) the patterns of increased risks (Fig. 6), resulting in large shifts in the comparative risks of very warm versus very cold springs.

Table 5 documents the areal coverage of significantly suppressed risks for seasonal temperature extremes given preceding El Niño or La Niña conditions. Coverage at the 5% significance level exceeds 30% for El Niño at all lead times with respect to spring temperatures. No other time of year or ENSO phase features such widespread suppression of extreme seasonal temperature risks, although La Niña winters come close (Table 5a; Fig. 9b). Consistent with the previous analyses, suppressed risk coverage for summer and autumn stays below 15% for almost all cases (Tables 5c,d). El Niño suppresses climate risks over a larger part of the United States than La Niña (Table 5), similar to results for significantly enhanced risks discussed earlier (Table 4). Further, a one-season lead for the SOI generally shows a stronger relationship than other leads for nationwide suppressed seasonal temperature risks.

The 50-yr analyses for suppressed risks (not shown here) are less stable than for enhanced risks. This, however, may be due in part to the stringent requirement that no events occur in the record for signficance at the 8.3% level (cf. Table 2b). Therefore, even a single outlier strongly affects a 50-yr analysis. Also, 50-yr risk analyses can be more difficult to interpret due to the occasional suppression of both warm and cold seasonal extremes during one ENSO phase.

c. Regional examples

Many parts of the country experience significant shifts in seasonal temperature risks following El Niño or La Niña conditions. We present here a few examples to illustrate how the odds of regional temperature extremes change in relation to ENSO conditions.

Compared to other parts of the country, Louisiana shows very symmetric ENSO relationships in winter, when the comparative risk of cold versus warm winters reaches 9:1 following El Niño autumns (Fig. 11a), while the comparative risk of warm versus cold winters is 8:1 after La Niña autumns (Fig. 11b). Significant comparative risks begin at the two-season lead (Fig. 11), consistent with the observation that ENSO summer conditions tend to persist into the winter.

Spring temperatures are particularly likely to be extreme for the state of Texas in association with ENSO phases. For the 20 El Niño winters of this study, Texas has experienced very cold springs 11 times, and no very warm springs at all (Fig. 12a), while La Niña winters have been followed by very warm springs eight times, and never by very cold springs (Fig. 12b). As is typical for most U.S. spring–ENSO relationships, significant changes in probabilities are common from the three-season lead onward (Fig. 12).

In contrast, Washington springtime temperature extremes show increased risks of very warm conditions in the wake of El Niño, while La Niña shows the opposite if somewhat smaller tendency (Fig. 13). Specifically, during El Niño 10 much above normal springs and no much below normal springs occurred in Washington (Fig. 13a). La Niña winters often precede cold springs at the one-season lead (Fig. 13b), while significant suppression of very warm springs follows La Niña conditions in fall (Fig. 13b).

6. Discussion

In this study, two primary approaches were used to identify relationships between ENSO and extremes in U.S. seasonal temperatures. In the first approach, seasons characterized by either much above or much below normal temperatures were first identified, and then the composite mean SOI determined for times relative to these extreme seasons. This enabled the recognition of regions where El Niño or La Niña conditions systematically precede or are concurrent with the extreme seasons. In the second approach, ENSO conditions were first identified, and then the numbers of extreme seasons occurring with specific ENSO conditions determined. This allowed definition of how particular ENSO conditions change the conditional probability, or risk, of extreme events. Both approaches were tested for robustness, by comparing early versus recent half-century records against the full 100-yr record. Many of the results discussed here were also confirmed with other ENSO indexes.

For both analyses, the strongest relationships were found in the winter and spring. Even for these seasons, the retrospective analyses showed that areal coverage where the mean SOI was significantly different from zero was modest, never higher than 30%. In most parts of the continental United States, extreme temperature seasons may occur over a wide range of ENSO conditions. In these regions, El Niño or La Niña conditions are not necessary for (or even strongly related to) the occurrence of the extreme temperature seasons. Conversely, in regions where the SOI was significantly different from zero, a strong systematic relationship exists between the occurrence of the extreme seasons and specific ENSO phases. Given preexisting El Niño or La Niña conditions, prospective risk increases (and suppressions) of seasonal extremes were most common in winter and spring. For autumn, extreme seasonal temperatures are possible under a variety of ENSO conditions, but if El Niño occurs in the previous summer, the risk of very cold autumns significantly increases over much of the northwestern half of the country.

Long-term trends in U.S. temperatures, as discussed by Karl et al. (1996), may influence some of the results, particularly the comparisons of the first and second 50-yr periods. However, it cannot be determined conclusively from our analyses whether these secular changes reflect true long-term trends or simply sampling variations. For example, the recently reduced areal coverage of significant t-test results in winter (Table 3a) is consistent with the decreasing importance of El Niño conditions to precede warm winters in the northwestern United States and cold winters in the Southeast, since both regions have been affected by long-term temperature trends in the same sense (IPCC 1996). However, possible secular changes in other seasons seen in our analyses are much harder to reconcile with long-term temperature trends.

For the period examined here, increased risk results generally show less secular variability between the first and second 50-yr periods than the t-test results. This may reflect a relatively greater sensitivity of the t-test results to variations in sample statistics (e.g., sample mean and standard deviation). For example, results for the 50-yr periods indicate that a few outliers can significantly reduce t-test values. However, the 50-yr suppressed risk results also can be strongly affected by outliers.

For some regions, under specific ENSO conditions one to three seasons in advance, very large shifts occur in the comparative risks of very warm seasons versus very cold seasons, sometimes changing comparative ratios from the unconditional value of 1:1 to 10:1 or more. This suggests that, at least in these regions, probabilistic forecasts of seasonal temperature extremes are feasible. Such predictions may have useful applications in a variety of sectors, for example, energy and utilities. Further work will be required to determine if increased skill in predicting seasonal extremes may be obtained through additional empirical predictors or through combined use with model forecasts.

Acknowledgments

Support for this research was provided through the NOAA Office of Global Programs. Kriste Paine (formerly of CDC) helped with initial processing and early analyses. Craig Anderson deserves thanks for digitizing the U.S. map used here. Discussions with Joe Barsugli, Marty Hoerling, Brant Liebmann, Prashant Sardeshmukh, and Jeff Whitaker (all at CDC) are gratefully acknowledged.

REFERENCES

  • Barnston, A. G., 1994: Linear statistical short-term climate predictive skill in the Northern Hemisphere. J. Climate,7, 1513–1564.

  • Chagnon, S. A., 1999: Impacts of 1997–98 El Niño-generated weather in the United States. Bull. Amer. Meteor. Soc.,80, 1819–1827.

  • Deser, C., and J. M. Wallace, 1987: El Niño events and their relationship to the Southern Oscillation: 1925–86. J. Geophys. Res.,92, 14 189–14 196.

  • ——, and ——, 1990: Large-scale atmospheric circulation features of warm and cold episodes in the tropical Pacific. J. Climate,3, 1254–1281.

  • Gershunov, A., and T. P. Barnett, 1998: ENSO influence on intraseasonal extreme rainfall and temperature frequencies in the contiguous United States: Observations and model results. J. Climate,11, 1575–1586.

  • Halpert, M. S., and C. F. Ropelewski, 1992: Surface temperature patterns associated with the Southern Oscillation. J. Climate,5, 577–593.

  • Hastenrath, S., 1991: Climate Dynamics of the Tropics. Kluwer, 488 + xx pp.

  • Hoerling, M. P., and A. Kumar, 1997: Origins of extreme climate states during the 1982–83 ENSO winter. J. Climate,10, 2859–2870.

  • ——, ——, and M. Zhong, 1997: El Niño, La Niña, and the nonlinearity of their teleconnections. J. Climate,10, 1769–1786.

  • Hughes, P., 1982: Weather, climate and the economy. Weatherwise,35, 60–63.

  • IPCC, 1996: Observed climate variability and change. Climate Change 1995. The Science of Climate Change, J. T. Houghton et al., Eds., Cambridge University Press, 133–192.

  • Karl, T. R., R. W. Knight, D. R. Easterling, and R. G. Quayle, 1996:Indices of climate change for the United States. Bull. Amer. Meteor. Soc.,77, 279–292.

  • Kiladis, G. N., and H. F. Diaz, 1989: Global climate anomalies associated with extremes in the Southern Oscillation. J. Climate,2, 1069–1090.

  • Livezey, R. E., and W. Y. Chen, 1983: Statistical field significance and its determination by Monte Carlo techniques. Mon. Wea. Rev.,111, 46–59.

  • ——, M. Masutani, A. Leetmaa, H. Rui, M. Ji, and A. Kumar, 1997:Teleconnective response of the Pacific–North American region atmosphere to large central equatorial Pacific SST anomalies. J. Climate,10, 1787–1820.

  • Mearns, L. O., R. W. Katz, and S. H. Schneider, 1984: Extreme high-temperature events: Changes in their probabilities with changes in mean temperatures. J. Climate Appl. Meteor.,23, 1601–1613.

  • NCDC, 1994: Time bias corrected divisional temperature–precipitation–drought index. Documentation for Dataset TD-9640, NCDC, 12 pp. [Available from Data Base Management Branch, NCDC, Federal Building, 37 Battery Park Ave., Asheville, NC 28801-2733.].

  • Palmer, T. N., and J. A. Owen, 1986: A possible relationship between some “severe” winters in North America and enhanced convective activity over the tropical West Pacific. Mon. Wea. Rev.,114, 648–651.

  • Piechota, T. A., and J. A. Dracup, 1996: Drought and regional hydrological variation in the United States: Associations with El Niño–Southern Oscillation. Water Resour. Res.,32, 1359–1373.

  • Riebsame, W. E., S. A. Changon Jr., and T. R. Karl, 1991: Drought and Natural Resource Management in the United States. Impacts and Implications of the 1987–89 Drought. Westview Press, 174 + xvi pp.

  • Ropelewski, C. F., and M. S. Halpert, 1986: North American precipitation and temperature patterns associated with the El Niño/Southern Oscillation (ENSO). Mon. Wea. Rev.,114, 2352–2362.

  • ——, and P. D. Jones, 1987: An extension of the Tahiti–Darwin Southern Oscillation index. Mon. Wea. Rev.,115, 2161–2165.

  • Spiegel, M. R., 1975: Probability and Statistics. Schaum’s Outline Series, McGraw-Hill, 372 pp.

  • Trenberth, K. E., and D. S. Shea, 1987: On the evolution of the Southern Oscillation. Mon. Wea. Rev.,115, 3078–3096.

  • ——, and G. W. Branstator, 1992: Issues in establishing causes of the 1988 drought over North America. J. Climate,5, 159–172.

  • ——, and J. W. Hurrell, 1994: Decadal atmosphere–ocean variations in the Pacific. Climate Dyn.,9, 303–319.

  • ——, G. W. Branstator, and P. A. Arkin, 1988: Origins of the 1988 North American drought. Science,242, 1640–1645.

  • Troup, A. J., 1965: The Southern Oscillation. Quart. J. Roy. Meteor. Soc.,91, 490–506.

  • van Loon, H., and R. A. Madden, 1981: The Southern Oscillation. Part I: Global associations with pressure and temperature in northern winter. Mon. Wea. Rev.,109, 1150–1162.

  • Walker, G. T., and E. W. Bliss, 1932: World Weather V. Mem. Roy. Meteor. Soc..,4, 53–84.

  • Wolter, K., and M. S. Timlin, 1998: Measuring the strength of ENSO events—How does 1997/98 rank? Weather,53, 315–324.

APPENDIX A

Construction of Seasonal SOI Time Series

Monthly SOI time series as defined by Troup (1965) were initally obtained on the Web at http://www.dnr.qld.gov.au/longpdk/lpsoidat.htm. This time series is serially complete for January 1900 through the present (July 1999). It is calculated based on the anomalous sea level pressure difference between Tahiti and Darwin, divided by the respective standard deviation for the same month. Earlier years (1895–99) were subsequently appended from a second Australian SOI, available at http://www.bom.gov.au/climate/current/soihtml.shtml. To ensure continuity, these SOI time series were compared against Ropelewski and Jones (1987) for the overlapping early record. We smoothed the monthly SOI binomially into five-monthly averages to yield seasonal indices. For instance, the spring values are calculated from (1 × FEB + 4 × MAR + 6 × APR + 4 × MAY + 1 × JUN)/16.

APPENDIX B

Monte Carlo Field Significance Testing

Field significance was assessed using Monte Carlo techniques based on Livezey and Chen (1983). The seasonal SOI was scrambled randomly 1000 times. This scrambling was achieved by 1) drawing for each run 100 (50) uniformly distributed random numbers, 2) sorting these numbers in ascending order (along with a yearly index for the SOI), and 3) reshuffling the SOI values according to the “sorted” yearly index. After each shuffle, the scrambled SOI values that matched the (unaltered) 20 (10) warmest and 20 (10) coldest years were tested anew for each gridbox (using the two-sided t test). The number of locally significant SOI t-test scores at the 95% level was recorded with each Monte Carlo run and tabulated for all 1000 runs in order to determine the 95% and 99% field significance thresholds.

Fig. 1.
Fig. 1.

Idealized (a) normal distribution and (b) shifted normal distribution. An upward shift in the mean of 0.5 σ translates into a doubled risk of warm extremes (shaded above 1 σ), compared to the unshifted case, and a halved risk of cold extremes (shaded below −1 σ).

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3255:STCEOT>2.0.CO;2

Fig. 2.
Fig. 2.

Significant t-test scores of the SON SOI (leading by one season) in relation to the (a) 20 warmest and (b) 20 coldest winters at each climate division within the Dec–Feb 1895/96–1994/95 period. Cross-hatching at light and heavy densities denotes the 95% and 99% significance levels, respectively, for positive SOI (El Niño conditions); light and heavy gray shading refers to the 95% and 99% significance levels for negative SOI (La Niña).

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3255:STCEOT>2.0.CO;2

Fig. 3.
Fig. 3.

As in Fig. 2, except with the DJF SOI leading by one season in relation to the (a) 20 warmest and (b) 20 coldest springs at each climate division.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3255:STCEOT>2.0.CO;2

Fig. 4.
Fig. 4.

As in Fig. 2, except for t-test scores of the JJA SOI leading by one season in relation to the 20 coldest autumns at each climate division.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3255:STCEOT>2.0.CO;2

Fig. 5.
Fig. 5.

Significantly increased risks of climate division winter temperature extremes for the SON SOI leading by one season during (a) El Niño and (b) La Niña conditions within the DJF 1895/96–1994/95 period. Relative risks at or above the 175% and 200% levels of either very cold or very warm extremes are shown here, corresponding to 93.6% and 98.2% significance levels.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3255:STCEOT>2.0.CO;2

Fig. 6.
Fig. 6.

As in Fig. 5, except for significantly increased risks of U.S. spring temperature extremes in relation to preceding DJF ENSO conditions.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3255:STCEOT>2.0.CO;2

Fig. 7.
Fig. 7.

As in Fig. 5, except for significantly increased risks of U.S. summer temperature extremes in relation to preceding MAM SOI La Niña conditions.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3255:STCEOT>2.0.CO;2

Fig. 8.
Fig. 8.

As in Fig. 5, except for significantly increased risks of U.S. autumn temperature extremes in relation to preceding JJA SOI El Niño conditions.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3255:STCEOT>2.0.CO;2

Fig. 9.
Fig. 9.

As in Fig. 5, except for significantly suppressed risks of U.S. winter temperature extremes in relation to the preceding SON SOI during (a) El Niño and (b) La Niña conditions. Risk reductions by 75% and 100% for either very cold or very warm extremes are shown here and represent the 95% and 99.3% significance levels.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3255:STCEOT>2.0.CO;2

Fig. 10.
Fig. 10.

As in Fig. 5, except for significantly suppressed risks of U.S. spring temperature extremes in relation to the preceding DJF SOI during (a) El Niño and (b) La Niña conditions.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3255:STCEOT>2.0.CO;2

Fig. 11.
Fig. 11.

Conditional probabilities of extreme quintile temperature categories in winter for Louisiana (average of all climate divisions) in relation to (a) El Niño and (b) La Niña during the period 1896–1995, for three-season to zero lead times. Quintile values refer to 20 out of 100 possible values. The unconditional probability of occurence (20%) is indicated by a solid line. Significant positive or negative departures from the expected probability (at the 93.6% level or higher) are denoted by an asterisk.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3255:STCEOT>2.0.CO;2

Fig. 12.
Fig. 12.

As in Fig. 11, except for Texas spring temperatures.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3255:STCEOT>2.0.CO;2

Fig. 13.
Fig. 13.

As in Fig. 11, except for Washington spring temperatures.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3255:STCEOT>2.0.CO;2

Table 1.

Classification of seasonal SOI values centered on Jan, Apr, Jul, and Oct for 1896–1995. Top 20 cases qualify for El Niño/La Niña in the full record, and 10 cases in the split sample record. The first digit of the code denotes the ENSO phase during the full record (W = El Niño, C = La Niña, N = “neutral”), while the second digit denotes the same for the 50-yr subsets 1896–1945 and 1946–95, respectively.

Table 1.
Table 2.

Probabilities (%) for sampling without replacement: (a) 20 samples from 20 extreme vs 80 other cases and (b) 10 samples from 10 extreme vs 40 other cases. Shown here are the individual odds P of sampling 0, 1, 2, . . ., n extreme cases as well as the threshold odds 1 − ΣP for sampling ≥n extreme cases.

Table 2.
Table 3.

Areal coverage (%) of the contiguous United States for which two-sided t-test scores of the SOI are significantly different from zero (high or low; at the 95% level), for SOI leading U.S. temperatures by up to three seasons: (a) DJF, (b) MAM, (c) JJA, (d) SON. The first entry refers to the full 100-yr base period (1896–1995), the second (third) entry refers to the early (late) 50-yr subset period. Field significance at the 95% and 99% levels is indicated by one and two asterisks, respectively.

Table 3.
Table 4.

Areal coverage (%) of the contiguous United States for which significantly increased risks of climate division seasonal extremes (warm or cold) occur in relation to preceding ENSO conditions, for SOI leading U.S. temperatures by up to three seasons: (a) DJF, (b) MAM, (c) JJA, (d) SON. Entries refer to the full 100-yr base period (1896–1995), at the one-sided 93.6% significance level (see Table 2).

Table 4.
Table 5.

Areal coverage (%) of the contiguous United States for which significantly suppressed risks of climate division seasonal extremes occur in relation to preceding ENSO conditions, for SOI leading U.S. temperatures by up to three seasons: (a) DJF, (b) MAM, (c) JJA, (d) SON. Entries refer to the full 100-yr base period (1896–1995), at the one-sided 95.0% significance level (see Table 2).

Table 5.
Save