Abstract

Daily sounding-derived atmospheric stability indices are typically employed for short-term severe weather forecasts. Over longer time periods, these indicators may convey changes in the potential for severe storm development over the United States. Daily (0000 UTC) observations from 48 radiosonde stations in the contiguous United States are extracted to assemble a ∼50 yr record of four common stability indices: the Lifted Index, the K-Index, convective available potential energy (CAPE), and the Air Force Severe Weather Threat Index. Because of radiosonde data inhomogeneities, the 1973–97 period is the focus of the analysis. Trends in the mean and extreme values of daily index observations are calculated for spring and summer seasons. In addition, climatological mean indices, as well as the mean frequency of index extremes, are determined for all U.S. regions. At stations free of obvious data discontinuities, the early part of the record (1948–65) is compared with more recent periods.

In spring, few significant changes in either index means or extremes are reported. However, in summer, widespread trends toward enhanced instability appear, particularly in the Lifted Index, the K-Index, and CAPE. Stations in the Plains and southern states show the most consistent increases in severe weather potential, while significant trends in index extremes are uncommon in the western region, possibly due to the rarity of severe weather there.

1. Introduction

Severe thunderstorms and their associated hazards, such as tornadoes, hail, lightning, and heavy rain, threaten lives and property in all regions of the United States, often inflicting losses in excess of $1 million per storm event (Changnon 2001a). Changes in climate may alter the statistical ensemble of storm-related weather ingredients, which may influence the temporal and spatial distribution of U.S. thunderstorms. For example, an increased environmental lapse rate, produced by surface warming, cooling of the mid to upper troposphere, or recurrent upper-level troughing could increase regional storm frequency. Given the damage potential of severe local storms, as well as concern about a climate-related increase in extreme weather events (Karl et al. 1999; Meehl et al. 2000; Houghton et al. 2001), an exploration of severe storm climatologies is overdue. However, temporal statistical analyses of twentieth-century thunderstorms and tornadoes have been limited by notorious data biases. It is difficult for weather stations to reliably document these highly localized phenomena, despite the impressive coverage of surface stations in the United States. Because many storm data archives are compiled from visual reports or damage assessments, time series of storm frequency likely do not represent changes in the physical climate. A number of factors can influence reporting and skew the storm record (Schaefer and Galway 1982; Kelly et al. 1985; Grazulis 1993; Easterling et al. 2000), including population density, urbanization, highway distribution, visibility problems, and observer perceptions. Distinguishing real climatological variations from these statistical artifacts remains a challenge. One way to estimate climate-driven changes in storms is to avoid the storm data altogether and examine larger-scale atmospheric conditions that promote storm genesis.

Atmospheric stability indices, derived from daily radiosonde observations, measure the potential for severe weather development and are considered representative of the synoptic-scale environment (Davis et al. 1997). Meteorologists commonly employ these indices to determine where conditions favor storm formation and where severe thunderstorm and tornado “watches” should be issued. In this study, stability indices will be transformed into climatological variables to assess possible changes in severe weather potential. Next to elusive storm-scale processes (such as the microscale winds that may induce tornadogenesis), these indices are the meteorological variables most closely associated with severe storm development. They have not previously been used for temporal climatological studies for the United States.

A stability index is designed to measure the ease with which an air parcel will rise through the atmosphere, using the parcel-to-environment temperature difference, a comparison of surface and upper-air temperatures, or the moisture content of the boundary layer (i.e., the moisture content of the air parcel). Those designed to forecast tornadoes also include parameters for wind shear that can induce rotation. This paper examines trends in four common stability indices: the Lifted Index (LI), the K-Index (K), the Air Force Severe Weather Threat (SWEAT) Index, and convective available potential energy (CAPE). These four indices primarily diagnose thermodynamic instability, though the SWEAT index also incorporates components for wind veering and upper-level wind speed. Trends are calculated at radiosonde stations across the country, for the 1973–97 period. Because most U.S. thunderstorms occur during the warmer months, this analysis is limited to the spring and summer seasons.

2. Data

The data source for this study is the Radiosonde Data of North America archive of daily soundings produced by the Forecast Systems Laboratory (FSL) and the National Climatic Data Center (NCDC), for the 1948–97 period. Forty-eight U.S. stations with the most complete records were selected (Fig. 1), and data were extracted for two 92-day seasons: spring (March through May) and summer (June through August). Observations are formally taken twice daily at 1200 and 0000 UTC. Only the 0000 UTC soundings were retained because they represent late-afternoon to earlyevening local times across the continental United States and most closely correspond to the peak in afternoon heating that can initiate thunderstorms.

Fig. 1.

Selected U.S. radiosonde stations.

Fig. 1.

Selected U.S. radiosonde stations.

a. Selected stability indices (T = temperature; Td = dewpoint; DD = dewpoint depression; υ = horizontal wind velocity; numerical subscripts denote atmospheric pressure surfaces)

Because a comprehensive stability index dataset does not currently exist, the four selected indices were calculated from the radiosonde archive according to the formulas detailed below. Established index thresholds [many derived from Miller (1972) and summarized in Gordon and Albert (2000)], which forecasters use as a guide to estimate thunderstorm and tornado probabilities, are also listed with each index formula. These thresholds classify index values according to weather conditions of graded severity.

1) Lifted Index

The LI (Galway 1956) expresses the temperature difference between a lifted parcel and the surrounding air at 500 mb. The parcel is lifted dry adiabatically from the surface to the lifted condensation level, and then wet adiabatically to 500 mb. In this study, the parcel assumes the average temperature and humidity of the lowest 100 mb of the sounding. The LI is negative when the parcel is warmer than its environment.

  • LI = T (500-mb environment) − T (500-mb parcel) (in °C)

  • LI:

  • 0° to 3°C stable

  • 0° to −3° marginally unstable

  • −3° to –6° moderately unstable

  • −6° to −9° very unstable

  • < −9° extremely unstable

2) K-Index

The K-Index (George 1960) serves as a predictor for thunderstorms that produce heavy rain and possibly flash flooding. The index is high, and thus intense rainfall is likely, when there is abundant moisture through midlevels, as well as a strong lapse rate. Note that when the dewpoint depression (DD) at 700 mb is small, the index increases.

  • K = Td850 + (T850T500) − DD700

  • K:

  • < 15 0% probability of thunderstorms

  • 15–20 20%

  • 21–25 20–40%

  • 26–30 40%–60%

  • 31–35 60%–80%

  • 36–40 80%–90%

  • >40 near 100%

3) Severe Weather Threat Index

This index (Miller 1972) is designed to distinguish between severe and nonsevere storms, because it includes variables for wind speed and vertical wind shear, in addition to thermodynamic variables such as lapse rate and low-level moisture. In this case, the wind shear component measures wind veering between 850 and 500 mb, which may signal warm air advection near the surface, help to sustain the storm by separating the downdraft from the updraft, and help to produce storm rotation:

  • SWEAT = Td850 + 20 (Totals − 49) + 2υ850 + υ500 + 125 (SHEAR + 0.2), where

  • Totals = (Td850T500) + (T850T500) and

  • SHEAR = sin[500-mb wind direction (deg) − 850-mb wind direction].

  • SWEAT:

  • > 300 potential for severe thunderstorms

  • > 400 potential for tornadoes.

4) Convective available potential energy

CAPE is a vertical integration of the parcel-to-environment temperature difference between the levels where the parcel freely rises [the level of free convection (LFC) to the level of neutral buoyancy (LNB)]. It expresses the energy a parcel will have when lifted (in J kg−1) and indicates the potential strength of updrafts within a thunderstorm (Bluestein 1993). High CAPE requires both high surface moisture and temperature and often requires a steep environmental lapse rate to maintain instability with height.

  • CAPE = gLNBLFC [Tp(z) − Te(z)/Te(z)], where

  • Tp(z) = the temperature profile of an air parcel ascending moist adiabatically (parcel assumes conditions of the lowest 500 m),

  • Te(z) = the temperature profile of the environment, and

  • g = acceleration due to gravity.

  • CAPE:

  • = 0 J kg−1 stable

  • = 1000–2500 moderately unstable

  • > 2500 very unstable; strong convection likely.

5) Addtional notes

The surface pressure at high-elevation stations can be lower than the 850-mb level required by the K and SWEAT index formulas. In such cases, the observation closest to the 850-mb level was substituted. In the calculation of CAPE, the virtual temperature correction was not applied and is not expected to affect the regional pattern of index means and trends reported in section 4. A random sample of stations, one from each region, indicated that the temperature correction would uniformly increase larger values of CAPE (those between 1000 and 2000 J kg−1) by 11.1% ± 2.3%.

b. Underlying changes in temperature and humidity

Temperature and humidity are the most important influences on the four indices chosen for this study. A single sounding (Lake Charles, Louisiana, at 0000 UTC on 31 May 1973) demonstrates how low-, mid-, and upper-level temperature and moisture perturbations alter atmospheric stability, as measured by each index. The original sounding, recorded around 6:00 p.m. local time, signals relatively stable conditions: a very warm but dry near-surface layer, a moderate lapse rate, and a large dewpoint depression throughout the column. Index calculations are shown in Table 1, both for the observed sounding and after altering temperature and dewpoint at selected heights.

Table 1a. Stability indices calculated for 0000 UTC 31 May 1973 at Lake Charles, LA. Temperature and dewpoint were adjusted by 1° and 10°C at various levels.

Table 1a. Stability indices calculated for 0000 UTC 31 May 1973 at Lake Charles, LA. Temperature and dewpoint were adjusted by 1° and 10°C at various levels.
Table 1a. Stability indices calculated for 0000 UTC 31 May 1973 at Lake Charles, LA. Temperature and dewpoint were adjusted by 1° and 10°C at various levels.

Adjusting single or even multiple levels by just 1°C has a minimal effect on overall stability, though such changes can help to explain the direction of long-term trends in stability indices. On the other hand, variations of 10°C have a dramatic effect on the indices. For example, increasing parcel (lowest 100 mb) temperature and dewpoint by 10°C shifts CAPE from 0 to 5885.04 J kg−1 and the LI from 2.99° to –10.96°C. Decreasing upper-level temperature compounds the effect.

CAPE and LI are tightly correlated (Table 1b) because both rely on a lifted parcel and upper-level temperatures, though CAPE is more dependent on temperature gradations throughout the air column. It is clear that altering the characteristics of the lifted parcel directly affects stability. The K and SWEAT indices are also tightly correlated because both formulas incorporate lapse rate (between 850 and 500 mb) and midlevel moisture, while neither includes surface measurements. For example, note that a 10°C decrease in upper-level temperature forces both indices into the “unstable” range. Increasing midlevel dewpoint also increases K and SWEAT.

Table 1b. Correlations among indices, based on Table 1a.

Table 1b. Correlations among indices, based on Table 1a.
Table 1b. Correlations among indices, based on Table 1a.

In general, Table 1 confirms that changes in stability indices strongly depend on changes in temperature and humidity at specific atmospheric pressure surfaces. All of the resulting shifts follow logically from the index formulas. Positive trends (i.e., toward enhanced instability) in the LI and CAPE are likely due to increases in near-surface temperature and humidity, or a decline in upper-level temperature. Positive trends in K and SWEAT may be forced by increases in midlevel temperature and humidity, or a decrease in upper-level temperature.

c. Radiosonde data inhomogeneities

The FSL radiosonde archive has been subjected to extensive quality control analyses for obvious errors and hydrostatic inconsistencies. However, inhomogeneities induced by changes in instrumentation, observation time, and recording procedures remain in the data and have been documented by other investigators (Elliott and Gaffen 1991; Gaffen 1993, 1994). Because temporal discontinuities related to changes in radiosonde observing methods can be misinterpreted as climate changes, four prominent dates with major historical shifts in these practices were selected for their possible influence on the long-term stability index record. These change points are described in Table 2 and include a change in observation time (1957), the introduction of a modified temperature sensor (1960), the initial reporting of very low relative humidities (1965), and the use of a faulty humidity gauge (1965–72).

Table 2.

Nationwide changes in radiosonde observing practices that may affect the stability index record.

Nationwide changes in radiosonde observing practices that may affect the stability index record.
Nationwide changes in radiosonde observing practices that may affect the stability index record.

While the LI and CAPE would be most sensitive to the temperature-related changes in the thermistor and observation time, all four indices could register artificially stable conditions during the 1965–72 “low humidity bias” interval. Moreover, a large proportion of missing humidity observations before 1965 could render that part of the record useless at several stations.

The effect of possible data inhomogeneities on the archive of daily stability indices was investigated. First, daily index values were tested for discrete shifts between the 5-yr intervals before and after the suspected change-point years (1957, 1960, 1965, and 1972), using the nonparametric Wilcoxon rank–sum test for difference in medians (Lehmann 1975; Wilks 1995). Results of this procedure are shown in Table 3. Though only a few stations show changes at 1957 and 1960, several stations show clear disruptions during the 1965–72 interval. The low humidity bias resulted in an artificial decrease in CAPE, K, and SWEAT and an increase in LI—all changes toward greater stability—before the defective humidity device was improved in 1972–73. Second, the percentage of missing values before and after 1965 was calculated for all four indices and averaged over all stations. This procedure confirmed that the K and SWEAT records also contain an unacceptable proportion of missing low humidity observations before 1965 (Table 4), with nearly one-third of possible daily observations omitted at many stations. The drier, more stable days are therefore “filtered out” of the earlier period, confounding inferences about climate change over the full record. This effect is illustrated in Fig. 2, which shows the number of missing spring K-Index observations (1948–97) at Lake Charles, a representative station.

Table 3.

Number of stations showing significant discontinuities in stability index records, before and after changes in radiosonde observing practices. (Does not include stations with records beginning after 1960)

Number of stations showing significant discontinuities in stability index records, before and after changes in radiosonde observing practices. (Does not include stations with records beginning after 1960)
Number of stations showing significant discontinuities in stability index records, before and after changes in radiosonde observing practices. (Does not include stations with records beginning after 1960)
Table 4.

Missing daily observations per season, averaged over all stations. Interstation variation is also shown.

Missing daily observations per season, averaged over all stations. Interstation variation is also shown.
Missing daily observations per season, averaged over all stations. Interstation variation is also shown.
Fig. 2.

Number of missing spring K-Index observations (of the 92 spring days) at Lake Charles, LA.

Fig. 2.

Number of missing spring K-Index observations (of the 92 spring days) at Lake Charles, LA.

In short, up to two-thirds of stations show discontinuities related to changes in humidity observing practices; there are no significant problems arising from the thermistor and time of observation changes. CAPE and LI are affected only by the low humidity bias in 1965–72, while SWEAT and K have the additional problem of missing data before1965. Because the radiosonde humidity record is especially problematic before 1973 (Elliott and Gaffen 1991), stability index trends are derived only for the 1973–97 time interval. The earlier (1948–65) period is incorporated into the study only for stations free of known underlying data problems.

3. Methods

Temporal changes in atmospheric stability were explored in two ways. To begin, “raw” daily index observations at each station were averaged over each season and region (Fig. 1), assembled into annual time series, and evaluated for significant trends. However, because such aggregations tend to dilute the meaning of each index (stations record a stable atmosphere on most days), they convey little about the frequency of severe weather. Stability index extremes, therefore, are the primary focus of this study. The daily index data were reduced to time series expressing the number of days that severe weather thresholds were exceeded per season. The resulting time series, which express the annual frequency of these extremes, were compiled separately for each index, season, and station.

a. Selection of appropriate threshold levels

Although the index thresholds described in section 2 are considered “common” in operational forecasting, there is a limited body of research to support these precise boundaries between severe and nonsevere conditions. However, it is a valid assumption that as atmospheric instability increases, so does the likelihood of severe weather. To define index extremes here, thresholds have been selected based on a compromise between the established guidelines (EGs) and previous studies classifying stability indices according to severe storm incidence.

Four threshold levels were chosen because they indicate a high potential for severe weather. Lifted Index values below –6°C suggest a “very unstable” atmosphere (EG) and were associated with 75% of the tornadoes examined by Johns et al. (1993). While moderate to strong convection occurs when CAPE falls between 1000 and 3000 J kg−1 (Bluestein 1993), a threshold of 1800 was selected because all storms with CAPE above ∼1820 were severe in Rasmussen and Blanchard (1998). A K-Index greater than 35 indicates a very high probability (80%–90%) of thunderstorms with heavy rains and possibly flash flooding (EG; Giordano 1994). A SWEAT threshold of 300 was chosen because Miller (1972) identified this value with severe thunderstorms and because David (1976) found that average the SWEAT Index associated with warm season tornadoes usually fell between 200 and 400.

b. Trend estimation and significance testing

Linear trends for each threshold series are estimated using the “median of pairwise slopes” method (Lanzante 1996; Hoaglin et al. 1983), which, unlike least squares regression, is nonparametric and resistant to outliers. The method requires calculating the slopes of lines joining all possible pairs of points in the time series; the median of those slopes then defines the trend magnitude. The trends are tested using another measure resistant to outliers, a two-tailed Spearman rank–order correlation coefficient, based on the linear association of the ranks of the two variables (the data series and the time vector). Gaffen and Ross (1999) also employed these methods in their study of recent trends in U.S. surface humidity and temperature.

4. Results

a. Raw index values: Regional means and trends

Regionally and seasonally averaged stability indices (Table 5, column A) show expected patterns: they confirm a more unstable atmosphere in summer and imply that severe weather development is more likely during the warmer season. For example, across all regions, the mean Lifted Index shifts downward from spring (4.2 to 10.8) to summer (−1.1 to 2.9); recall that instability increases as the LI decreases. Furthermore, the mean indices emphasize the thunderstorm-prone regions of the south and plains, which show the lowest mean LI and the highest mean CAPE, K, and SWEAT in nearly every case. One exception is the western region in spring, where frequent upper-air troughing and moderate surface moisture produce a lower LI and higher K than is recorded in other parts of the country.

Table 5.

Analysis of daily indices, by season. Column A shows the mean observed daily index for the period of record (1973–97), with the interannual standard deviation. Column B shows the annual trend in the mean daily index. Column C shows Spearman's Rho.

Analysis of daily indices, by season. Column A shows the mean observed daily index for the period of record (1973–97), with the interannual standard deviation. Column B shows the annual trend in the mean daily index. Column C shows Spearman's Rho.
Analysis of daily indices, by season. Column A shows the mean observed daily index for the period of record (1973–97), with the interannual standard deviation. Column B shows the annual trend in the mean daily index. Column C shows Spearman's Rho.

Trend analyses performed on regional means (Table 5, column B) also reveal consistent patterns. Spearman's Rho (Table 5, column C), the rank correlation coefficient on which the trend tests were based, gives an indication of the strength of the trend. All significant trends are toward greater instability (i.e., the LI decreases, the other three indices increase) and emerge mainly in the summer season. The most widespread trends appear in the LI and K, where increased instability is suggested in all but one region (the Northeast LI) during summer. Both the South and Plains regions show an increased CAPE in summer. In the western half of the country (Midwest, Plains, and West), a rise in the mean SWEAT Index also appears in the summer season.

Regional and temporal aggregations of daily index data reflect temperature and humidity characteristics within atmospheric layers over large geographic areas. However, because the indices were designed to forecast convective storms on a single day, trends in these broad averages should not be interpreted as a change in the frequency of severe weather. For example, while a moderately to strongly negative LI is considered meaningful for daily thunderstorm forecasts, the LI is positive on most days. The seasonal frequency of stability index extremes, therefore, will be the focus of this investigation.

b. Frequency of index extremes

The number of “high instability” days recorded per season, based on the severe weather thresholds established above, provides a more useful measure of severe weather frequency. Clearly, this unit does not identify individual storm cells but is analogous to measures such as “thunder days” or “tornado days.” Because an extreme value of a stability index does not guarantee storm development, the trends presented in this section are best described as trends in “days of high severe weather potential.”

1) Climatological frequency of index extremes

It is useful to understand how severe weather potential varies among U.S. regions, by examining the mean frequency of index extremes for the 1973–97 period. The climatological charts (Fig. 3) depict the mean number of days on which stability index thresholds were exceeded. Each regional value represents an average over all stations in the region; the 1-σ error bar conveys interstation variability. The South and Plains regions experience high instability days most frequently, followed by the Midwest and then the Northeast. The West is least likely to experience severe weather in spring and summer.

Fig. 3.

Average number of days in which stability index thresholds are exceeded in spring (white bars) and summer (black bars). Results are shown for (a) LI < −6°C, (b) CAPE > 1800 J kg−1, (c) K > 35, and (d) SWEAT > 300.

Fig. 3.

Average number of days in which stability index thresholds are exceeded in spring (white bars) and summer (black bars). Results are shown for (a) LI < −6°C, (b) CAPE > 1800 J kg−1, (c) K > 35, and (d) SWEAT > 300.

For the Lifted Index (Fig. 3a), days below −6 (°C) are relatively rare, occurring on an average of only 2 spring days and 6 summer days in the South and Plains. Nonetheless, the dominant season (summer) and regions (South and Plains) indicate the role that high surface temperature and moisture play in reducing values of the LI.

Days in which CAPE exceeds 1800 J kg−1 (Fig. 3b) are clearly most frequent in the South in summer, where surface temperature and humidity tend to be highest. The typically high surface specific humidity and resulting low lifted condensation levels in this region explain why the moist adiabatic lapse rate can remain smaller than the environmental lapse rate, maintaining instability through the vertical.

High values of the K Index are far more common in summer than in spring because K is sensitive to high tropospheric water vapor. The frequency of K > 35 (Fig. 3c) is highest in the plains, specifically at the four Texas stations, because of moisture advection from the Gulf of Mexico via the low-level jet.

Of the four indices, SWEAT best reflects the position of the jet stream and the convergence of contrasting air masses, because it measures upper-level wind speeds and wind shear and compares surface and upper-air temperatures. It is therefore not surprising that the South and Plains regions show the highest frequency of SWEAT > 300 (Fig. 3d) in spring, while the Plains, Midwest and Northeast show the highest frequency in summer, given the northward migration of the jet stream through the warm season.

These climatological means are necessary background for understanding the trends in severe weather potential that are presented in the following section. Regions where atmospheric conditions rarely favor the development of severe storms will likely not show an obvious increase or decrease in such conditions. For example, the observed LI at the Salem, Oregon, station never crossed the –6°C threshold, with the exception of single days during the summers of 1986, 1990, and 1991. These three days did not constitute a trend. Discernible trends are most likely where high values of the indices are common.

2) Trends in index extremes

Figures 4 and 5 depict trends in the number of days in which stability index thresholds were exceeded for the 1973–97 period. Black triangles symbolize positive trends; open circles show negative trends. Significance (at the 0.01 and 0.05 levels) is indicated by the size of the symbols. Because these results convey trends in the number of “unstable days” rather than in raw index values, all positive values express a tendency toward more frequent instability.

Fig. 4.

Trends (1973–97) in the number of days that index thresholds were exceeded: (a) LI in spring, (b) LI in summer, (c) CAPE in spring, and (d) CAPE in summer. Positive trends (I) are black triangles; negative trends (D) are open circles. Significance (at the 0.01 and 0.05 levels) is indicated by the size of the trend symbols. Gray crosses represent a trend of zero.

Fig. 4.

Trends (1973–97) in the number of days that index thresholds were exceeded: (a) LI in spring, (b) LI in summer, (c) CAPE in spring, and (d) CAPE in summer. Positive trends (I) are black triangles; negative trends (D) are open circles. Significance (at the 0.01 and 0.05 levels) is indicated by the size of the trend symbols. Gray crosses represent a trend of zero.

Fig. 5.

Same as in Fig. 4, but for (a) K in spring, (b) K in summer, (c) SWEAT in spring, and (d) SWEAT in summer.

Fig. 5.

Same as in Fig. 4, but for (a) K in spring, (b) K in summer, (c) SWEAT in spring, and (d) SWEAT in summer.

In general, far more stations show trends in summer because severe weather is common over a larger part of the country in the warmer season. Where trends exist in either season, they are predominantly positive and occur east of the Rockies. The western and northern tier states show the fewest significant changes, possibly because index thresholds are rarely exceeded in these regions.

There are, however, some distinctions among specific indices, in terms of the magnitude and regional pattern of the trends. The Lifted Index shows no clear regional pattern in spring (Fig. 4a), although significant positive trends (in the number of days that the LI falls below –6) are limited to stations in the South and the southern Plains. In summer (Fig. 4b), significant upward trends appear across the eastern two-thirds of the country. All of the changes in the LI are positive, with the exception of the Little Rock, Arkansas, station in summer. Given the dependence of a strongly negative LI on high surface temperature and humidity, it is not unexpected that changes (in either direction) would be limited to warmer parts of the country during spring and that these signals expand across the country in summer. However, it is notable that most of the trends are positive.

Significant trends in the number of days with CAPE greater than 1800 J kg−1 are scarce in spring (Fig. 4c) and are found primarily along the Gulf Coast. This pattern is similar to that seen in the LI, though most changes are restricted to the extreme South, where thermodynamics are the major determinant of severe weather. In summer (Fig. 4d), significant positive trends in CAPE extend from the southern Plains to the Northeast, with two western stations also showing increases (Salt Lake City, Utah, and Denver, Colorado). These trends may result from a more frequent strengthening of the lapse rate (from the advection of cooler air aloft) or an increase in very warm, humid days.

Increases in the number of spring days that the K Index exceeds 35 (Fig. 5a) occur in the country's agricultural interior and in the southern Plains, while negative trends appear in the mid-Atlantic region. In summer (Fig. 5b), positive trends appear at nearly every station. Note that significant trends are again clustered east of the Rockies and are absent in the western and northeastern states. These increases in K suggest changes over broad layers of the atmosphere, including widespread increases in lower-tropospheric moisture and/or enhanced lapse rates (the apparent explanation at dry western stations), at some central U.S. stations in spring and across the country in summer.

The SWEAT Index shows the fewest significant changes (in the number of days exceeding 300). In spring, these are primarily negative throughout the eastern third of the country (Fig. 5c), suggesting a recent decline in the potential for tornadogenesis here. This pattern departs from spring patterns seen in the other three indices, in which southern stations show increases. Positive summer trends (Fig. 5d) are apparent in the classic “tornado alley” region (where wind shear is a dominant factor in severe weather development) extending east through the Great Lakes (near the climatological mean position of the jet stream in summer). Increases also appear along the Carolina coasts and in the Rocky Mountain states. While this pattern is comparable with the other index maps for summer, positive trends in SWEAT are absent along the U.S. Gulf Coast.

c. The western region: Reduced index thresholds

The absence of trends in the number of extremely unstable days in the western region may simply reflect a rarity of severe thunderstorms—the small-scale, thermodynamically intense thunderstorms common east of the Rockies. In this section, the index thresholds are lowered for the western stations, which could reveal changes in atmospheric stability relevant to nonsevere weather. Figure 6 shows 1973–97 trends in the number of days exceeding new thresholds for three indices: the LI < −3°C, CAPE > 1000 J kg−1, K > 25, and SWEAT > 200. Significant increases, in both seasons, seem to be concentrated in Rocky Mountain states and eastern Washington, possibly indicating enhanced precipitation or storm activity. The overall effect of lowering the thresholds is to increase the probability that an extreme day will be recorded, and to reduce the number of years with “zero” extreme days.

Fig. 6.

Same as in Fig. 4, but for trends (1973–97) in index extremes in the western region. (a) LI < −3 in spring, (b) LI < −3 in summer, (c) CAPE > 1000 in spring, (d) CAPE > 1000 in summer, (e) K > 25 in spring, (f) K > 25 in summer, (g) SWEAT > 200 in spring, and (h) SWEAT > 200 in summer.

Fig. 6.

Same as in Fig. 4, but for trends (1973–97) in index extremes in the western region. (a) LI < −3 in spring, (b) LI < −3 in summer, (c) CAPE > 1000 in spring, (d) CAPE > 1000 in summer, (e) K > 25 in spring, (f) K > 25 in summer, (g) SWEAT > 200 in spring, and (h) SWEAT > 200 in summer.

Finally, it should be noted that both upper and lower index extremes were tested for the fall and winter seasons at the four West Coast stations—no changes in these variables were found. While it is possible that these parameters are less sensitive to the large-scale winter rainstorms that typically strike the West Coast, hail and tornadoes can be embedded within these larger storm systems, particularly in southern California (Hales 1985). Therefore, stability indices can be a reliable weather analysis tool in the western region (Braun and Monteverdi 1991).

d. Incorporating early data: 1948–65

The 25-yr dataset used to derive index trends provides a narrow view of climate changes related to atmospheric stability; radiosonde stations with full records dating back to 1948 would help to enhance that picture. However, because the humidity bias years (1965–72) are unusable, comparing the earliest period (1948–65) with later periods of the same length (1973–89 and 1981–97) allows the radiosonde record to be expanded for this study.

Differences in index extremes between the earlier and later periods are computed and then tested using a Wilcoxon rank sum test. Because this method examines only raw differences between periods, it is more sensitive to temporal changes than the trend estimation method. Twenty suitable stations, distributed uniformly across all regions, exist for a longer-term analysis of the LI and CAPE. The K and SWEAT indices are examined only at the few stations (8 and 11, respectively) not corrupted by a high number of missing humidity observations before 1965 (less than an average of 10 missing days for the 1948–65 period). Results are presented only for summer because a sufficient number of valid stations could not be found for spring.

Stacked bar graphs depict the proportion of stations showing increases and decreases in index extremes between the early (1948–65) period and the two later periods, 1973–89 (period 1) and 1981–97 (period 2). Figure 7a presents shifts in index means, while Fig. 7b shows shifts in the number of extreme days.

Fig. 7.

Changes in index (a) means and (b) extremes between the 1948–65 period and two later periods: 1) 1973–89 and 2) 1981–97. Significant increases (I*) are shown in black, increases (I) in gray, decreases (D) in white, and significant decreases (D*) in patterned fill. Results are presented as a percentage of the stations examined, for the summer season only.

Fig. 7.

Changes in index (a) means and (b) extremes between the 1948–65 period and two later periods: 1) 1973–89 and 2) 1981–97. Significant increases (I*) are shown in black, increases (I) in gray, decreases (D) in white, and significant decreases (D*) in patterned fill. Results are presented as a percentage of the stations examined, for the summer season only.

Decreases (D) in mean LI, in most cases significant at α = 0.01, occur at nearly every station examined. These are accompanied by increases (I) in the number of days in which the LI falls below –6. This shift toward enhanced atmospheric instability, as measured by the LI, is particularly stark for the 1981–97 period. While there is variation among stations, increases in CAPE (mean and extremes) also outnumber decreases. Consistent with the recent trends in the previous section, significant increases in CAPE are most common at the southern stations in summer, while decreases appear at some stations in the Northeast and Plains regions (not shown). Changes in both the K and SWEAT indices, for the few stations examined, also suggest enhanced instability from the early to the later periods. The changes are not as uniform or widespread given the data limitations.

Annual century-long temperature trends (Houghton et al. 2001; Jones et al. 2001) reveal a modest midcentury cooling over the United States from the mid-1940s through the mid-1970s, followed by rapid warming after 1975. Stability index data encompassing these time intervals show an overall shift toward increased instability, in general agreement with the temperature patterns.

e. Summary of results

In summary, within each season, the spatial distribution of positive and negative trends in stability index extremes is fairly consistent among the four indices. (Henceforth, all trends toward greater instability will be described as “positive” or “increasing,” including a decline in the mean Lifted Index.) In spring, stations in the northern and western parts of the country show little change or a decline in the potential for severe weather, while the stations in the deep south show increases. In summer, positive trends in index extremes are widespread, with the exception of New England and the West Coast.

The overall pattern of change conveyed by the four indices is one of enhanced, or more frequent, atmospheric instability. Regional changes in raw mean index values are largely mirrored by the spatial distribution of trends in index extremes. It appears that the number of days with a high severe weather potential has risen, particularly in the eastern two-thirds of the country during summer.

5. Discussion

Among the four indices, spatial relationships in the trend patterns can be discerned, given that some of the measured parameters overlap. For example, nearly identical LI and CAPE patterns emerge because both rely heavily on air parcel characteristics; increases in surface temperature and humidity may play an important role in these trends. Summer trends in the K-Index are similar to those of the LI and CAPE, but extend farther westward, possibly reflecting increased surface temperatures (850 mb is very close to surface level at mountainous western stations) or increased tropospheric moisture. The K and SWEAT patterns would be expected to coincide, but they deviate in the southeastern United States, perhaps as a result of changes in the wind parameters that are included in the SWEAT formula.

a. Related climate changes

Stability index trends can be compared with documented changes in related climate variables within the United States to assess the consistency of the patterns identified here. For example, 1973–95 trends in precipitable water (Ross and Elliott 2001), also derived from radiosonde data, show a regional pattern similar to that of the K-Index: positive trends limited to the eastern half of the United States in spring, but widespread in summer. This correspondence is not surprising given that the two parameters incorporate humidity measurements within proximal atmospheric levels (precipitable water is measured from surface to 500 mb; the K uses humidity at 700 and 850 mb). Furthermore, recent increases (1961–95) in surface humidity and dewpoint (Gaffen and Ross 1999) seem to mirror the regional pattern of increases in the LI, CAPE, and to a lesser extent K (both the index means and extremes) in summer. A comparison of mean index trends to seasonal average surface temperature trends for the 1976–2000 period (Jones et al. 2001) reveals only weak correspondence. Positive temperature trends seem to be largest in the southwest in both seasons (in contrast to the index trends), although a modest spring and summer temperature rise was reported in parts of the southern and eastern United States. However, widespread increases in high temperature extremes since 1960 (DeGaetano and Allen 2002) may help to explain the increased frequency of stability index extremes in the eastern United States.

It should be noted that there is a clear disconnect between the stability indices and some climate variables in the southern Plains. This region showed significant trends in all four indices (means and extremes) but little change in surface temperature, warm extremes, and surface moisture. Conversely, while all of the studies cited above show strong climate changes in the western United States, these have not translated into an increase in severe weather potential for the 1973–97 period. However, index means (particularly those including the 1948–65 data) and nonsevere storm indicators appear to have increased in the West. Despite some regional discrepancies, stability index trends in the United States fit the overall pattern of change in related climate variables.

b. The role of atmospheric teleconnections

Changes in index extremes, and by extension thunderstorm potential, appear to be broad, consistent, and positive in the eastern two-thirds of the country. This regional pattern appears to be more indicative of the rarity of severe weather in the west and not the discreet regional signatures of atmospheric circulation regimes. Until recently, extratropical teleconnection patterns linked to the El Niño–Southern Oscillation were defined only for the cold season, when wind fields are strongest. So, it is difficult to draw parallels between known teleconnections and warm-season trends in atmospheric instability. The dominance of the Pacific–North American (PNA) pattern from 1976–88 (Trenberth and Hurrell 1994) and subsequent transition to the negative phase of the tropical Northern Hemisphere (TNH) pattern after 1988 (Robertson and Ghil 1999) would appear to coincide with some of the 1973–97 trends reported here, given the height and temperature anomalies involved (the PNA is associated with a deep trough into the southeast United States; the TNH with a strong ridge over the Northeast). For example, the 1973–97 decreases in the SWEAT Index found in the southeast United States in spring may be due to a northward displacement of the jet stream, with the transition away from the PNA pattern. Surface warming in the eastern United States, along with increased moist air advection into the Southeast, could also be associated with this transition. However, both patterns, particularly the TNH, are considered most relevant during the winter months.

In contrast, the North Pacific Index (NPI) is believed to modulate summer climate. The principal mode (warm North Pacific sea surface temperature anomaly) was recently associated with above-normal precipitation in the east/southeast United States and below-normal precipitation through the Plains and northwestern regions for the 1979–99 period (Lau et al. 2004). Still, the North American height anomalies identified with the summer NPI contradict some of the results reported here, particularly the increased instability found in the southern Plains. At this time, established circulation regimes show no apparent link with warm-season trends in atmospheric instability.

c. Stability index trends and thunderstorm climatology

A comparison of stability index trends to known thunderstorm trends is limited by the dearth of climatological research relating to severe local storms. However, some broad remarks can be made regarding existing research, given that trends in the mean index values (Table 5) and lower threshold extremes (Fig. 6) may relate to trends in nonsevere thunderstorms. Changnon (2001b) evaluated the 50-yr record of annual thunder days in the United States and found that increases in thunder days occurred from the southern Plains (Texas and Oklahoma) through the western United States, with decreases in the East. This pattern contrasts with the results reported here, where fewer changes occur in the far West, while CAPE, LI, and K show clear increases in the East. Another study by Changnon (2001c) shows widespread increases in annual thunderstorm rainfall across the country. These results coincide with the positive trends in CAPE, LI, and K in summer. However, significant positive trends in heavy storm rainfall are again prominent in the West. A simple explanation for the western inconsistencies may be the differences in the time periods examined; some western stations show increased instability when the 1948–65 period is included.

While the large-scale conditions measured by radiosonde-derived stability indices promote severe weather development, even optimum conditions do not always produce a severe storm. For example, Rasmussen and Blanchard (1998) found that sounding-derived forecast parameters have a very high false alarm rate; only a fraction of the extreme values of the parameters they studied were associated with supercells and tornadoes. They also showed that indices based on storm relative helicity are strong predictors of severe storms and may therefore best serve as surrogates for specific storm types. However, because these parameters rely on an accurate estimate of the storm motion vector, they may not be suitable for climatological investigations of the scope presented here, which employed over 360 000 soundings. Nonetheless, given the forecast utility of these wind shear parameters, particularly when they are combined with CAPE, future climatological studies are encouraged.

6. Conclusions

With the common severe weather indicators selected for this study, no attempt was made to distinguish among storm types. As atmospheric conditions evolve from “stable” to “unstable” to “very unstable,” it is likely that the probability of severe weather development increases (Schultz 1989). This study may provide some insight into climatological changes in severe storms by depicting temporal changes in severe weather potential. In summer, large portions of the country show significant increases in stability index means, and in the number of days crossing upper thresholds of LI, CAPE, and K. While spring trends are often toward enhanced instability, they are sporadic, with most stations showing no change. Furthermore, the SWEAT Index shows a marked decrease in spring over much of the eastern United States.

The most consistent patterns are restricted to two regions: 1) The Plains states (particularly from Texas to Kansas) exhibit significant positive trends in nearly every variable examined; 2) the West shows few changes in upper index extremes, although the index means and lower-threshold extremes indicate some movement toward enhanced atmospheric instability at some stations. Finally, this analysis suggests that, for the 1973–97 period, the number of very unstable days has increased across the eastern two-thirds of the country and so has the potential for severe thunderstorm development.

Acknowledgments

This research was supported in part by a NASA Graduate Fellowship in Earth System Sciences. The author also wishes to thank Dr. Orman Granger and two anonymous reviewers for their thoughtful comments, as well as Steve Chiswell and Vadim Yushprakh for their assistance with radiosonde data and software.

REFERENCES

REFERENCES
Angell
,
J. K.
,
W. P.
Elliott
, and
M. E.
Smith
,
1984
:
Tropospheric humidity variations at Brownsville, Texas and Great Falls, Montana, 1958–80.
J. Climate Appl. Meteor.
,
23
,
1286
1295
.
Bluestein
,
H. B.
,
1993
:
Synoptic-Dynamic Meteorology in Midlatitudes.
Vol. 2, Oxford University Press, 594 pp
.
Braun
,
S. A.
, and
J. P.
Monteverdi
,
1991
:
An analysis of mesocyclone-induced tornado occurrence in northern California.
Wea. Forecasting
,
6
,
13
31
.
Changnon
,
S. A.
,
2001a
:
Damaging thunderstorm activity in the United States.
Bull. Amer. Meteor. Soc.
,
82
,
597
608
.
Changnon
,
S. A.
,
2001b
:
Thunderstorms Across the Nation. Changnon Climatologist, 93 pp
.
Changnon
,
S. A.
,
2001c
:
Thunderstorm rainfall in the coterminous United States.
Bull. Amer. Meteor. Soc.
,
82
,
1925
1940
.
David
,
C. L.
,
1976
:
A study of upper air parameters at the time of tornadoes.
Mon. Wea. Rev.
,
104
,
546
551
.
Davis
,
R. E.
,
T. M.
Stanmeyer
, and
G. V.
Jones
,
1997
:
A synoptic climatology of tornadoes in Virginia.
Phys. Geogr.
,
18
,
383
407
.
DeGaetano
,
A. T.
, and
R. J.
Allen
,
2002
:
Trends in twentieth-century temperature extremes across the United States.
J. Climate
,
15
,
3188
3205
.
Easterling
,
D. R.
,
J. L.
Evans
,
P.
Groisman
,
T. R.
Karl
,
K. E.
Kunkel
, and
P.
Ambenje
,
2000
:
Observed variability and trends in extreme climate events.
Bull. Amer. Meteor. Soc.
,
81
,
417
426
.
Elliott
,
W. P.
, and
D. J.
Gaffen
,
1991
:
On the utility of radiosonde humidity archives for climate studies.
Bull. Amer. Meteor. Soc.
,
72
,
1507
1520
.
Gaffen
,
D. J.
,
1993
:
Historical changes in radiosonde instruments and practices. Rep. 50, WMO/TD- No. 541, World Meteorological Organization, 123 pp
.
Gaffen
,
D. J.
,
1994
:
Temporal inhomogeneities in radiosonde temperature records.
J. Geophys. Res.
,
99
,
3667
3676
.
Gaffen
,
D. J.
, and
R. J.
Ross
,
1999
:
Climatology and trends of U.S. surface humidity and temperature.
J. Climate
,
12
,
811
828
.
Galway
,
J. G.
,
1956
:
The lifted index as a predictor of latent instability.
Bull. Amer. Meteor. Soc.
,
37
,
528
529
.
George
,
J. J.
,
1960
:
Weather and Forecasting for Aeronautics.
Academic Press, 673 pp
.
Giordano
,
L. A.
,
1994
:
A fingertip guide to key stability and shear index values used in evaluating severe weather and flash flood potential. National Weather Service Eastern Region, ERH Tech. Attachment 94-4, 7 pp
.
Gordon
,
J. D.
, and
D.
Albert
,
2000
:
A comprehensive severe weather checklist and forecast guide. National Weather Service Central Region, NWS Tech. Service Publication TSP-10.
Grazulis
,
T. P.
,
1993
:
A 110-year perspective of significant tornadoes. The Tornado: Its Structure, Dynamics, Prediction and Hazards, Geophys. Monogr., No. 79, Amer. Geophys. Union, 467–478
.
Hales
,
J. E.
,
1985
:
Synoptic features associated with Los Angeles tornado occurrence.
Bull. Amer. Meteor. Soc.
,
66
,
657
662
.
Hoaglin
,
D.
,
F.
Mosteller
, and
J.
Tukey
,
1983
:
Understanding Robust and Exploratory Data Analysis.
Wiley, 527 pp
.
Houghton
,
J. T.
,
Y.
Ding
,
D. J.
Griggs
,
M.
Noguer
,
P. J.
van der Linden
,
X.
Dai
,
K.
Maskell
, and
C. A.
Johnson
,
2001
:
Climate Change 2001: The Scientific Basis.
Cambridge University Press, 881 pp
.
Johns
,
R. H.
,
J. M.
Davies
, and
P. W.
Leftwich
,
1993
:
Some wind and instability parameters associated with strong and violent tornadoes, part II. The Tornado: Its Structure, Dynamics, Predictions and Hazards, Geophy. Monogr., No. 79, Amer. Geophys. Union, 583–590
.
Jones
,
P. D.
,
T. J.
Osborn
,
K. R.
Briffa
,
C. K.
Folland
,
E. B.
Horton
,
L. V.
Alexander
,
D. E.
Parker
, and
N. A.
Rayner
,
2001
:
Adjusting for sampling density in grid-box land and ocean surface temperature time series.
J. Geophys. Res.
,
106
,
3371
3380
.
Karl
,
T. R.
,
N.
Nicholls
, and
A.
Ghazi
,
1999
:
Weather and Climate Extremes:.
Changes, Variations, and a Perspective from the Insurance Industry. Kluwer Academic, 349 pp
.
Kelly
,
D. L.
,
J. T.
Schaefer
, and
C. A.
Doswell
,
1985
:
Climatology of nontornadic severe thunderstorm events in the United States.
Mon. Wea. Rev.
,
113
,
1997
2014
.
Lanzante
,
J. R.
,
1996
:
Resistant, robust and non-parametric techniques in the analysis of climate data: Theory and examples, including applications to historical radiosonde station data.
Int. J. Climatol.
,
16
,
1197
1226
.
Lau
,
K-M.
,
J-Y.
Lee
,
K-M.
Kim
, and
I-S.
Kang
,
2004
:
The North Pacific as a regulator of summertime climate over Eurasia and North America.
J. Climate
,
17
,
819
833
.
Lehmann
,
E. L.
,
1975
:
Nonparametrics: Statistical Methods Based on Ranks.
Holden-Day, 457 pp
.
Meehl
,
G. A.
, and
Coauthors
,
2000
:
An introduction to trends in extreme weather and climate events: Observations, socioeconomic impacts, terrestrial ecological impacts, and model projections.
Bull. Amer. Meteor. Soc.
,
81
,
413
416
.
Miller
,
R. C.
,
1972
:
Notes on analysis and severe-storm forecasting procedures of the Military Warning Center. Air Weather Service (MAC), Tech. Rep. 200, Scott Air Force Base, IL, 181 pp
.
Rasmussen
,
E. N.
, and
D. O.
Blanchard
,
1998
:
A baseline climatology of sounding-derived supercell and tornado forecast parameters.
Wea. Forecasting
,
13
,
1148
1164
.
Robertson
,
A. W.
, and
M.
Ghil
,
1999
:
Large-scale weather regime and local climate over the western United States.
J. Climate
,
12
,
1796
1813
.
Ross
,
R. J.
, and
W. P.
Elliott
,
2001
:
Radiosonde-based Northern Hemisphere tropospheric water vapor trends.
J. Climate
,
14
,
1602
1612
.
Schaefer
,
J. T.
, and
J. G.
Galway
,
1982
:
Population biases in the tornado climatology. Preprints, 12th Conf. on Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., 51–54
.
Schultz
,
P.
,
1989
:
Relationship of several stability indices to convective weather events in northeast Colorado.
Wea. Forecasting
,
4
,
73
80
.
Trenberth
,
K. E.
, and
J. W.
Hurrell
,
1994
:
Decadal atmosphere-ocean variations in the Pacific.
Climate Dyn.
,
9
,
303
319
.
Wilks
,
D. S.
,
1995
:
: Statistical Methods in the Atmospheric Sciences. Academic Press, 467 pp
.

Footnotes

Corresponding author address: Dr. Diana DeRubertis, 2195 Station Village Way, Apt. #1117, San Diego, CA 92108-6516. Email: diana.derubertis@gmail.com