1. Introduction

Among the possible effects of climate changes is the higher frequency of extreme temperature and atmospheric moisture events (Houghton et al. 2001; Delworth et al. 1999). Heat waves of extreme high temperature and humidity are one of the major natural hazards that cause severe human mortalities and economic damages in the United States and worldwide (Kalkstein 1991; De et al. 2004). It is also believed that a human-induced climate change would increase summer mortality despite the development of acclimatization (McMichael et al. 1996; Kalkstein and Greene 1997). Greene and Kalkstein (1996) show that the relationship between mortality and high heat/humidity varies regionally across the United States. However, Davis et al. (2004) indicate that climate-induced mortality calculated on a monthly scale during the 1990s showed little variation across the United States.

Heat waves have a typical duration of a week and often occur in connection with long-term drought, which may extend for months or even years (Chang and Wallace 1987). There are cases, however, in which heat waves develop in the absence of drought (Lyon and Dole 1995). As precipitation deficits increase, daytime high temperatures increase because of the increase in the sensible heat fluxes as a result of the depletion of soil moisture and the reduction of evapotranspiration (ET) from vegetation. Huang and van den Dool (1993), for example, showed that there is a negative relationship between precipitation and drought across the Midwestern United States. Namias (1982, 1991) has suggested that a positive feedback loop can be set up whereby long-term drought conditions promote and maintain anticyclonic conditions that further inhibit precipitation. Research by Lyon and Dole (1995) and others indicates that there is a complex interplay between local (e.g., decreased ET and a higher sensible heat flux) and remote forcings in the development and maintenance of heat-wave droughts. In the investigation of two heat-wave droughts in the central United States, they found that remote forcings played a relatively greater role in the earlier stages of the event as local forcing became more significant later in the event. In one of the events, anomalously dry conditions during the prior spring contributed to the desiccation of vegetation later in the summer and an increase in the sensible heat flux (i.e., hotter daytime temperatures).

Several studies (Kunkel et al. 1996; Livezey and Tinker 1996; Palecki et al. 2001) have identified the synoptic conditions contributing to the buildup of excessive heat and moisture in two deadly Midwest heat waves. They identified the 1) presence of an upper-tropospheric ridge that induces descending air and adiabatic warming; 2) advection of warm air; and 3) strong solar insolation over an urban heat island. It is interesting, however, that these heat waves were not tied to dry conditions in the boundary layer (i.e., drought), but rather to abundant soil moisture in the upwind direction, which caused vigorous evapotranspiration and moistening of the lower atmosphere. Furthermore, a low-level capping inversion associated with subsidence aloft effectively trapped the atmospheric water vapor, allowing dewpoint temperatures to reach record levels. In the southern United States, Henderson (1995) showed that warm outbreaks are tied to a strengthened Bermuda high and increased subsidence, which promote clearer skies and greater solar insolation.

It is not clear if hot episodes in the southeastern United States correlate with drought or exceptionally dry or moist boundary layer conditions. Moreover, the associations between the synoptic environment and the occurrence of unusually warm conditions (hereinafter referred to as hot events) are unclear. More specifically, the role of the surface and boundary layer conditions in mediating these relationships is not known. The objective of this study, therefore, is to identify relationships between temperature/humidity, and synoptic/boundary layer conditions in a large sample of hot events in the Piedmont region of North Carolina. The following two questions will be addressed: 1) What concurrent/antecedent synoptic features and boundary layer conditions best distinguish exceptionally hot and moist days from moderately warm summer days? On what temporal scales do the greatest distinctions occur? 2) What variables most effectively distinguish hot, dry days from hot, moist days?

2. Study area

Hot events in this study are based on the surface weather observations from the Raleigh–Durham International Airport (RDU), which is situated in a rapidly urbanizing region between the cities of Raleigh and Durham in east-central North Carolina (Fig. 1). This area lies in the eastern Piedmont, a region of gently rolling terrain with elevations ranging between 60 and 250 m. The northeast–southwest-trending Blue Ridge Mountains are situated approximately 100 km northwest of the area. The RDU site was chosen because it has an excellent record of hourly observations of temperature and dewpoint. In a pilot study, the same analyses were also carried out for Charlotte, another Piedmont city about 270 km southwest of Raleigh. Because of the great similarities found in both synoptic patterns and boundary layer conditions, it was decided that the situations responsible for the hot events defined at RDU are representative of those for the Piedmont region in general. With exception to the last several years of the study period, the local area around RDU was largely rural in character with patches of fields and early successional forests. Any urban heat island effects were therefore confined to those associated with the sensible heating of the airport's structures and runways. To test for urban heat island effects, comparisons were made between the mean annual summer season temperatures at RDU relative to three cooperative observer sites in the broader region. The comparisons revealed that the temperatures at RDU were slightly warmer than the cooperative observer sites on average (0.8°C); however, no secular trend was noted during the study period. In particular, the highest mean temperature differences (1.7°C) were noted from the mid-1960s to the mid-1970s, while the lowest differences (0.4°C) were found in the late 1950s to early 1960s and the late 1980s and early 1990s. These differences may be related to several very localized movements of the weather station during the study period.

3. Data and methods

Mean daily values of surface air temperature and dewpoint at the National Weather Service (NWS) RDU station were calculated from the 3-hourly data from a quality-controlled dataset by Robinson (1998). These data were extracted for the summer months (June–August) of the period of 1951–93. The temperature data alone were organized into time series and used to identify the hot events for this study. A moving-windows approach was used to identify temperature maxima in the time series. In particular, the temperature mean for each day of time series was compared with the daily temperature means for the 2 days prior to and following the given day. If the temperature mean was greater, the day was defined as a hot event. Each event was therefore defined by the occurrence of a daily mean temperature within a surrounding 5-day-or-greater time window (e.g., Fig. 2). This definition ensures that any two hot events are at least 3 days apart. In most cases, the 3-day-or-greater separation period ensures that adjacent events are not connected with the same synoptic-scale circulation feature. However, extended periods of hot weather (e.g., heat wave) typically have durations on the order of a week (Chang and Wallace 1987), and therefore could conceivably be represented by two consecutive hot events in the temperature time series. It should be noted that a hot event is not necessarily hot in an absolute measure, but is the warmest day within a 5-day period or greater. As a result, these events are evenly distributed across the summer months, rather than clustering in the hottest month of July. Using these criteria, 634 events were identified during the study period.

Among these hot events, we have focused on the extreme ones in terms of temperature and humidity or combinations of both variables. Four extreme-event types are defined, with the number of events in parentheses in each category, as indicated below:

1. hottest (63)—temperature at or above the 90th percentile,

2. moistest (63)—dewpoint at or above the 90th percentile,

3. hot, humid (40)—both temperature and dewpoint in the top quartile, and

4. hot, dry (40)—temperature in the top quartile and dewpoint in the lower quartile.

Comparisons were made between the events in each category and the events in the remainder of the sample (e.g., the 40 hottest events were compared with the remaining 594 events.) These comparisons reveal the combination of ingredients that most distinguish a given event type.

The daily surface weather data were supplemented by hourly precipitation data obtained from the National Climatic Data Center (NCDC). Additionally, hourly weather conditions were obtained from NCDC Surface Airways Data CD-ROMs (EarthInfo 2001). To identify the synoptic features associated with hot events, gridded (2.5° latitude × 2.5° longitude mesh), twice-daily synoptic fields were extracted from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) 40-year reanalysis dataset (Kalnay et al. 1996). These fields (Table 1) were spatially interpolated onto a 1332 km × 1332 km floating grid at a 222-km interval centered over the eastern Piedmont of North Carolina. Using the 0000 and 1200 UTC gridded synoptic fields, a temporal interpolation was undertaken to estimate field values for the 1500 LST, which is typically the map time immediately prior to the hottest time of the day. Synoptic fields were extracted for the 6-day period prior to each event. An inverse distance technique was used to carry out all spatial and temporal interpolations.

Composite (i.e., mean) synoptic fields were created for each category and then compared. These composites highlight synoptic features common to each category. Twice-daily radiosonde data from the station at Greensboro (GSO, WBAN 13723), North Carolina, were used to represent the vertical profiles of temperature, humidity, and wind associated with each event in the sample. Greensboro is situated about 96 km west of the Raleigh–Durham observing site (Fig. 1); it is the nearest site in the radiosonde network. It should be noted that mesoscale variations in the vertical profiles exist between the two sites, which cannot be readily estimated in this work. Composite vertical profiles were also produced. The earliest available sounding data were in 1954 (i.e., no data for the events during the first year of the study period). To assess antecedent conditions in the boundary layer, temperature, dewpoint, and sky cover were averaged for a range of time periods (i.e., 2–30 days) prior to each event. Additionally, rainfall hour and precipitation totals were determined for each antecedent time period. Comparisons were made between the means of the variables in order to identify the variables that most effectively distinguish the boundary layer character between each extreme-event type and the remainder of the sample.

4. Results

5. Discussion and conclusions

In this study, a large sample of hot events were defined in the Piedmont region of North Carolina and described climatologically in terms of their connections with the present as well as antecedent surface and atmospheric conditions. In particular, the environments associated with the most extreme hot and moist events were distinguished from the events constituting the remainder of the sample. Additionally, the environments of hot, dry events were distinguished from those associated with hot, moist events. Composite analyses revealed that the synoptic circulations were generally similar across the different event types. Most events occurred immediately downstream of a 500-hPa ridge, and in some cases, beneath this ridge. At the surface, a quasi-stationary Bermuda high was well established over the western Atlantic Ocean with an arm extending into the Gulf of Mexico. This pattern is similar to that observed by Henderson (1995) in warm outbreaks identified across the southern United States. The southwesterly to westerly low-level flow associated with this circulation is adiabatically warmed because of its descent off the lee of the Blue Ridge Mountains to the west. Composite 200-hPa divergence fields showed persistent convergence, suggesting subsidence and adiabatic warming in the midlevels of the troposphere. Sounding composites at 1200 UTC indicated a low-level inversion, which could be attributed to the combined effects of adiabatic warming aloft and the remnants of radiatively cooled air at the surface. The relatively stable environment characterizing the events encouraged relatively clear skies and strong solar insolation. Additionally, the more stable conditions inhibited deep convective mixing, thus inhibiting the ventilation of the heat and moisture in the daytime boundary layer.

The mean wind in the hot event sample veered from southwesterly at the surface to west-northwest in the upper levels, suggesting the occurrence of warm advection; however, the composites showed no significant thermal advection immediately before or during the event. In most events, this lack of warm advection can be tied to the relative absence of a north–south thermal gradient. This stands in contrast with the July 1995 Midwestern U.S. heat wave, where a southwesterly lower-tropospheric flow produced warm air and moisture advection that contributed to the high heat indices (Livezey and Tinker 1996).

The hottest events in the sample are most distinguished by a strong westerly wind component in the lower troposphere. It is estimated that 65% of the events displayed a sufficiently strong westerly wind component to cross the Blue Ridge Mountains and warm adiabatically on the leeside. Our analysis suggests that downslope adiabatic warming is a contributor to the hot conditions in the Piedmont region of North Carolina. Upper-level divergence also effectively distinguishes these events; it contributes to middle-tropospheric subsidence and adiabatic warming. However, it is unclear how much of this adiabatically warmed air is actually mixed down to the surface and can therefore contribute to the surface heating. The hottest events were also distinguished by a low sky cover (i.e., strong solar insolation), little precipitation, and relatively fewer precipitation hours during the antecedent period leading up to the event. This finding implies a high Bowen ratio promoted by the high levels of sensible warming resulting from the relatively dry soils. The relatively dry antecedent conditions are similar to those observed in the heat-wave droughts of 1980 and 1988 in the Southern Great Plains and Midwestern United States, respectively (Lyon and Dole 1995).

The moistest events in the sample are most distinguished by high levels of convergence in the upper troposphere, which promote middle-tropospheric subsidence and lower-tropospheric stability. These events showed the lowest mean inversion height (at 1200 UTC), which can be tied to stronger subsidence aloft. The stability may act to inhibit the ventilation of moisture out of the daytime boundary layer as discussed above. The moistest events also show a significantly higher coverage of clouds and precipitation totals during the period leading up to the event, which implies moister soils and more active vegetation growth. As a result, evapotranspiration is relatively greater, thereby providing more water vapor to the boundary layer. This atmospheric pattern was identified in the very humid heat waves that occurred over the Midwestern United States during the 1990s (Kunkel et al. 1996; Livezey and Tinker 1996; Palecki et al. 2001).

Comparisons of the hot, moist and hot, dry events revealed the importance of antecedent precipitation. The hot, moist events were associated with much higher precipitation totals (4.57 cm) during the antecedent 30-day period as compared with the hot, dry events. The increased precipitation provides wetter soils and increased vegetation growth, which in turn increases rates of evapotranspiration. Mean temperature over the past 30 days is also 1.2°C higher in the hot, humid events, which further encourages evapotranspiration. Note that these higher temperatures result from the relatively higher positive skewness in the temperature distribution of the hot, humid events. There were no significant differences identified in moisture advection between the hot, dry and hot, moist event samples. This suggests that the moisture source is the increased evapotranspiration resulting from wet soils and active vegetation because of the higher precipitation totals in the weeks leading up to the event. This pattern was observed in the Midwestern U.S. heat wave of 1995; however, the exceptionally moist conditions occurred upstream, which provided water vapor that was then advected into the region (Livezey and Tinker 1996).

Although the relationships between summer weather and health have not been explicitly examined in the Piedmont region of North Carolina, excessive heat and humidity are known to affect mortality and morbidity. The antecedent atmospheric conditions associated with hot, moist events in this work (i.e., those associated with dangerously high heat indices) provide valuable prognostic information in the advance issuance of heat-wave warnings. In particular, this work suggests that the occurrence of positive monthly precipitation anomalies can set the stage for a potentially hot and moist event. If a northwesterly circulation develops aloft concurrently with these positive precipitation anomalies, the probability of exceptionally hot and moist conditions is greatly increased. Future work should be aimed at the development of predictive relationships between antecedent atmospheric patterns and temperature/dewpoint conditions.

Additionally, this work suggests a connection between hot events and adiabatic warming via downsloping westerly flow immediately east of the Appalachian Mountains. It is unclear, however, how much this effect contributes to the hot temperatures. Confounding influences, such as adiabatic warming resulting from synoptic-scale subsidence, make this determination difficult. Moreover, the spatial scale of this effect downstream of a mountain range is unclear. Detailed analyses of circulation trajectories are needed to address these questions.