Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
 The tropical lower stratosphere is an important region of the atmosphere, where strong convective activity in the underlying troposphere affects both its chemical and dynamical properties. Temperatures near the tropopause influence the input of water vapor from the troposphere and act as an indicator of the dynamical properties of the region. This paper addresses long-term trends in the temperature of the tropical lower stratosphere. Correlations with recent changes in tropical stratospheric water vapor are also noted. Special attention is given to the convectively active tropical western Pacific Ocean, where sea surface temperatures (SSTs) are among the highest in the world. The region contains several island radiosonde stations with records reliably extending over several decades. Results show only weak cooling trends occurred before the 1990s, but a strong and rapid cooling of 4° to 6°C took place in the mid-1990s and has persisted since that time. The properties of the temperature records during and following this cooling event are discussed, and a significant anticorrelation with SST anomalies in the underlying ocean is noted. The rate of ocean warming increased in the early 1990s, coinciding approximately with the mid-decade cooling event, while individual monthly anomalies in both time series are also anticorrelated. Past work has shown that cooling of the tropical lower stratosphere is a dynamical result of tropospheric convection, which in turn partially depends upon sea surface temperatures. Convection may therefore be the link between the ocean and the stratosphere, and the increased cooling may be an indication of strengthening tropical convection.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The tropical lower stratosphere and the underlying tropopause are regions of particular interest in connection with issues of global climate change. The lower stratosphere, roughly defined as the region between pressure levels of 100 and 10 hPa, contains most of the stratosphere's ozone content, while the extreme cold of the tropical tropopause acts as a filter for water vapor, freezing out most of the water that reaches the tropopause, thus preventing it from entering the stratosphere. The details of the interaction between the troposphere and the stratosphere remain unclear, but a fairly recent upsurge in both observational and theoretical activity provides some hope that the important processes will soon be clarified.
 The basic balance in the lower stratosphere is between the radiative heating and adiabatic cooling due to upwelling [Corti et al., 2005; Gettelman et al., 2004]. Convective clouds are important to both processes, and radiative cloud heating can provide an important feedback source that can act to strengthen both the Walker and Hadley circulations [Sohn, 1999] and thus influence the lower stratosphere. Because convective activity itself is affected by the temperature of the underlying ocean, a connection between SSTs and temperatures in the upper troposphere and lower stratosphere (UTLS) would not be surprising. Evidence for such a connection will be discussed below.
 It is well established that the seasonal cycle of water vapor entering the stratosphere is largely the result of the seasonal cycle in temperature near the tropopause [Mote et al., 1995, 1996]. Exactly how long-term changes in tropical near-tropopause temperatures relate to the amount of water vapor that makes it into the stratosphere is currently unclear [Rosenlof et al., 2001]. The NOAA Boulder (40°N) long-term frost point research balloon record shows increases in water over a 20–25 year period, while the temperature record shows cooling at the tropical tropopause [Randel et al., 2000, 2006]. What is interesting to note is that the temperature trends at the tropical tropopause determined from the radiosonde record shown by Randel et al.  are ∼0.5 K/decade, while the annual cycle has an amplitude of ∼3 K [Reed and Vleck, 1969]. This translates to a maximum to minimum difference of ∼6 K, which is an order of magnitude greater than the temperature changes noted over a 10-year period. Because of the difference in scales, it is not surprising that the response in water vapor is much clearer on an annual basis than for long-term trends. However, it is important to ascertain how changes in temperature near the tropical tropopause relate to those in the entry values of stratospheric water vapor, because of possible radiative impacts in the stratosphere.
 The stratosphere itself is known to be cooling, approximately in line with what is expected from known changes in ozone and greenhouse gases. Shine et al.  have compared the results of several model simulations of stratospheric temperature trends, using both global and zonal averages, and have compared the model results to observations. While there is generally reasonable agreement with observations and among the models themselves, significant discrepancies do exist, particularly in the tropics, where observed cooling trends are larger than the simulated trends. In the case of the lower stratosphere (from the tropopause to the 10-hPa level), which is our concern in this paper, ozone depletion appears to be the major radiative cause of cooling, but the influence of dynamical changes remains much less clear. Additionally, Shine et al.  note that changing assumptions about water vapor trends could improve model/observational agreement in temperature, indicating that understanding processes that change stratospheric water vapor on long timescales is important in understanding the radiative budget of the stratosphere.
 In this paper we shall discuss the long-term trends in temperature in the tropical lower stratosphere, with particular emphasis on recent anomalous variations. While these variations have been noted in earlier publications [e.g., Randel et al., 2004; Thompson and Solomon, 2005] no clear picture of their cause and their relationship to other climatic parameters has emerged. Such trends and anomalies have important implications for the minor constituent composition of the stratosphere, particularly for water vapor. As noted above, water vapor primarily enters from the convectively dominated tropical troposphere and is assumed to be freeze-dried to the saturation vapor pressure at the point of minimum temperature (the “cold point”), as originally suggested by Brewer . The freeze-drying mechanism is strongly supported by the so-called “tape recorder” effect seen in measurements of stratospheric water vapor by the Halogen Occultation Experiment (HALOE) on the Upper Atmosphere Research Satellite (UARS), in which the annual cycle of cold point temperatures is duplicated by a similar annual cycle in water vapor just above the tropopause [Mote et al., 1996]. The detailed quantitative agreement between temperatures and water vapor concentrations, however, is not perfect and undoubtedly depends on small-scale transport processes [Rosenlof, 2003].
 The anomalous behavior we will focus on in this paper is of relatively recent origin and contains unique features that have no parallel in earlier periods, at least since regular radiosonde measurements of tropical stratospheric temperatures began in the 1950s. In what follows, we shall describe evidence for a possible link to variations in sea surface temperature (SST) in the western tropical Pacific. Possible connections to the major El Niño–Southern Oscillation (ENSO) event of 1997–1998 might be expected, but no strong evidence for such a connection has yet emerged.
2. Data Sources
 The chief sources of data for this study are the NCAR/NCEP Reanalysis Project (obtained from the NOAA/CIRES Climate Diagnostic Center) and archived radiosonde data provided in the Integrated Global Radiosonde Archive (IGRA) described by Durre et al. , which superseded the Comprehensive Aerological Data Set (CARDS) produced by NOAA's National Climatic Data Center in Asheville, North Carolina. Use has also been made of data from the UARS satellite mentioned above.
 Radiosonde measurements provide an excellent source of information on temperature, winds, and geopotential heights in the lower stratosphere, but care must be taken in using the archived data to derive long-term trends, as has been pointed out frequently [e.g., Lanzante et al., 2003a, 2003b; Randel and Wu, 2006]. Instrumental changes are a particular source of problems [Seidel et al., 2004] but can often be identified readily by sharp discontinuities in the record. Atmospheric temperature trends derived from radiosonde measurements have been the subject of many studies by Angell and his collaborators [e.g., Angell and Korshover, 1983; Angell, 2003], who have devoted a great deal of effort to determining the usefulness of individual radiosonde stations and continuity of their archived records. Continuity is more of a problem with the stratosphere than with the lower levels, simply because the maximum altitude reached by a balloon is variable and the probability of loss of signal increases with distance from the recording station. In addition, confidence in the reliability of data from a given station is greatly increased if there are other stations in the same geographical area for comparison. For that reason, the radiosonde data used in this study have been confined largely to the island stations of the western tropical Pacific Ocean, which also happen to lie mostly within the so-called “warm pool,” where climatological sea surface temperatures (SSTs) exceed 28°C. The region is also one of the three major global areas of deep tropospheric convection on an average basis and thus one of the main contributors to the tropospheric Hadley and Walker circulations.
 The island stations are shown in the map in Figure 1, together with their location relative to the SST warm pool. The archived temperature records from all of the stations shown extend back to the late 1950s, and most of the individual daily soundings reach well into the lower stratosphere, extending to at least the 10-hPa pressure level in most cases. The four stations of Koror, Yap, Truk, and Majuro form a nearly longitudinal chain, while Pago Pago in Samoa is included as a single Southern Hemisphere representative, and Lihue at 22°N latitude represents the Northern Hemisphere border region between the tropics and the subtropics. Lanzante et al. [2003a] and Randel and Wu  have identified discontinuities in some of these stations. However, for our analysis, and in particular for changes noted post-1990, these discontinuities are not sufficient to preclude the use of these stations.
3. Data Analysis
 The analysis here is largely concerned with monthly mean temperature anomalies rather than the actual temperatures. Anomalies are derived on a monthly mean basis by subtracting the long-term average temperatures for each calendar month of the year from the actual monthly mean measured values, thus eliminating the pronounced annual temperature variation. Diurnal variations at these altitudes are small and have not been taken into account in this preliminary study.
Figure 2 shows the time series of monthly mean temperature anomalies at a pressure level of 70 hPa recorded by radiosonde instruments at the four island stations in the western tropical Pacific Ocean between 1970 and early 2006 and the time series at all the available pressure levels at Yap. As mentioned above, the stations were selected on the basis of their long-standing and reliable records, with similar instrument packages and a minimum of operational problems. The four stations making up the western chain in Figure 1 are all between 5°N and 10°N latitude and between 120°E longitude and the date line. Since they lie within the 28°C climatological SST contour, which roughly defines the western Pacific warm pool, their data can be taken as representative of this important climatic region, while the overall similarity of the time series in Figure 2 shows that the assumption of homogeneity over the warm pool region is reasonable.
 The earlier part of the record at all stations shows little long-term trend, the dominant component of the time variation being the temperature response to the quasi-biennial oscillation (QBO) in the tropical stratospheric winds. Figure 3 compares the 70-hPa temperature anomalies at Koror with the monthly mean QBO zonal wind speeds at the higher 30-hPa level. The relation between temperature and the QBO is in phase and fairly clear prior to about 1990 but changes character at later times. It is interesting to note that the in-phase QBO/70-hPa relationship appears to be strong again for the period 1999–2001 and then is weaker again following 2001 to the present, when the 70-hPa temperatures seem to settle in at an average 3°C cold anomaly with a range from 2° to 5°Colder than the temperatures for the 1980–2006 period. The overall character of the temperature record began to change for all stations at roughly the same time, with a pronounced cooling by about 4°C over a 10-year period. In early 1999 all stations showed an abrupt rise in temperature by about 4°C to a level close to that of the pre-1995 long-term average, remaining at that level for nearly 2 years, then equally abruptly dropping by about 6°C to a level several degrees colder than any recorded during the previous 30 years. Figure 3 shows that the 1999 warming event accompanied the westerly swing of the QBO and the strong cooling accompanied the return of the easterly phase. The peak easterly winds were not unusually strong, but the corresponding temperature anomalies were the coldest of the entire record, presumably through superposition of the overall cooling of the earlier 1990s on the QBO effect. The record through 2005 shows possible indications of a recovery to more normal levels and a reappearance of the QBO signal in a muted form. Although earlier records show multiyear cooling periods in the lower tropical stratosphere, sometimes apparently linked to major volcanic eruptions or to the ENSO cycle, none of these has the magnitude or detailed structure of the events shown here.
 Although the main emphasis in this discussion is on the western tropical Pacific warm pool region, the global distribution of the temperature anomalies is also an interesting question. Because of the sparse distribution of reliable tropical radiosonde stations, the best global information comes from data assimilations, which include information from radiosondes and satellites. Figure 4 shows the zonally averaged temperature anomalies in the 10°N to 10°S latitude belt derived from the UARS/UKMO data assimilation [Swinbank and O'Neill, 1994]. The pressure level resolution of the assimilation is less than that of radiosondes, but the main features at the 70-hPa level are clearly similar to those shown in Figure 2. The warming event centered starting in 1999 is seen to be confined between about 90 and 50 hPa with a maximum at 70 hPa and is followed by a strong cooling. This cooling appears to disrupt the QBO signal below about 50 hPa, in agreement with the behavior shown in Figure 2.
 Since the UKMO/UARS temperatures presented in Figure 4 are zonal averages, a natural question is whether the western Pacific variation is dominating the global record, while the rest of the tropics is behaving normally, or whether the zonally averaged record does in fact reflect the variation occurring at all longitudes. The UKMO/UARS temperatures show relatively uniform anomalies with respect to longitude in the tropics. However, examination of a few reliable radiosonde records from other tropical longitudes shows that some of the some features of the record do appear in other locations but not with the same magnitude as over the warm pool. The records at the subtropical station of Lihue (22°N) and the Southern Hemisphere tropical station of Pago Pago (14°S) (presented in Figure 9 in section 5) show similar features to those of the northern tropical stations but rather less sharply defined. They are, however, surprisingly well correlated with each other, considering that they lie in different hemispheres, separated by more than 4000 km. However, as will be discussed further in this paper, there are other locations that show different temporal features.
4. Sea Surface Temperature Connection
 The time series of monthly temperature anomalies at the 70-hPa level over Koror is shown in Figure 5, together with the time series of SST anomalies averaged over a large segment of the equatorial western Pacific warm pool (7.5°S to 4.5°N, 120°E to 180°). Both time series can be taken as typical of temperature variations over the lower stratosphere and the ocean surface over the warm pool, since both variables have a high degree of spatial correlation over the relevant distances.
 Before about 1994 the dominant variation in the 70-hPa temperature anomalies was the response to the quasi-biennial wind oscillation (QBO), with a peak-to-peak variation of roughly 6 K, while the SST anomalies show a fairly random variation with a standard deviation of about 0.25 K, superimposed on a slow warming trend of about 0.1 K/decade, indicated by the least squares trend line in Figure 5. The cross correlation between the two time series in the 1960–1993 period is about −0.15 at zero lag, which is barely significant, considering the autocorrelation in the stratospheric values. SST variations during this 34-year time period thus account for only about 2 percent of the stratospheric temperature variance. Clearly, the major contributor to this variance is the QBO response, making it difficult to detect any other component without removing the QBO signal. This has not been done in this study. The strength of the QBO signal changes with time, becoming significantly weaker after the mid-1980s and changing as noted below.
 The period following 1993 shows significantly different behavior from that of the earlier period. For the present discussion it has been divided into the two sections 1994–2000 and 2001–2005, indicated by the vertical lines in Figure 5. The 70-hPa temperature anomalies showed a marked cooling during 1994–1996, reaching a value lower than any recorded during the preceding 35 years. The QBO signal that had dominated the earlier period was strongly attenuated, following the weakening that began in the mid-1980s. It was still present, however, as indicated by the short horizontal bars over the positive phases of the QBO signal in 1995–1996, 1997–1998, and 1999–2000. The last of these appearances returned the 70-hPa anomaly to a positive value close to its precooling level but the following negative phase brought the temperature anomalies down to a new record low level of about −6 K. Figure 5 shows that it was accompanied by a cooling event in the SST record of similar duration but moderate magnitude. The major event in tropical SSTs during this period was the intense El Niño of 1997–1998 (the strongest on record), which had several unusual properties [McPhaden, 1999], including a sharp termination and transition to a La Niña event, described by McPhaden  as “explosive,” in the middle of 1998. These dramatic SST phenomena appeared to have had little effect on the lower stratospheric temperatures, at least over the western tropical Pacific, unlike other ENSO events recorded by the same radiosonde stations [e.g., Reid, 1994].
 The final period, from 2001 through 2004, shows the 70-hPa anomalies in a new mode, with the dominant feature being an annual variation. The temperature anomalies reach −6 K at their sharply marked minima, each of which corresponds to a positive maximum in the SST anomalies. The SST anomalies are the highest of the entire record, and the 70-hPa temperature anomalies are the lowest of the entire record. Possible mechanisms and implications of this cyclical correspondence will be discussed in the final section. Meantime we turn to a discussion of the trends in the stratospheric cold point and the consequent variations in water vapor entering the lower stratosphere.
5. Trends in Height and Temperature of the Cold Point and Tropopause
 The standard World Meteorological Organisation (WMO) definition of the tropopause, i.e., the lowest point at which the lapse rate drops below 2°C km−1 and remains there for at least 2 km, can be applied in the tropics as well as in midlatitudes, but tropical radiosonde profiles often reveal the coldest point in the profile at some higher level [e.g., Selkirk, 1993; Seidel et al., 2001]. The term cold point has come into common usage to define this level. This level corresponds to the minimum saturation vapor pressure of water being transported upward from the upper troposphere. These cold values are a major player in determining the water vapor concentration in the lower stratosphere. The reasons for the existence of the cold point and its relationship to the lapse rate tropopause are not well understood, and radiosonde measurements show that the cold point temperature and height are highly variable from day to day. Monthly mean values thus have a questionable meaning, particularly for the water vapor concentration, since the saturation vapor pressure is highly sensitive to temperature in this cold environment. Similarly, monthly zonal means are also of questionable value [Dessler, 1998]. Nevertheless, we shall assume that monthly mean cold point temperatures provide us with at least a qualitative estimate of the vapor pressure of water vapor entering the lower stratosphere. This is supported by the existence of the tropical tape recorder signal in water vapor [Mote et al., 1995] and also by changes in water vapor described in the following section of this paper. In actuality, the minimum temperature a parcel encounters on its transit to the stratosphere is likely the controlling factor in determining the water vapor concentration that enters the stratosphere. Looking at monthly mean cold point temperatures over a long period allows us to more easily examine correlations and trends. Note that the minimum temperature does not have to be encountered in vertical upward transit at all longitudes, but, as demonstrated initially by Holton and Gettelman , horizontal transport through a cold region is sufficient to dry air to stratospheric concentrations even if the cold region is longitudinally restricted in the tropics.
 The dramatic decrease of tropical tropopause/lower stratosphere temperatures during the 2000–2001 time period is global in nature, as evident in Figure 4, and appears in the NCAR/NCEP reanalysis product (not shown here), as well as in the UKMO/UARS assimilation, and in global averaged radiosondes. This is shown in Figure 6, which plots the anomaly time series of 100 hPa and cold point temperatures averaged over 52 radiosonde stations in the 20°N to 20°S latitude belt. These particular radiosonde stations were chosen because they had mostly continuous temperature records over the period of interest in the lower stratosphere and did not display unusual discontinuities in time. All data came from the Integrated Global Radiosonde Archive (IGRA) described by Durre et al. . A list of the stations used to construct Figure 6 is given in Table 1. As seen in Figure 6, the drop in the cold point temperatures at the end of 2000 is considerably more pronounced than the corresponding cooling at the 100-hPa level, close to the WMO lapse rate tropopause, indicating a change in the lower stratospheric lapse rate as well. A corresponding change in lower stratospheric water vapor also occurred in conjunction with this extended cooling episode; this is discussed in the following section.
Table 1. List of Stations Used to Construct Figure 6
Fatuna Island Pago Pago Island
Cocos Island Airport
Manaus Ponta Pelada
Penang Bayan Lepas
Ponape Caroline Island
Koror Caroline Island
Truk Caroline Island
Yap Caroline Island
Trinidad Piarco Airport
San Andres Island
Point A Pitre Raizet
Sint Martin Juliana
San Juan Isla Verde
Chiang Mai International
Mexico City International Airport
Figure 7 shows the geographical distribution of the changes in NCEP tropopause temperatures between the 1990s and the 2000s; the difference shown is between 1995–1997 and 2001–2003. These particular years were chosen to straddle the period of the large temperature drop, and so that each average is not biased toward a particular phase of the El Nino/Southern Oscillation. The largest cooling appears to have taken place in the equatorial western Pacific and eastern Indian Ocean region, roughly corresponding to the maritime continent of the Indonesian islands, although there is a somewhat weaker and less extensive cooling in the eastern equatorial Pacific. The 3-year periods differenced in Figure 7 included one warm, one cold, and one neutral El Niño year. The geographic differences in cooling are also evident in individual radiosonde stations data (Figure 8), not only in the NCEP/NCAR assimilated temperatures. Of interest is the fact that the cold point (or tropopause) temperature drop at the end of 2000 is prominent in the warm pool region and over Africa at all seasons but not a major feature in the Caribbean or the mid-Pacific. Over South America, the drop is evident during the December–January–February period but not at other times of the year. The consistent feature at these tropical stations is the downward temperature trend from 1992 through 1997, except at Hilo, although the drop does appear at Lihue, in close proximity to Hilo.
 Trends in cold point temperature and geopotential height anomalies are illustrated in Figure 9 for five of the radiosonde stations in Figure 1, covering the time period from 1966 through the first half of 2005. All five stations show a fairly steady increase in cold point height, consistent with a warming troposphere and a cooling stratosphere [Santer et al., 2003]. The rise in height can be reasonably fit by a least squares straight line in each case, with slopes varying from 60 m per decade at the stations closest to the center of the warm pool to 90 m per decade at the more outlying stations. A straight-line fit to the height difference between Lihue and Yap (not shown here) shows that Lihue cold points were about 150 m lower than those at Yap during the 1960s and were at about the same height in 2005, suggestive of a greater tropospheric warming and/or stratospheric cooling. It should be noted, however, that the difference is small compared with the annual variation in height at either station, which has a total range of more than 1 km.
6. Changes in Stratospheric Water Vapor
 Water vapor in the stratosphere is important for both radiative and chemical processes. Forster and Shine  showed in a modeling study how changes in stratospheric water can have a significant impact on stratospheric temperatures. Their model results showed that observed changes in water vapor from 1979 to 1997 caused changes in stratospheric temperature that were comparable with those caused by observed trends in carbon dioxide and ozone during the same period. The changes were most significant in the lower stratosphere but reached their greatest magnitude in polar regions.
Shine et al.  reviewed estimates of 20-year temperature trends in the stratosphere and concluded that the differences between modeled and observed temperature trends in the lower stratosphere are reduced if the cooling effects of increased stratospheric water are included. Unfortunately, accurate global measurements of stratospheric water vapor do not exist for an extended period of time. There are indications of a significant long-term trend prior to 2000 [Kley et al., 2000; Oltmans et al., 2000; Rosenlof et al., 2001] in the Northern Hemisphere middle latitudes, while a sharp decrease occurred in both tropical and middle latitudes in about 2000. This decrease is clearly seen in the tropical HALOE observations from the UARS satellite as shown in Figure 10 and coincides approximately with the rapid cold point cooling seen in the radiosonde measurements shown in Figures 6 and 10. Figure 11 shows the long-term correlation between tropical tropopause temperature averaged zonally over the latitude belt from 5°S to 5°N and the zonally averaged water vapor concentration centered at the 82-hPa level as measured by the HALOE instrument on UARS. Peak correlation occurs when the water vapor plot lags the temperature plot by about 2 months, presumably the traveltime from the tropopause to the 82-hPa level. The drop in water vapor content is translated upward in the tropics and reaches the 10-hPa level approximately 1.2 years later, as shown in Figure 12. The drop in water vapor content in the tropics that took place about 2001 is also seen in midlatitude (40°N) HALOE measurements and is evident as shown in the NOAA ESRL GSD (former CMDL) frost point balloon measurements at 70 hPa, with a decrease in early 2001, bringing values back down to mixing ratios comparable to those seen in 1990.
Randel et al.  attribute the drop in water vapor to the drop in tropical temperatures. The change in the 82-hPa HALOE water vapor anomalies is ∼0.4 ppmv. HALOE average water vapor mixing ratios at 82 hPa are ∼3.25 ppmv. This corresponds to a saturation mixing ratio at 190 K and 100 hPa. Reducing the temperature to 189 K gives a saturation mixing ratio of ∼2.75 ppmv, or a reduction of ∼0.5 ppmv. The individual station cold point drops shown in Figures 8 and 9 range from 1° to 2°, while the zonal average drop shown in Figure 11 is ∼1.5°. The observed tropical temperature drop and water vapor decrease do appear to be reasonably in line with each other. The actual temperature drop at the end of 2000 appears to be correlated with an increase in the Brewer Dobson circulation, also noted by Randel et al. . An estimate of the anomalies in the 10°S–10°N zonal tropical upwelling is shown in Figure 13. These were calculated by first estimating the radiative heating rates using input constituents from HALOE and temperatures from UARS/UKMO and then calculating the TEM residual circulation stream function as described by Rosenlof . There is a marked increase near the tropopause level at the end of 2000 that is not evident at higher levels in the stratosphere. This is consistent with the observation in Figure 4 that the cooling post-2001 is in a relatively narrow layer near the tropical cold point. It therefore can have a large impact on transport of water vapor into the stratosphere.
 Examination of the distribution of temperature change in the tropics around the globe shows that the step function like decrease in temperature is not zonally uniform. This is demonstrated in Figure 14, which shows the lagged correlation of monthly averaged tropical tropopause temperatures as a function of longitude with the zonally averaged HALOE water vapor time series along with a similar anomaly correlation. Tropopause temperatures lead 2 months relative to the HALOE water time series to yield the maximum correlation. The correlation is longitudinally uniform and on average the correlations coefficient is 0.82 for the monthly average time series correlations. This high correlation is a consequence of the strong annual cycle in both tropopause temperatures and water vapor and results in the tape recorder signal noted by Mote et al. [1995, 1996]. However, as noted by Rosenlof , the annual cycle in tropical tropopause layer temperatures does not correlate highly with an annual cycle in zonally averaged tropical SSTs, in that maximum near tropopause temperatures are in July-August when SSTs are at a minimum, but the minimum tropical tropopause layer temperatures occur in December–January, not in April, when SSTs are at a maximum. For the case of zonally averaged anomalies, SST and tropopause temperatures are also not well correlated; the coefficient is −0.23 using NCEP 10°S–10°N tropopause temperatures and the Optimal Interpolation V2 SST data from NOAA/CIRES Climate Diagnostics Center.
 If we examine the anomaly time series correlation, also shown in Figure 14, we see that the water vapor and temperatures are best correlated in a limited longitudinal region; only from 171°–200° longitude is the correlation coefficient greater than 0.5. This is a longitude region just to the west of the Indonesian subcontinent (warm pool region). Although the temperature change appears limited in longitudinal scope, it can still have an impact on zonal distribution of lower stratospheric water vapor, as demonstrated in the model of Holton and Gettelman . An increase in upwelling in an isolated region may also impact other trace species in the lower stratosphere, if they have a sufficiently long lifetime. Thus a change in upwelling induced possibly by convective changes in one longitudinal region could impact ozone in a zonally averaged sense, and at that point, radiative processes as described by Randel et al.  could impact the overall circulation as well. If we further look at correlations between SST anomalies and the NCEP tropopause temperature anomaly between 171° and 200° longitude, we find the maximum negative correlation (of −0.44) to be with SSTs just to the east, from 139° to 171° longitude. The SST and NCEP tropopause temperature time series are shown in Figure 15. The correlation coefficient is −0.44 if one considers the entire time period, however, if one correlates times prior to 2000, or after 2001, the correlations coefficients are quite small. The correlation is exclusively a consequence of the decrease in tropical tropopause temperatures of ∼2°C in 171°–200° longitude band coincident with an increase in SSTs of ∼0.4°C in the 139°–171° tropical longitude band. It is possible that there is a convection forced wave response that produces the UTLS temperature response, but assessing cause requires additional modeling work to ascertain.
 This paper has been concerned with the long-term changes in the temperature of the tropical lower stratosphere. Geographically, the study has been mainly focused on the so-called warm pool region of the western tropical Pacific Ocean, where climatological sea surface temperatures exceed about 28°C. This particular choice of location was made on the basis of two factors: first, the region is noted as one of the principal regions of deep convective activity in the troposphere (in addition to equatorial Africa and the Amazon basin of South America), and hence of rapid upward transfer of material from the boundary layer to at least the base of the stratosphere, and second, the existence of a few island radiosonde stations with long records of reliable data extending several decades in time and allowing a check on artificial temperature changes by comparison between the stations.
 The results have shown relatively weak cooling throughout the 1970s and 1980s, with a strong QBO-related variation. The cooling at the 70-hPa pressure level and at the tropopause is consistent with what is expected in the tropics from ozone depletion, water vapor content, and greenhouse gas increases. A more substantial cooling event occurred in the 1990s, however, and has continued into the current decade. Temperature anomalies from 2001 to the present are at a level roughly 4°C below the pre-1990 values. The main features of this event and its continuation have been described, including a significant muting of the QBO signal, which also showed a transient return to “normal” levels for about 2 years from 1999 to 2001. This event took place during the westerly phase of the QBO at the 30-hPa level and seems to have been centered at about the 70-hPa level and confined to pressure levels below about 50 hPa (see Figure 4).
 The most interesting feature of the anomalous cooling in recent years is an apparent anticorrelation between the stratospheric temperature anomalies and sea surface temperature anomalies in the underlying western tropical Pacific Ocean. Data obtained from the NOAA/CIRES Climate Diagnostics Center show that this region of the global ocean warmed steadily at a rate of about 0.1°C per decade until about 2000, when the mean SST anomaly increased fairly rapidly by about 0.25 K (see Figure 5). Since then a pronounced annual signal has dominated the SST anomalies, accompanied by an antiphased annual signal in the 70-hPa temperature anomalies. The warmest peaks in SST occur in the Northern Hemisphere winter, as do the coldest dips in the stratospheric temperatures. In both cases these peaks or dips represent the warmest or coldest temperature anomalies in the entire 34-year period of the record.
 This result strongly suggests that the underlying ocean can have a fairly direct influence on the lower tropical stratosphere, and one can speculate, though with no quantitative evidence, that the connecting link is probably deep convection in the troposphere. Deep convective towers lift material, such as water vapor, to the upper troposphere, where it is in a position to rise, albeit slowly, into the stratosphere. While this could in principle have a radiative effect on stratospheric temperatures, the time delay between the upper troposphere and the 70-hPa level is too long for the connection to appear as nearly simultaneous as it does. Furthermore, the connection did not appear until the SST anomalies had increased to a level higher than any previously recorded in this data set. One possibility, again unsupported by observation, is that the warmer underlying ocean leads to more powerful convection, through an increased water vapor content and latent heat release. Although most convective towers release their water vapor and die out well below the tropopause, occasional cloud tops in the tropics are known to reach heights well into the lower stratosphere. It is not unreasonable to suggest that a warmer ocean could lead to an increase in the number of these anomalously high clouds and that the normal picture of a tropopause layer located between the cloud tops and the tropopause could be replaced by one in which the tropopause layer is thinner and perhaps even nonexistent much of the time as the ocean warms.
 Other scenarios are certainly possible, both radiative and dynamical. For example, an increase in the coverage of high convective cloud tops would reduce the terrestrial infrared contribution to the stratospheric heat budget. Replacing the warm ocean by a thick cold cloud cover could cause significant radiative cooling. On the dynamical side, stratospheric cooling in the tropics was proposed by Reed and Vleck  to be the result of the upward motion associated with the Northern Hemisphere Hadley circulation, but more recent explanations have invoked the meridional Brewer-Dobson circulation [e.g., Holton et al., 1995], which is thought to be driven by extratropical wave-mean flow interaction. The direct connection between warm SST anomalies and cool stratospheric temperature anomalies described above, however, suggests that at least part of the intensified upward motion is forced locally within the tropics instead of remotely by wave activity. The concept of local forcing is consistent with a recent analysis by Kerr-Munslow and Norton  of conditions near the tropical tropopause using ECMWF data. Their results attributed the cooling in the lowermost stratosphere to upward motion forced by mass divergence out of the tropics in the overlying stratosphere.
 One of the principal reasons for investigating the temperature variability of the lower stratosphere and the tropopause region is the fact that temperatures at these levels influence the entry of water vapor into the stratosphere, where it has a significant effect on both the chemistry and the thermal properties of the region. There may also be feedbacks at play, in that the increased cooling can thus be amplified by the removal of water vapor, which is a radiative heat source in the lower stratosphere.
 Satellite observations have shown that a rapid decrease in stratospheric water vapor occurred in both tropical and midlatitude regions in about 2000 in the same time frame as the sharp cold point cooling, in agreement with expectations from the freeze-drying mechanism. Maximum correlation between zonally averaged tropical tropopause temperatures and zonally averaged water vapor concentration at the 82-hPa level occurred when the water vapor time series lagged the temperature time series by 2 months, presumably representing the time needed for vertical transport of the water vapor from the tropopause to the 82-hPa level and thereby adding one more strong piece of evidence supporting the freeze-drying mechanism.
 Clearly, the circulation of the tropical stratosphere contains several aspects that are not yet well understood. The influence of the Brewer-Dobson circulation, forced by extratropical wave action, is generally accepted, but its relationship to the Hadley circulation, forced by the large-scale global temperature distribution, requires more study [Salby and Callaghan, 2005], as do the effects of locally forced meridional circulations and the major disturbances caused by the ENSO cycle. As mentioned above, the period reported on in this paper included the El Niño event of 1997–1998, which was both very intense and had some unusual properties [McPhaden, 1999], including an anomalously rapid termination in May 1998. While some link to the El Niño event can probably be deduced from the plots in Figure 2 and from the zonal mean cooling above about 70 hPa in Figure 3, there does not appear to be a direct connection with later developments in the stratospheric time series beyond the termination of the SST anomalies in 1998 and also prior to their onset in 1997. A major focus of further study, however, is the shift in the mode of temperature variance in the lower stratosphere, which has occurred in the last few years and which appears to be linked to the warming of the tropical oceans.
 Data used in this paper have come from a variety of sources. Boulder frost point balloon measurements are from NOAA ESRL GMD, courtesy of Sam Oltmans and Holger Vömel. HALOE measurements and the UKMO assimilation are from the NASA Upper Atmosphere Research Satellite (UARS) project. NCEP assimilation and sea surface temperature data were obtained from the NOAA-CIRES Climate Diagnostics Center, Boulder, Colorado, (http://www.cdc.noaa.gov/). We also appreciate our anonymous reviews and conversations with Andrew Gettelman.