By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
With a mean elevation of over 4000 m and a surface area in excess of 3000000 km2, the Tibetan Plateau (TP) is the world's highest and largest plateau (Figure 1). Ringed by high mountains including those of the Himalaya and the Karakoram, the plateau is often referred to as the ‘roof of the world’.
Beginning with the work of Blanford (1884), the TP has been recognized as having a profound impact on both local weather and global climate. These impacts arise because of its role as a topographic barrier as well as an elevated thermal source (Yanai et al., 1992; Molnar et al., 2010). As a topographic barrier, it acts as an obstacle to the interaction between extratropical air to its north and tropical air to its south (Krishnamurthy and Kinter, 2002). It also acts during the winter to separate the subtropical jet stream into southern and northern branches (Liu et al., 2007). The quasi-stationary upper tropospheric anticyclone that exists over the plateau during the summer months is evidence of its role as an elevated heat source (Yanai and Li, 1994). This heating supports the meridional temperature gradient that contributes to the maintenance of the Indian Summer Monsoon (Yanai et al., 1992; Yanai and Li, 1994) as well as contributing to deep convection over the plateau that enhances exchange of atmospheric constituents between the troposphere and stratosphere (Yang et al., 2004; Fu et al., 2006). During winter, the TP acts as an elevated cold land surface that allows for snow accumulation and glaciers to occur at latitudes south of what is typical in other regions of the Northern Hemisphere. Inter-annual variability in this accumulation has been shown to have an inverse relationship to the intensity of the Indian and the East Asian Summer Monsoons (Blanford, 1884; Robock et al., 2003; Wu and Qian, 2003; Pu and Xu, 2009). The glaciers of the TP also act as a water source for approximately 50% of the world's population (Barnett et al., 2005).
The TP has recently been warming at a faster rate than the entire Northern Hemisphere, with evidence of a cool period during the 1950s before the onset of the current warming period (Liu and Chen, 2000). However, the lack of instrumental records prior to the 1950s limits our ability to place this recent warming in a longer-term context. Palaeoclimate reconstructions of temperature over the TP are also scarce. However, a temperature reconstruction using tree ring data from the TP indicates a warming of approximately 0.3°C over the period from the 1870s to the 1990s (Wu, 1995). In addition, an ice core pollen record from the central TP shows a summer warming of approximately 0.15°C from 1860 to 2001 (Yang et al., 2008).
Atmospheric reanalysis projects that make use of modern numerical weather prediction systems to assimilate historical observations into consistent and homogeneous datasets have revolutionized the diagnosis of atmospheric processes and their impact on many aspects of the climate system (Kalnay et al., 1996; Bengtsson et al., 2004; Simmons et al., 2010; Dee et al., 2011a). The recently completed 20th Century Reanalysis (20CR) project—a reanalysis dataset that assimilates only surface pressure observations back to 1871–provides the first three-dimensional representation of the state of the troposphere in the period prior to the establishment of the synoptic-scale upper-air network in the late 1940s (Compo et al., 2006, 2011).
In a hydrostatic atmosphere, warming in a layer increases the pressure in that layer and hence it follows that the regional warming that occurs below the height of a plateau, like the TP, is reflected as an increase surface pressure over the plateau (Toumi et al., 1999). The use of surface pressure over the plateau to document regional warming has several advantages. Most notably, surface pressure is an easily measured and robust field that, unlike temperature, is relatively insensitive to small-scale fluctuations and therefore more representative of the large-scale atmospheric conditions (Trenberth et al., 1987; Jarvinen et al., 1999; Compo et al., 2011).
In this paper, we will use surface pressure and temperature observations from the Global Historical Climatology Network (Peterson and Vose, 1997), the annual mean Northern Hemisphere surface air temperature time series from the HADCRUT3 dataset (Brohan et al., 2006), as well as the surface pressure and 2 m air temperature fields from the 20CR and the Interim Reanalysis (ERAI) from the ECMWF (Dee et al., 2011b) to document the relationship between surface pressure of the TP and regional and hemispheric surface air temperatures.
Geophysical time series of the sort used in this paper tend to be temporally autocorrelated, resulting in so-called ‘red noise’ behaviour that leads to a reduction in the degrees of freedom, which can have an impact on the significance of trends and correlations (Allen and Smith, 1996; Ghil et al., 2002). To take this into account, the statistical significance of the trend in a given time series was estimated with a resampling technique that randomizes the phase of the components of its Fourier decomposition to generate 1000 synthetic time series that preserve the power spectra of the original time series (Rudnick and Davis, 2003). The distribution of the trends from these synthetic time series was then used to estimate the statistical significance of the trend of the original time series. For the 20CR and ERAI trend calculations, this technique was applied at each grid point. A similar approach was used to estimate the significance of the regressions of the annual mean surface pressure fields from the two reanalyses with the regional and hemispheric annual mean surface air temperature time series.
Singular spectral analysis (SSA) is a non-parametric analysis technique, similar to empirical orthogonal functions employed in spatial analysis, that uses data-adaptive functions to separate a time series into components that are statistically independent and that maximize the variability in the original time series that is described (Ghil et al., 2002). The non-parametric nature of the method allows for the identification of generalized trends and oscillatory signals that may not be captured using conventional techniques that rely on a decomposition into trigonometric functions (Ghil et al., 2002).
Figure 2 shows the annual mean surface air temperature and surface pressure time series at Lhasa in Tibet (29.67°N, 91.13°E; 3649 m asl) and at New Delhi in northwest India (28.58°N, 77.20°E; 216 m asl) as well as the least squares linear fit over the entire period of the records. Refer to Figure 1 for the location of the two stations. The magnitudes of the trends and their significance, estimated using the resampling method described in the Methods Section, are shown in Table 1.
Table 1. Trend in surface temperature and surface pressure time series at Lhasa and New Delhi. Bold trends are statistically significant at the 99% level, while the italicized trend is statistically significant at the 95% level.
0.35°C per decade
0.24 mb per decade
0.12°C per decade
−0.22 mb per decade
With regard to the Lhasa data, both time series have a positive trend that is statistically significant at the 99% level. In contrast, the New Delhi surface air temperature time series has a positive trend, statistically significant at the 99% level, while the surface pressure time series has a decreasing trend that is statistically significant at the 95% level. Table 1 also indicates that the warming trend at Lhasa is approximately three times larger than that at New Delhi, a result consistent with previous work that showed the TP is warming at a faster rate than the rest of the Northern Hemisphere (Liu and Chen, 2000). There is also variability in these time series that is not reflected in these simple long-term trends. For example, the Lhasa surface air temperature time series had a period of cooling near the start of the record before the recent warming (Liu and Chen, 2000).
In Figure 3, we show the scatterplots of the annual mean surface pressure as a function of the annual mean surface air temperature at Lhasa and New Delhi. There is also a positive correlation, r = 0.41, statistically significant at the 95% level, between the two time series at Lhasa; while at New Delhi there is a slight anti-correlation, r = −0.10 that is not statistically significant. The least squares regression coefficient between the two time series at Lhasa is 0.36 mb/°C, while at New Delhi it is −0.17 mb/°C.
These results presented so far indicate that both Lhasa and New Delhi have experienced a warming since the middle of the 20th century that is amplified over the TP. However, the relationship between the surface air temperature and surface pressure time series at the two stations is markedly different. In particular, the surface pressure at Lhasa is increasing and correlated with surface air temperature, while at New Delhi the surface pressure is decreasing and does not exhibit any statistically significant correlation with surface air temperature. The existence of these differences in the magnitude of the temperature trend and in the sign of the surface pressure trend drives home the markedly different climate that exists between these two sites that are only 1300 km apart but differ in elevation by over 3400 m (Figure 1).
As described by Toumi et al. (1999) and elaborated upon in the Appendix, the surface pressure in regions of high elevation, like the TP, contains a record of the regional warming. The statistically significant correlation between the annual mean surface pressure and surface temperature at Lhasa shown in Figure 3 is consistent with this hypothesis. Further evidence of this relationship is provided by the grid point regression of the annual mean surface pressure against the annual mean 2 m air temperature from both the 20CR and the ERAI shown in Figure 4. For both datasets, there is a statistically significant positive relationship between the two fields over the TP. It should be emphasized that the two reanalyses are independent and that the 20CR does not assimilate any temperature data and so its 2 m temperature field is solely a function of the underlying numerical model. For both regressions, the two fields are anti-correlated in the region surrounding the TP. This is consistent with the results obtained at New Delhi. However, unlike what occurred over the TP, the statistical significance of this anti-correlation is not uniform throughout the surrounding region.
As discussed in the Appendix, the magnitude of the regression between the two fields is predicted to be an increasing function of height above sea level. In Figure 5, we stratify the results presented in Figure 4 by the height above sea level of the grid points in the respective reanalysis. In each case, the regression coefficient is an increasing function of height. The theoretical dependence derived in the Appendix plateaus at heights above approximately 4 km and this is also the case with regression coefficients from both reanalyses. However, both have a systematic underestimation of the magnitude of the regression coefficient at a given height that is most pronounced for the 20CR.
Figure 6 shows the annual mean surface pressure time series at Lhasa as extracted from the 20CR as well as the annual mean Northern Hemisphere surface air temperature time series from the HADCRUT3 dataset (Brohan et al., 2006). Also shown are the low-frequency reconstructions determined using SSA (Ghil et al., 2002); see the Methods Section for information on this technique. The low-frequency reconstructions filter out all variability on time-scales less than 20 years. The correlation between these two time series over the period 1871–2008 is 0.37, while the correlation between the two low-frequency reconstructions is 0.65. Both correlations are statistically significant at the 99% level. As expected, the correlation between the low-frequency reconstructions is higher, because of the reduction in the degrees of freedom.
The annual mean surface pressure time series and its low-frequency reconstruction clearly show evidence of the trend towards higher surface pressure since 1871 with a recent acceleration in this trend. For example, the trend since 1871 is 0.04 mb per decade, while since 1980 it is 0.54 mb per decade. Both of these trends are statistically significant at the 99% level based on the significance test described in the Methods section. In addition, there is evidence of oscillatory behaviour with a periodicity of the order of 50–75 years. Included in this low-frequency variability is a period from approximately 1950 to 1970 where there was a tendency towards a reduction in surface pressure that preceded the recent amplified tendency towards higher surface pressures. For the most part, this low-frequency variability in surface pressure is similar to that observed in the Northern Hemisphere surface air temperature time series, with the exception of a period of lower surface pressures around 1930 that is not captured in the temperature time series.
In Figure 7 we show the grid point regression of the annual mean surface pressure from the 20CR against the annual mean Northern Hemisphere surface air temperature time series from the HADCRUT3 dataset over the period 1871–2008. Over the TP, there is a positive regression, statistically significant at the 99% level, regression with values ranging from 0.6 to 0.8 mb/°C. As was found to be the case in the regression against the 2 m air temperature (Figure 4), there is generally an anti-correlation in the region surrounding the TP. Although not shown, the same regression using the surface pressure field from the ERAI is similar to that shown in Figure 7 for the 20CR.
In Figure 8, we stratify the regression coefficients of the annual mean surface pressure from the 20CR against Northern Hemisphere annual mean temperature for the period 1871–2008 as a function of the height above sea level of the respective grid points. Also shown is the same stratification but for the surface pressure field from the ERAI for the period 1989–2010. One can see that the observed functional dependence follows the characteristics of the theoretical dependence on height against sea level. In particular, the magnitude of the observed regression coefficient increases with height above sea level with evidence of a tendency for a flattening in this dependency at heights above 4 km. There is, however, a systematic difference between the predicted and observed regression coefficients for a given height of the order of 20–25% for the 20CR that is effectively eliminated for the ERAI.
We have provided evidence of a correlation between the surface pressure over the TP and regional and hemispheric temperature fields. This, combined with the robustness of the observability of the surface pressure field, suggests that this field contains a record of surface warming over the plateau. The 20CR assimilates only this field and so provides evidence of trends in this field over the plateau extending back into the 19th century. With this thought in mind, we show in Figure 9 the trend in the annual mean surface pressure field from the 20CR over the period 1871–2008 as well as from the ERAI for the period 1989–2010. The statistical significance of the trends was estimated using the resampling technique described in the Methods section. It should be noted that there is some question regarding trends in reanalysis datasets (Bengtsson etal., 2004; Thorne and Vose, 2010). However, there is increasing evidence that these datasets, including the 20CR, contain useful trend information (Gillett et al., 2003; Simmons et al., 2010; Compo et al., 2011; Bengtsson and Hodges, 2011). The figure clearly shows a trend, statistically significant at the 95% level, towards higher annual mean surface pressures over the TP and a compensatory trend towards lower pressures in the area surrounding the TP in both the 20CR and the ERAI. There is an approximate fivefold increase in the magnitude of the trend in the ERAI as compared to the 20CR. The magnitude and sign of the ERAI surface pressure tendency trends at Lhasa and New Delhi are consistent with observations shown in Figure 2 and Table 1.
In this paper, we have used station and reanalysis data to argue that there is a positive correlation between the surface pressure over the TP with regional and hemispheric temperatures. This correlation can be understood in terms of the response of a hydrostatic atmosphere to a warming in an atmospheric layer that acts to increase the pressure in that layer. As a consequence, as first proposed by Toumi et al., (1999) and elaborated upon in the Appendix, the surface pressure over a high-elevation region like the TP contains a record of the regional warming that occurs in the layer below it. In addition, assuming time-invariant lapse rates, the surface pressure and surface temperature over the TP should also be positively correlated. Furthermore, the magnitude of the signal in surface pressure arising from a given change in temperature is predicted to increase with height.
The observed positive and statistically significant correlation between the annual mean surface temperature and pressure at Lhasa (Figures 2 and 3 and Table 1) is consistent with this hypothesis. In contrast at New Delhi, a nearby station that is close to sea level, there is a small anti-correlation that is not statistically significant. The regression of the annual mean surface pressure and annual mean 2 m air temperature in the 20CR and the ERAI both show a positive and statistically significant correlations over the TP (Figure 4) with the magnitude of the regression coefficient between the two increasing with height (Figure 5). The agreement with the theoretical dependence of the regression coefficient with height is better with the ERAI. This may be the result of the fact that the 2 m air temperature in the 20CR is unconstrained by observations and is solely dependent on the underlying numerical model. However, the similarity in the magnitude of the regression coefficient for the two different time periods and models shown in Figure 5 provides one with confidence that there is a physical basis for this correlation.
In addition to a long-term trend towards higher values since 1871, the annual mean surface pressure time series from the 20CR at Lhasa has also been shown to exhibit variability on the multi-decadal time-scale that is similar to that observed in the annual mean Northern Hemisphere surface air time series from the HADCRUT3 (Figure 6). Included in this variability is a period of decreasing surface pressure from the 1950s to the 1970s that was followed by a period of rapid increase in surface pressure. Indeed these time series and their low -frequency reconstructions have a positive correlation that is statistically significant at the 99% level.
We have furthermore extended this result to show that there is a statistically significant correlation between the surface pressure from both the 20CR and the ERAI with the annual mean Northern Hemisphere surface air time series across the TP (Figure 7), with a magnitude of the regression coefficient that increases with height in a manner that is similar to the theoretical dependence derived in the Appendix (Figure 8). Again, the results from the ERAI are in better agreement with the prediction than those from the 20CR. This suggests that there are still systematic issues with the 20CR over the TP. This not unexpected given the data-sparse nature of the region, especially in the early part of the record. Nevertheless, the existence of any signal in this data-sparse region confirms the usefulness of a reanalysis like the 20CR that assimilates only surface pressure.
Finally, we have shown that in both the 20CR and the ERAI there is a dipole in the annual mean surface pressure tendency between the TP and northwest India that has, since the 1870s, resulted in increasing surface pressure over the TP and decreasing surface pressure in the surrounding area (Figure 9). The magnitude of the trend over the shorter period of the ERAI is approximately five times larger than that over the longer period of the 20CR and is consistent with the idea that there has been a recent acceleration in the surface pressure tendency over the TP. This acceleration has also noted in the surface pressure time series from Lhasa (Figure 6).
This long-term trend in surface pressure confirms previous research that compared unique observations of surface pressure made during the 1924 British Expedition to Mount Everest with more current data, concluding that there has been an increase in surface pressure in the region since 1924 (Moore et al., 2011). It is also consistent with a long-term trend towards higher pressures at the summit of Mount Everest that, on decadal to centennial time-scales, is of physiological significance in that hypoxic environment (Moore and Semple, 2009).
Given the documented correlation between the surface pressure over the TP and regional and hemispheric temperatures that is in agreement with theoretical predictions, it follows that the trend in surface pressure over the TP is consistent with the conclusion that the TP has been warming since the 1870s with a recent acceleration that began in the 1980s. For the entire period of the 20CR (1871–2008) the estimated change in annual mean surface air temperature at Lhasa is estimated to be between 0.3 and 0.45°C, while over the period of the most recent warming (1980 onwards) the estimated warming is between 0.75 and 1.2°C. The lower bound is based on the observed correlation between the annual mean surface pressure and surface pressure at Lhasa, while the higher bound is based on the theoretical prediction derived in the Appendix. The long-term estimate of the warming is consistent with both tree ring and ice core pollen data from the TP (Wu, 1995; Yang et al., 2008). The long-term warming estimated from the surface pressure tendency over the plateau is of the same order as the long-term hemispheric warming (Liu and Chen, 2000), suggesting that the recent acceleration in warming over the plateau is not a long-term phenomenon. The recent estimate of the warming is in agreement with the observed warming at Lhasa, i.e. from the data presented in Figure 2 and Table 1, of the order of 1°C.
It should also be noted that in the region surrounding the TP there is an anti-correlation between the surface pressure and both regional and hemispheric temperatures. There appears to be no theoretical basis for this anti-correlation. It is proposed that it is the result of two unrelated processes: namely the negative pressure tendency in this region, which may be a mass balance response to the increasing pressure over the TP, and the surface warming that is also occurring in this region.
Finally, this approach may have applicability to other high-elevation data-sparse regions of the world such as western North America, Greenland, the Andes and Antarctica.
The author would like to thank the associate editor and reviewers for comments that improved the manuscript. The GHCN data were provided by the US NOAA Climate Data Center. The 20th Century Reanalysis Project data were provided by the US NOAA Earth System Research Laboratory. The HADCRUT3 data were provided by the CRU at the University of East Anglia. The ERAI data were provided by the ECMWF. This research was funded by the Natural Sciences and Engineering Research Council of Canada.
The hydrostatic equation describes the variation in pressure Pz with height z in an atmosphere at rest:
where T is the atmospheric temperature, R is the gas constant for dry air and g is the acceleration due to gravity (Andrews, 2010).
The environmental lapse rate Γ is defined as
and under the assumption that it is a constant, (A1) may be integrated to obtain
where P0 is the sea-level pressure, T0 is the sea-level temperature and .
If we now let T0 = T00 + ΔT0, where T00 is a constant reference sea-level temperature and ΔT0 is the variation in sea-level temperature about this constant, and furthermore assume that ΔT0 is small compared to T00, then
The above expression is equivalent, after the assumption that ΔT0<< T00, to that derived by Toumi et al. (1999) and expresses the fact that for a fixed G and P0,Pz increases with ΔT0 at a rate that increases with increasing height.
Assuming a stably stratified atmosphere and representative values for P0 (1000 mb) and T00 (300°C), the slope in (A4) for representative heights of the TP (3000–4000 m) is of the order of 0.8–1.0 mb/°C.
It also follows from (A4) that for a time-invariant lapse rate
Assuming a time-invariant lapse rate, (A4) and (A5) also hold for the variation in the surface temperature ΔTS.
Although it is difficult to qualify the assumption of a time-invariant lapse rate, there is observational evidence that there has been no significant trend in this parameter in the tropical region since 1958 (Gillett et al., 2000). In addition, the dependence of the slope in (A4) or (A5) with lapse rate is small.