The debate regarding the impact of anthropogenic climate change on Earth's remote ecosystems is complicated by the fact that historical weather records are limited in their length, especially at high altitudes where the sensitivity to global climatic change may be paramount (Liu and Chen, 2000; Bagla, 2009). Furthermore, large-scale orographies such as the Himalayan and Tibetan Plateau (H/TP), can also play an important role in influencing climate—most particularly with respect to monsoon circulation (Liu and Chen, 2000; Yasunary et al., 2004; Kang et al., 2010), especially by acting as an obstacle to southward flow of cool, dry air (Molnar et al., 2010).
Increases in positive temperature trends have been observed in southern pre-Himalayan areas since the 1970s (Shrestha et al., 1999), as well as over the Tibetan Plateau (TP) in more recent decades (Yimin et al., 2007; Liu et al., 2009). This may induce a faster melting of some Himalayan glaciers depending on their own individual behaviours (Inman, 2010). However, permafrost temperatures are mainly controlled by elevation and latitude (together with geothermal heat flux), and concerns are only potentially raised in relation to the possible transfert of the climate warming into permafrost temperature increases and degradation on the TP (Wu et al., 2010). Important teleconnections between El Niño-Southern Oscillation (ENSO) and temperature anomalies of the Tibetan Plateau have been temporally identified by Yin et al. (2000) and Liang et al. (2008) but, in general, temperature teleconnections are not yet well understood across Asia. This is mainly due to sparsely distributed meteorological measurement stations over such remote and mountainous areas (Xie et al., 2008). Today, the Himalayan area benefits from the recently installed instrumental observations (Ueno et al., 2008) operated by Ev-K2-CNR Committee (effective in this area since 1989, http://www.evk2cnr.org, with measured data becoming available by 1994) and, in particular, from the Pyramid Automated Weather Station (hereafter Pyramid AWS), placed at the foot of Mount Everest at 5050 m a.s.l. (86.80°E and 27.95°N).
The present study establishes the longest temperature (1901–2009) and indices series ever for this high-elevation area, taking advantage of both high-resolution gridded temperature data made available by the Climate Research Unit (CRU TS3) (Brohan et al., 2006) and global datasets arranged via Climate Explorer web-GIS by both the Hadley Centre for Climate Prediction and Research and the National Climatic Data Center (van Oldenborgh et al., 2009). The annual trends in reconstructed temperature time series at the Pyramid AWS since 1901 was used to document the interdecadal-scale variability patterns over TP and to study potential temperature teleconnections.
A variety of datasets was evaluated for the purposes of temperature modelling and reconstruction (Table I). In the following sections, their use is elaborated upon in detail.
Table I. Availability and source of temperature datasets used for model calibration and validation, reconstruction of the temperature time series, and testing (out-of-sample validation)
Annual temperature data for Pyramid AWS before 1994 were estimated from monthly CRU TS3 dataset (Brohan et al., 2006). An overlapping period over 1994–2005 was available with temperature records from both Pyramid AWS and the ∼50-km gridded CRU TS3 dataset (available on a monthly basis). Interpolating points (longitude: 86.75–87.25°E, latitude: 27.75–28.25°N) of the CRU TS3 dataset are in the Tibetan Plateau, within a range of elevations compatible with the Pyramid AWS location. After extracting the temperature for the same coordinates of Pyramid AWS—by Climate Explorer interpolation (van Oldenborgh et al., 2009)—a relationship was found between the two series for the calibration dataset (1994–2001). From the same CRU TS3 dataset, data from 2002 to 2005 were used for validation. Monthly mean temperatures (Tm, °C) at the Pyramid AWS were estimated using a linear model approximated by:
where a = − 3.8 ( ± 0.11) and b = 0.75 ( ± 0.02) are model parameters (R = 0.98), and Tm(CRU) is the monthly temperature ( °C) of CRU TS3 dataset. Least-square regression analyses were run to estimate the parameters of Equation (1). The entire process was assessed interactively using Microsoft® Office Excel 2003 with the support of Statistics Software–R modules (Wessa, 2009).
2.2. Evaluation of temperature series
The estimated temperature monthly values by Equation (1) were compared against observational values. The agreement between estimates and Tm-values was evaluated using the modelling efficiency by Nash and Sutcliffe (Nash and Sutcliffe, 1970) as performance statistic (ranging from negative to positive unity, the latter being the optimum value; positive values indicating that the model is estimating better than the average of the observed data). For a visual inspection of the quality of results, a set of comparative scatterplots and histograms (four graphs) are also presented based on temperature simulations to evaluate how the model works. A comparison against long-term records was performed using data from Kathmandu Airport, Nepal (the nearest available station), provided for 1950–2009 by the Climate Explorer database (WMO station code 44 454, van Oldenborgh et al., 2009).
2.3. Climate change investigations
Continuous-time signals of climate change and trends in the reconstructed series were investigated by using a Morlet mother wavelet based on the same procedure described in Torrence and Compo (1998). Wavelet transforms decompose complex information and patterns into elementary forms. They are efficient in determining the damping ratio of oscillating signals and are resistant to the noise in the signal, changing adaptively to the time and frequency resolution (e.g. Ding, 2008).
The following climate indices were also used to characterize and explain the influential role of local processes and teleconnections that influence over TP temperatures: Pacific Decadal Oscillation (Mantua et al., 1997), Atlantic Multidecadal Oscillation (Goldenberg et al., 2001), and Asian-Pacific Oscillation (Zhao et al., 2007).
3. Results and discussion
3.1. Evaluation of reconstructed temperature data
A general agreement is apparent between observed data and Tm distributional estimates with both calibration, 1994–2001, and validation, 2002–2005, datasets (Figure 1, graphs A and B, respectively). Nash-Sutcliffe index of 0.97 reflects negligible departures of the data points from the 1:1 line. Similar values of Nash-Sutcliffe index (0.96) obtained also with validation dataset indicate certain stability in the model performance. These results also reveal a bimodal distribution in monthly temperature values (Figure 1, graphs A and B), whilst the histograms of residuals are close to quasi-normal distributions (Figure 1, graphs A1 and B1).
In order to verify if temperature reconstruction from monthly records was also able to capture interannual variability at the Pyramid AWS, predicted and observed annual temperatures (mean of monthly temperatures) were compared (Figure 2). The years 1994–2006 define a period of overlap with reanalysis data provided by the National Centers for Environmental Prediction (NCEP), National Oceanic and Atmospheric Administration of the United States Department of Commerce-Earth System Research Laboratory (http://www.cdc.noaa.gov).
The NCEP reanalysis represents an updated source providing relatively coarse spatial data (about 1°× 1°). For the period investigated, NCEP temperature fluctuations are largely approached by our reconstruction, and both series do not differ substantially from the observed profile (Figure 2, graph A).
Figure 2, graph B displays the comparison, for the testing period 1950–2009, of the predicted temperatures at Pyramid AWS and the observations taken at Kathmandu station (Nepal). The latter is placed 100 km apart and about 4000 m lower than the Pyramid station, thus, lack of correspondence between the two series is not unexpected. However, the overall trend of predicted temperatures can be considered to agree sufficiently well with the observed data series of Kathmandu. This also corresponds to the distributions of seasonal and annual temperature trends shown by Shrestha et al. (1999) for Kathmandu and high-elevation sites of the Himalayan region.
3.2. Warming trends
The Himalayan climate is characterized by a considerable significant temperature increase, referred to as a core of + 0.5 °C for the period 1971–2005 in comparison to 1901–1960 (Figure 3A1). Similarly, satellite tropospheric temperature data detected from 1979 and 2007 revealed small (positive) warming trends over the TP compared to the strong warming trend observed in the western Himalayan region (Gautam et al., 2009), although there are major physical differences between surface and tropospheric temperatures (Parker, 2000).
Focusing on Mount Everest, Figure 3A2 shows the temperature anomaly series for the Pyramid AWS statistically reconstructed back to 1901 using a linear regression between gridded (CRU TS3) and Pyramid AWS station data updated until 2009 (Figure 3B1). The temperature series at this high elevation appear affected by an increasing trend with irregular steps and sudden shifts in amplitude that took place particularly during 1960s and 2000s. This was also confirmed by a spectral analysis (Morlet mother wavelet), which detected two significant temperature increase trends (marked by rectangles in Figure 3B2). The comparison with temperature proxy records collected in the southeast TP area by Yang et al. (2009) adds further confirmation to our temperature reconstruction shown in Figure 3B1. The reconstruction performed by these authors also shows a significant warming trend after the 1950s. Although the unprecedented warming occurred during the 2000s decade, several times over the past two millennia the climate was considerably warmer in the TP, and for longer periods, than during the late 20th century (Liu et al., 2002; Feng and Hu, 2005). Furthermore, no trend and no important changes are visible in agreement with the global warming paused over 2000–2009 (Kerr, 2009; Camuffo et al., 2010; Simmons et al., 2010; Solomon et al., 2010).
Albeit the TP is a small part of the North Hemisphere (NH), it may respond in a different way to climate forcing (also due to its complex surrounding topography), and then it is interesting to assess the degree of teleconnections between NH and TP, taking advantage of both CRU TS3 and NCDC-NH (land + ocean) temperature series and TP temperature pattern. The result highlights a moderate but long and important coupling between NH warming and TP temperature in the period 1901–1960 (Figure 4A). This outcome is supported by a previous study performed by Bhutiyani et al. (2007) indicating, for the 20th century, similar epochs of temperature variation between global and northwestern Himalayas. Nowadays, although a coupling in the northwestern area of the TP appears clearly and with certain persistence and it can be related to global warming, at the same time, on the rest of southeastern region across the Pyramid AWS, the unvarying TP temperature with NH warming is an unexpected and hard-to-figure behaviour (Figure 4). It is the southeastern TP, in particular, that seems to resist the mechanisms of global warming after 1970 (Figure 4B).
This is also revealed by a correlation pattern between the global and TP temperatures in the latitude-height cross-section (Figure 5, graphs A and B), based on the NCEP reanalysis. In particular for 1971–2009 (graph B), it is visible that while temperatures to the north of the TP and in Nepal agree to the global warming (green and yellow areas), the southern part of the TP (circle) resists this global forcing, indicating the existence of different driving forces between northern and southern plateau zones. Not entirely supported by Shrestha et al. (1999), whose study was limited to 1971–1994, our results are consistent with an updated investigation by Ding and Zhang (2008), who found that the recent acceleration of warming in the TP has occurred later than in the north of the Yangtze River (east China), especially in winter and spring. In the recent decades, some unknown forcing working in a decoupled way, seems to have triggered the southeastern TP to depart from the NH climate. This decoupling emerging from the CRU TS3 estimates reveals the existence of spatial heterogeneity of regional forcing factors affecting current warming (Bradley et al., 2003).
3.3. Temperature teleconnections
Climate indices may influence the climate at regional scale, although areas in which important changes are detected can be scattered around the globe and lack consistency in time (Sterl et al., 2007).
The Atlantic Multidecadal Oscillation (AMO, defined from the patterns of sea surface temperatures), for instance, may have played an important role in influencing the TP climate, since it appears to be part of a wider phenomenon driven by variation in the surface heat flux and coherent with parallel fluctuations in the strength of the thermohaline circulation (Burroughs, 2003). This is indicated by a positive AMO–temperature correlation pattern, prominent and more powerful in 1971–2005 than in the previous longer period 1901–1960 when some significant correlation was already apparent limited to the southeastern TP (Figure 6A, B).
With equal emphasis and extension, a confounding change in the correlation patterns was also detected between Pacific Decadal Oscillation (PDO, detected as warm or cool surface waters in the Pacific Ocean) and temperature (Figure 7A, B), and for which a high predictability was found in relation to global climate change on decadal time scales (Mochizuki et al., 2010) as above. This was, in turn, consistent with the Asian-Pacific Oscillation (a zonal teleconnection pattern over the extratropical Asian-Pacific region)—which is associated with the out-of-phase relationship in atmospheric heating—and that shows a decadal variation according to the high index polarity registered before 1975 and to the low index polarity registered afterwards (Zhao et al., 2007). Conforming to the warming phase 1971–2005, the AMO and PDO effectively marked the last decades, and Figures 6 and 7 show that the switch of both AMO and PDO to their warm cycles from a precedent cool cycle (1901–1960) has become firmly established.
Although Sheffield and Wood (2008) found weak correlations between drought and climate indices over the warm period 1950–2000 with El Niño-3.4 and AMO, a significant correlation was found for TP temperature and AMO.
The northwestern TP is highly sensitive to global warming, while the southeastern part seems less prone to these changes. However, the relative importance of these regional factors is not clear, particularly because neither the response to regional forcing nor the regional forcing itself are well known for the 20th century (Shindell and Faluvegi, 2009). It is possible that climate–environment interactions in the Tibetan area coincide with global warming at a sub-regional level, thus leading to a speculative interpretation (Qian and Xue, 2010). For instance, You et al. (2010) suggested that change in regional atmospheric circulation is one of the important factors contributing to the recent climate warming of the TP. Considerable changes in teleconnection strength were also detected in recent decades compared to 1901–1960. Changes in cloud amount (Duan and Wu, 2006) and land use (Zhang, 2007) can also be considered important factors contributing to the recent climate warming in the TP, although local and global greenhouse gas emissions cannot be excluded. The combined action of tropospheric and stratospheric volcanic aerosols, due to their attenuation of incoming solar radiation, may also have a cooling effect on surface air temperature (e.g. Balling and Idso, 1991).
4. Summary and conclusions
In this study, we investigated the dynamic patterns of surface warming in the Tibetan Plateau, by correlating temperature series reconstructed in situ with large-scale (Northern Hemisphere) temperature variability. While some results seem hardly explainable, we highlight that the complex interactions between climate and temperature variability over the TP are still poorly understood and the mechanisms driving future temperature changes are not completely clarified.
The TP is located in a unique geographical zone with peculiar altitudinal characteristics and low average air temperature. Basing on our study, its climate since 1901 has changed both in time and in space indicating the non-uniform effects of global warming. Global warming started in around1850, after the end of the Little Ice Age, but its pattern could have changed character since the 1970s. The changes of air temperature in the last decades exhibit a zonational pattern not always significant. The start of a new decade (from 2010 onward) appears promising, as observed and predicted climates reveal them in a way not seen in the past 100 years. A challenge and priority for future research lies in attaining a much better understanding of what is causing the global warming. In particular, more comparative studies between different types of data sources are possible, and certainly also desirable for the reason that remote regions (such as the TP) include inhomogeneities resulting from disposal or relocation of stations and changes in the local environment. Moreover, further investigation to seasonal scales would help in the process of finding explanations for the trends and teleconnections. This will be possible as more records become available in situ, and homogenized data will be provided by the CRU TS3 on a monthly basis.
The authors are grateful to specialist colleagues Casey Saenger (Department of Geology and Geophysics at Yale University) and Carlo Casty (Risk Management and Reserves of Zürich, Switzerland) who have provided helpful comments while the research for this paper was in progress, and on draft versions of the manuscript. The authors are also grateful for the possibility offered them of using predictor data from various sources, and mainly from the Royal Netherlands Meteorological Institute (in the person of Andreas Sterl) for supplies by the Climate Explorer database. We also thank the Ev-K2-CNR Committee for the use of Pyramid AWS data and the work carried out to maintain the monitoring activity in the Everest area.