Climate change and variability using European Centre for Medium-Range Weather Forecasts reanalysis (ERA-40) temperatures on the Tibetan Plateau

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 110,, doi: /2004jd005230, 2005 Climate change and variability using European Centre for Medium-Range Weather Forecasts reanalysis (ERA-40) temperatures on the Tibetan Plateau Oliver W. Frauenfeld, Tingjun Zhang, 1 and Mark C. Serreze National Snow and Ice Data Center, Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, Colorado, USA Received 13 July 2004; revised 22 October 2004; accepted 15 November 2004; published 19 January [1] Surface air temperature measurements from meteorological stations on the Tibetan Plateau are compared to 2-m temperatures from the European Centre for Medium- Range Weather Forecasts (ECMWF) reanalysis (ERA-40) to assess the accuracy of this reanalysis product. We focus on ERA-40 grid cells containing at least four stations. The reanalysis temperatures are consistently lower, by as much as 7 C. However, temporal correlations are high, indicating that ERA-40 captures the interannual variability very well. The temperature bias is almost exclusively due to differences between the grid cell and the station elevations. We conclude that the ERA-40 temperatures provide better spatial fields of temperature than is possible with stations in this topographically complex, data-sparse area. Using this spatially and temporally continuous data set, we provide a temperature climatology and assess longterm climate trends. The high elevations of the western plateau are generally 10 C cooler than the eastern plateau. During winter, 2-m temperatures are below 0 C on the entire plateau, with summer values of only 0 C in the west and 10 C in the east. While station records point to a long-term climate warming trend, no trends are observed in ERA-40. This could be due to inadequacies of the reanalysis data, although we see no evidence of anomalous nonclimatic shifts. The significant trends in station data may reflect the extensive land use change and industrialization that has occurred on the Tibetan Plateau. Reanalysis data are less influenced by these local effects. Citation: Frauenfeld, O. W., T. Zhang, and M. C. Serreze (2005), Climate change and variability using European Centre for Medium-Range Weather Forecasts reanalysis (ERA-40) temperatures on the Tibetan Plateau, J. Geophys. Res., 110,, doi: /2004jd Introduction [2] The Tibetan Plateau (TP) is the highest (averaging >4000 m) and largest plateau on Earth (Figure 1) and is surrounded by Earth s highest mountains. It represents some of the most complex terrain of the globe. The TP provides an anomalous midtropospheric summer heat source and is therefore also a key factor in generating the Asian monsoon [e.g., Hsu and Liu, 2003] which drives the climate of the Asian continent, impacting more than half of the world s population. It exerts a strong influence on global climate and has been argued to be a critical region for climate change detection. High-altitude areas seem to have warmed more, and perhaps sooner than the rest of the globe [Beniston et al., 1997; Diaz and Bradley, 1997; Aizen et 1 Also at Institute for Plateau Meteorology, China Meteorological Administration, Chengdu, China. Copyright 2005 by the American Geophysical Union /05/2004JD005230$09.00 al., 1997; Liu and Chen, 2000]. The TP in particular is reported to have warmed significantly in recent decades [Liu and Chen, 2000], which has been attributed to greenhouse gas loading [e.g., Chen et al., 2003]. [3] Understanding the complex land-air processes that govern the TP s influence on local to hemispheric scales is hampered by the lack of reliable and (spatially and temporally) continuous long-term data. Station records for the TP are not readily available, and none have been incorporated in global surface air temperature data sets such as those by Jones et al. [1991] and Jones and Moberg [2003], or the Global Historical Climatology Network (GHCN) [Vose et al., 1992]. A potentially valuable source of continuous long-term data for the TP is provided by atmospheric reanalysis efforts. [4] Reanalysis is a form of numerical weather prediction (NWP). NWP systems use an atmospheric model to generate forecasts of free-air atmospheric variables such as tropospheric pressure heights, humidity and winds, as well as surface and near surface fields like 2-m temperature, precipitation, and radiation fluxes. In an ongoing data 1of9

2 Figure 1. Map of south central Asia depicting the location and topography of the Tibetan Plateau. Shading corresponds to DEM elevation. Note that the elevation of the western plateau (west of 90 E) is greater than 5000 m, while the average elevation of the eastern plateau is m. assimilation and forecast cycle, fields from a short-term forecast (6 12 hours) are optimally blended with observations from rawinsonde profiles, satellite retrievals, aircraft reports, and other sources including some surface observations. These so-called analyses are used as the initial state to generate the next set of forecasts. In an operational setting, assimilation and forecast systems are continually refined in an effort to improve forecast skill. This can result in non- Figure 2. Map of Tibetan Plateau, as outlined by the m DEM contours. The symbols represent the 161 meteorological stations; open symbols denote stations below 2000 m, and solid symbols denote those above 2000 m. The grid cells outlined in black represent the 14 grid cells (labeled 1 14) with four or more stations used in this analysis. The grid cells outlined in blue and red correspond to the 54 grid cells used to represent the TP for the plateau-wide analyses. Blue indicates the western grid cells/stations, and red indicates the eastern grid cells/stations. 2of9

3 Figure 3. Map of the Tibetan Plateau as defined by the ERA m model elevation contours. Circles represent the 161 stations; solid circles are those 48 stations that could have been assimilated by ERA-40. climatic shifts and anomalies in archived fields. Reanalysis systems, such as those by National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) [Kalnay et al., 1996] and the new European Center for Medium-Range Weather Forecasts (ECMWF) ERA-40 project [ECMWF, 2002], provide long time series using data assimilation and forecast systems that are unchanged over the study period. This results in more temporally homogeneous fields, although inconsistencies are still present due to changes in the amount and quality of assimilation data. [5] Here we (1) assess the reliability of ERA-40 2-m temperature fields via comparisons of long-term and monthly means and interannual variability from station records on the TP, (2) provide a plateau-wide annual and seasonal temperature climatology based on ERA-40, and (3) assess long-term temperature trends from both the stations and ERA-40 at the grid cell level, for the whole TP, and separately for the data-sparse western plateau and more data-rich eastern plateau. 2. Data and Methods [6] We obtained daily surface air temperature measurements for 161 stations located throughout the TP from the Institute of Plateau Meteorology at Chengdu, China Meteorological Administration (Figure 2). ERA-40 daily 2-m air temperatures (2.5 latitude 2.5 longitude grid) were obtained from the ECMWF web site ( int/). We also use 30-arc sec (1-km) gridded, qualitycontrolled global Digital Elevation Model (DEM) data from the Global Land One-km Base Elevation (GLOBE) Project ( The ERA-40 temperatures are available for 1 September 1957 to 31 August The station data vary in record length but are generally available from 1 January 1951 to 31 December Our study is based on the overlapping period of September 1957 to December [7] Our focus is on monthly timescales. Therefore both the ERA-40 and station data were averaged into monthly means. On the basis of the distribution of stations on the TP we restrict tests of ERA-40 performance to grid cells containing at least four stations (Figure 2). Cells are taken to be 2.5 boxes centered over each ERA-40 grid location. The station values in each grid cell are simply averaged. This aggregation makes the station records more compatible with the spatial scale of the ERA-40 grid cell values. We make use of both raw time series and standardized time series whereby the annual cycles are removed using each month s mean and standard deviation. [8] The TP is characterized by very complex terrain. While the stations vary widely in elevation, they tend to be biased to the populated, lower-elevation areas. By comparison, the ERA-40 temperatures correspond to the elevation of the model grid cells. For realistic comparisons of temperature, it is desirable to account for such differences. For each of the analyzed grid cells we obtain the ERA-40 model elevation, the mean elevation based on the GLOBE DEM data, and the mean elevation of the stations. This allows us to account for differences in temperature between the two products due to elevation. The reanalysis model elevations can be quite different from the DEM elevations. However, the ERA-40 project also uses four additional variables to capture terrain variability: angle, anisotropy, and slope of the subgrid-scale orography, and the standard deviation of orography. 3of9

4 Table 1. Mean Temperatures for ERA-40 and Stations and Their Differences (Biases) Averaged in Grid Cells 1 14 a Grid Cell Temperature, C ERA Stations Difference Elevation, m DEM Stations Difference a Also included are the DEM/grid cell and station elevations and their differences. Grid cells 1 14 are outlined in black in Figure 2. [9] An important difference between the NCEP-NCAR and ERA-40 reanalysis is the treatment of surface observations. While NCEP-NCAR does not assimilate surface observations such as temperature, moisture, and wind [Kalnay and Cai, 2003], ECMWF does incorporate these variables when available. This further complicates our analysis since a subset of the 161 station temperature records were potentially used in ERA-40. Hence our comparisons of station and reanalysis temperatures are not entirely independent. It is unclear exactly which stations were accessible during different periods of the reanalysis. However, ECMWF has precise criteria for the use of surface observations. The temperature output for a particular grid cell from ERA-40 s atmospheric model (based on 6-hour forecasts) is adjusted by the 50 closest stations within a 1000 km radius. However, the station elevations have to be within 300 m of that grid cell s model elevation to be assimilated. Furthermore, the incorporation of an e-folding distance causes stations farther away than 300 km from a grid location to receive negligible weights. [10] Replicating the ERA-40 data assimilation scheme is beyond the scope of our study. However, we can obtain a first-order approximation of how strongly the data assimilation impacts our comparisons. We prepared separate time series based on aggregating stations within each of the 14 cells that (1) are within 300 m of the ERA-40 model elevations, and (2) differ from the ERA-40 elevation by more than 300 m. This procedure recognizes that the strongest weights would be given to nearby stations (i.e., within the grid cell and hence within the e-folding distance of 300 km). On the basis of these estimates, only 48 of the 161 stations would have been assimilated and accorded strong weights (Figure 3). By comparing each of these two time series, (1) and (2), with the ERA-40 data, we have at least some information on model performance with and without the impacts of surface data assimilation. 3. Results and Discussion 3.1. Grid Cell Comparisons [11] Fourteen ERA-40 grid cells contain four or more temperature stations (Figure 2). One of these cells, 1 (100 E, 40 N), is mostly north of the TP, while cell 6 (105 E, 35 N) is partially east of the TP. However, including these grid cells can provide a relative measure of ERA-40 performance on and off the TP and they are therefore retained. Most of the 14 grid cells are located on the eastern TP, with the exception of the cell centered at 90 E, 30 N, which is in the south central TP and contains Tibet s capital, Lhasa (Figure 1). [12] Comparing the long-term grid cell means (averaged over all months) indicates that ERA-40 temperatures are consistently lower, with the exception of grid cell 1 (Table 1). These biases range from C to 6.94 C and are statistically significant according to paired samples t tests (not shown). The smallest biases are for the two grid cells not entirely on the plateau (1 and 6) and the largest is for the Lhasa grid cell, 10. As evaluated over the entire record, the standardized grid cell time series are nevertheless highly correlated, with coefficients ranging from 0.74 to 0.89 (Table 2). To further illustrate some of these points, Figure 4 shows the raw ERA-40 and observed time series Table 2. Correlations Between Standardized Monthly ERA-40 and Averaged Station Temperature Time Series for Grid Cells 1 14, September 1957 to December 2000 a Grid Cell Correlations All Stations Assimilated N/A Number of Stations Total Assimilated N/A a Separate correlations are presented using stations within each grid cell that would ( Assimilated ) and would not have been assimilated ( N/A ) by ERA-40. All correlations are all statistically significant (a < 0.05). Also included is the total number of stations and the number of stations that would and would not have been assimilated (as in Figure 3). 4of9

5 anisotropy, and slope of the subgrid-scale orography, and the standard deviation of orography. Therefore we assume that ERA-40 temperatures apply at the true DEM grid cell averaged elevations. This assumption appears to be valid. [15] Analysis for individual months yields similar results, although correlations tend to be lower during summer and autumn and higher during winter and spring for some grid cells (Table 3). This holds true mainly for cells in the northeastern TP. However, even for months with the weakest correlation, the coefficients are statistically significant at the a level. Accounting for elevation differences, ERA-40 therefore not only captures the monthly variability, but also the absolute temperatures. Since ERA-40 temperatures seem to more accurately reflect the average elevations of the grid cells, it could be argued that the reanalysis provides a better description of large-scale temperature variations across the TP than possible with the sparse station observations. Figure 4. (top) Station (black) and ERA-40 (gray) time series for grid cell 6 (105 E, 35 N) with the smallest temperature bias ( C) and (bottom) same for grid cell 10 (90 E, 30 N) with the largest temperature bias (6.94 C); long-term means are indicated by the horizontal lines. Horizontal dashed lines represent 0 C. for grid cells 6 and 10, while Figure 5 gives the standardized time series for grid cell 10. While the raw time series are offset by almost 7 C, the standardized time series, with the bias thereby removed, agree remarkably well. [13] Closer inspection of the temperature differences reveals that the higher the DEM grid cell elevation, the larger the bias (Table 1). The Lhasa grid cell, 10 (90 E, 30 N), with the largest bias, has the highest mean elevation. Grid cell 1 (100 E, 40 N) has the lowest elevation (north of the TP in the western Gobi desert) and one of the smallest biases. The mean elevation of the grid cells and the mean elevation of the stations are substantially different in some cases (Table 1). Grid cells 10 and 11 (90 E, 30 N and 100 E, 30 N) have the greatest difference of 902 m and 942 m, respectively. These cells also have the largest bias of 6.94 C and 5.17 C, respectively. Similarly, grid cell 6 (105 E, 35 N) has the smallest elevation difference ( 35 m) and virtually no temperature bias ( C). [14] In support, the correlation between the 14 grid cells temperature bias (station minus ERA-40 temperature) with the 14 grid cells elevation bias (mean station minus DEM elevation) is 0.94, indicating that most of the temperature bias is due to the elevation differences. While presumably some of the bias arises in that we are comparing 2-m temperatures with screen level values (somewhat closer to the surface), this effect would appear to be minor. As further evidence, applying a moist adiabatic lapse rate of 6 C/1000 m to the elevation differences removes nearly all of the biases (not shown). Recall that ERA-40 also uses four additional variables to capture terrain variability: angle, 3.2. Assimilated Versus Unassimilated Stations [16] The obvious caveat to these results is that ERA-40 assimilates surface observations. The good correspondence between station temperatures and ERA-40 temperatures may therefore not be surprising. As discussed earlier, we created time series within each grid cell by averaging the stations that would have been given strong weights by ERA-40 (assuming they were in fact available to ECMWF) based on ERA-40 s elevation and e-folding criteria (Figure 3). We also compiled time series by averaging the stations within each grid cell that would not have been assimilated. Correlations between the assimilated station time series with the ERA-40 record ought to be higher, if those stations were indeed included. [17] In all 14 grid cells, ERA-40 correlations with both assimilated and unassimilated time series are of approximately equal strength (Table 2). In fact, time series corresponding to the unassimilated data are in general slightly more strongly correlated with ERA-40. Since one would expect the opposite (assimilated stations more strongly correlated with ERA-40), this suggests that the stations classified by us as assimilated were either not available for ERA-40, or not assimilated. The finding that, Figure 5. Standardized station (black) and ERA-40 (gray) time series for grid cell 10 (90 E, 30 N) illustrating strong correspondence (R = 0.87). 5of9

6 Table 3. Midseason Monthly Temperature Bias and Temporal Correlations Between ERA-40 and Station Data for the 14 Grid Cells a Grid Cell Station Data-ERA-40 Bias, C January April July October Station Data-ERA-40 Correlations January April July October a Correlations are all statistically significant (a < 0.05). in general, unassimilated stations correlate slightly more strongly with ERA-40 is potentially a statistical artifact, arising because unassimilated stations within each grid cell are more numerous. The greater sample size provides a more representative depiction of grid cell-wide temperatures, even if those stations were not assimilated. [18] In grid cells 7, 10, and 11, none of the station data would have been assimilated (although as with other grid cells, some weight would be given to stations outside the cell boundaries). Correlations in those grid cells are nevertheless statistically significant and strong (Table 2). This provides further evidence that ERA-40 provides a good depiction of temperature variability on the TP even where station data are sparse or absent. From hereon, all stations are used regardless of whether they may, or may not, have been assimilated in ERA Plateau-Wide Temperature Climatology [19] Here, we examine the ERA-40 annual and seasonal temperature climatology. Annual mean temperatures across virtually all of the TP are below 0 C (Figure 6). The western TP, with elevations generally above 5000 m, is characterized by means around 10 C. Temperatures on the eastern TP are higher, with those at lower elevations surrounding the TP much higher. Means of approximately 15 C are found north and east of the plateau. Values over 25 C are found to the south. As would be expected, the margin of the TP is characterized by sharp temperature gradients. [20] The spring and autumn climatologies closely resemble the annual mean pattern (Figure 7). However, during winter the TP is characterized by temperatures around 25 Cinthe west, and 15 C in the east. Gradients along the TP s periphery are still strong, albeit weaker than for spring and autumn, with temperatures to the south and east of the TP >20 C and 5 C, respectively. The gradient to the north is also weaker, where temperatures in the deserts are around 5 C. In the summer season, the entire TP s temperatures rise above 0 C. The east warms up to between 5 10 C, and high elevations of the west warm up to 0 5 C. The deserts north of the TP as well as regions to the east warm up to approximately25 C.AreassouthoftheTParequitewarm(30 35 C) Temperature Trends [21] The station data indicate positive and statistically significant annual temperature trends in the northeastern TP (grid cells 1 6), as well as the southeastern TP (grid cells 11 and 13). These annual trends range from C decade 1 (Table 4), indicating a warming of 0.86 to 1.3 C for the 43-year period 1958 to Annual trends are largely determined by changes during autumn and winter, when positive trends range from C decade 1 (a warming of 0.86 to 2.6 C for ). No significant changes are observed in the central TP (grid cells 7 10), or the extreme southeast (grid cells 12 and 14). [22] Trend assessment using reanalysis data is dangerous due to changes in the amount and quality of assimilation data. For example, a major change occurred in the late 1970s, when the modern satellite data stream came online (although some satellite data were available back to 1972). Before this time, rawinsondes represent the primary data source. Such changes can result in nonclimatic jumps and hence spurious trends. However, we see no pronounced features in the reanalysis time series that are not present in the station data (see Figures 4, 5, and 8). Nonetheless, ERA-40 trends should be interpreted cautiously. Similarly, station trends must be interpreted cautiously because stations are biased toward low-lying populated regions. While neither record is ideal for assessing long-term trends on the TP, they represent the only available data sources. [23] The annual ERA-40 time series indicate a significant positive trend in only four of the grid cells: the northeastern Figure 6. Annual temperature climatology of the TP, as depicted by the m DEM height contours. 6of9

7 Figure 7. Seasonal temperature climatology of the TP: (top left) winter, (top right) spring, (bottom left) summer, and (bottom right) autumn. TP (grid cells 5 6), the Lhasa grid cell (10), and the extreme southeastern grid cell (14). These are of similar magnitude to the station trends (Table 4). Grid cell 7 has a significant negative trend. The trends at grid cells 6 and 14 are observed across all seasons, while the trend of grid cell 5 is dominated by the cold season. Significant positive winter trends only are evident in some of the northeastern grid cells. Overall, the correspondence between station and ERA-40 trends is poor Plateau-Wide Comparisons and Climate Change [24] We next assess the results aggregated for the entire TP. The aggregate long-term temperature means for the two time series representing the TP (the 54 ERA-40 grid cells Table 4. The Temperature Trends for the 14 Individual Grid Cells, Averaged From All 14 Grid Cells, the Entire TP (All), and the Eastern and Western Plateaua Grid Cell Annual Winter Spring Summer Autumn Station 0.05 Annual Winter Spring Summer Autumn ERA Temperature Trend, C decade Temperature Trend, C decade a Statistically significant trends (a < 0.05) are in bold. 7 of Average All Eastern Western

8 (with the exception of the western TP). Again, none of the plateau-wide ERA-40 trends are significant. This suggests that, if the ERA-40 trends are reliable, the climate change reported for the TP is valid only for low-lying sites near meteorological stations and hence the populated regions. Figure 8. Average annual station (black) and ERA-40 (gray) time series for the Tibetan Plateau. The station time series shows a statistically significant positive temperature trend of C decade 1, while no trend is evident for ERA-40 temperatures. The two time series are correlated at R = covering the TP, and the 97 stations located on the TP above 2000 m; see Figure 2) are different by 3.9 C, with the ERA-40 temperatures lower ( 0.5 C) than the station data (+3.4 C). This difference is attributable to the elevation differences discussed earlier. The two standardized time series are correlated at R = 0.87, again indicating very good temporal correspondence despite the fact that nearly all of the stations are concentrated on the eastern TP. [25] To assess the impact of the uneven station distribution, we divide the grid cells and stations into east (Figure 2, red) versus west (Figure 2, blue) along E longitude. In the data sparse west there are only 21 stations, mostly in the southern TP, of which only three might have been assimilated in ERA-40 (Figure 3). In this region, the mean temperature bias is 5.4 C, with ERA-40 and observed means of 2.3 C and 3.1 C, respectively. The correlation between the standardized time series for the west is lower than for the plateau-wide case, but still strong and significant (R = 0.74). In the eastern TP, where elevations are lower, the mean bias is smaller (2.0 C), with ERA-40 and observed means of 1.5 C and 3.5 C, respectively. Not surprising given the greater number of stations (76), the correlation for the eastern plateau is stronger (R = 0.95). This also suggests that the plateau-wide correlation of 0.87 would be even higher if more stations were available in the west. [26] In light of reported climate change on the TP, we calculated the long-term trends for each temperature product. As assessed for the entire TP, the stations show a positive trend for the period (Figure 8), statistically significant at + C decade 1. However, ERA-40 shows no significant trend. This discrepancy could be reconciled if the plateau-wide trend is dominated by a strong positive trend in the east (where most of the stations are located) and a negative trend in the west. However, trends computed separately for the eastern and western plateau produce the same result. The station based trends for the east and west are a statistically significant + C decade 1 and C decade 1, respectively, while there are again no significant trends in ERA-40 (Table 4). The plateau-wide station trends are dominated by the cold season and the east versus west trends are significant for the annual case only 4. Discussion and Conclusions [27] ERA-40 2-m temperatures are compared to station data to assess the reliability of this reanalysis product over the topographically complex, data-sparse TP. We find good agreement with the station records. Temporal correlations are high. While there is an apparent low bias in ERA-40 temperatures, this is primarily due to elevation differences. Some surface observations have been assimilated in ERA-40 and our comparisons are therefore not entirely independent. However, we separately analyze those station records that would likely not have been assimilated by ERA-40 and still find good agreement. Since station records, concentrated in the east, represent isolated point values in nonrepresentative areas of the plateau, ERA-40 arguably provides a better assessment of large-scale temperature variability across the plateau. [28] On the basis of ERA-40, virtually the entire TP is characterized by negative temperatures on an annual basis, with the eastern TP between 0 C and 5 C, and the western TP approximately 10 C. The spring and autumn climatologies are similar to the annual case, but winter temperatures range from 15 C in the east to 25 C in the west. The summer warm-up of C across the TP results in temperatures of 5 10 C in the east, and 0 5 C in the west. [29] Time series based on aggregating all station data on the TP show a statistically significant positive trend of C decade 1, as also reported by Liu and Chen [2000]. The stations, however, are biased toward low-lying populated regions in the eastern and southern TP. No trends are evident in the ERA-40 data for the plateau as a whole. [30] Long-term trends based on reanalysis products must be interpreted cautiously, and therefore so must this trend discrepancy. However, temperature trends in the station data seem to be confined locally, to populated low-lying areas. A potential explanation for the difference between reanalysis and station trends is the extensive local and regional land use change that has occurred across the TP over the last 50 years. [31] Over 62% of the plateau is used for agriculture: farmland, forest, and orchards, and a majority (80%) is used for livestock grazing [Du et al., 2004]. As the nomadic husbandry practices on the TP are at a primitive level, all livestock depend on the natural grasslands and no animal feed is imported. Over the last 30 years, livestock numbers across the TP have increased more than 200% due to inappropriate land management practices and are now at levels that far exceed the carrying capacity of the region [Du et al., 2004]. Consequently, overgrazing has caused land degradation and desertification at an alarming rate [Zhu and Li, 2000; Zeng et al., 2003]. The consequences of these changes on the local and regional climate have not been assessed for the TP. However, in other parts of the world, land degradation due to overgrazing has been shown to cause significant local temperature increases [e.g., Balling et al., 1998]. 8of9

9 [32] Urbanization, which can result in 8 11 C higher temperatures than in surrounding rural areas [e.g., Brandsma et al., 2003], has also occurred extensively on the TP. The construction of a gas pipeline in the 1970s and highway expansion projects in the early 1980s have resulted in a dramatic population influx from other parts of China, contributing to both urbanization and a changed landscape. For instance, the original Tibetan section of Lhasa (i.e., the pre-1950 Lhasa) now only comprises 4% of the city, suggesting a 2400% increase in size over the last 50 years. Similar population increases have occurred at other locations across the TP; however, even villages and small towns can exhibit a strong urban heat island effect, especially during the cold season. As in high-latitude regions [e.g., Hinkel et al., 2003], in the high-altitude environment of the TP this will be accompanied by earlier snowmelt and increased thickness of the thawed layer, resulting in permafrost degradation and thus a further altered land surface. The population increases have also resulted in more civil and engineering construction such as airports and communication networks. Construction of the Qinghai-Xizang railroad is underway, to be completed in 2007, which will provide additional changes to the Tibetan landscape as well as further encourage population increases. [33] Widespread land degradation due to deforestation, overgrazing, and cultivation of grasslands, and heightened urbanization are therefore plateau-wide problems, especially in low-lying areas and on hillsides. We submit that these local changes are reflected in station temperature records. Kalnay and Cai [2003] have shown that differences between surface data and NCEP-NCAR reanalysis data, because they incorporate no surface observations, can be successfully used to assess surface changes such as urbanization and land use. Although some surface observations over the TP are assimilated in ERA-40, potentially introducing these biases to the resulting temperature fields, the lack of trends suggests that ERA-40 temperatures are largely free of such surface contamination. We stress that other explanations could also be invoked to account for the differences in trends. For instance, ERA-40 s treatment of boundary layer mixing processes could dampen the surface temperature contamination, resulting in a reduction in trends. It is the focus of our ongoing research to quantify these effects and to test our land use change hypothesis. [34] Acknowledgments. ERA-40 2-m temperature data were obtained from ECMWF s public web server ( We thank Andrew Slater and Roy Jenne for their helpful discussions. This work was partly supported by the International Arctic Research Center, University of Alaska Fairbanks, under the auspices of NSF cooperative agreement number OPP , the Institute of Plateau Meteorology at Chengdu, China Meteorological Administration, the U.S. National Science Foundation through OPP , and the NOAA Arctic Research Office for ERA-40 validation. References Aizen, V. B., M. Aizen, J. M. Melack, and J. Dozier (1997), Climatic and hydrologic changes in the Tien Shan, central Asia, J. Clim., 10(6), Balling, R. C., Jr., J. M. Klopatek, M. L. Hildebrandt, C. K. Moritz, and C. J. Watts (1998), Impacts of land degradation on historical temperature records from the Sonoran desert, Clim. Change, 40(3 4), Beniston, M., H. F. Diaz, and R. S. Bradley (1997), Climatic change at high elevation sites: An overview, Clim. Change, 36(3 4), Brandsma, T., G. P. Konnen, and H. R. A. Wessels (2003), Empirical estimation of the effect of urban heat advection on the temperature series of De Bilt (the Netherlands), Int. J. Climatol., 23(7), Chen, B., W. C. Chao, and X. Liu (2003), Enhanced climatic warming in the Tibetan Plateau due to doubling CO 2 : A model study, Clim. Dyn., 20(4), Diaz, H. F., and R. S. Bradley (1997), Temperature variations during the last century at high elevation sites, Clim. Change, 36(3 4), Du, M., S. Kawashima, S. Yonemura, X. Zhang, and S. Chen (2004), Mutual influence between human activities and climate change in the Tibetan Plateau during recent years, Global Planet. Change, 41(3 4), European Centre for Medium-Range Weather Forecasts (ECMWF) (2002), Workshop on re-analysis, 5 9 November 2001, ERA-40 Proj. Rep. Ser. 3, 443 pp., Reading, U. K. Hinkel, K. M., F. E. Nelson, A. E. Klene, and J. H. Bell (2003), The urban heat island in winter at Barrow, Alaska, Int. J. Climatol., 23(15), Hsu, H.-H., and X. Liu (2003), Relationship between the Tibetan Plateau heating and east Asian summer monsoon rainfall, Geophys. Res. Lett., 30(20), 2066, doi: /2003gl Jones, P. D., and A. Moberg (2003), Hemispheric and large-scale surface air temperature variations: An extensive revision and an update to 2001, J. Clim., 16(2), Jones, P. D., S. C. B. Raper, B. S. G. Cherry, C. M. Goodess, T. M. L. Wigley, B. Santer, P. M. Kelly, R. S. Bradley, and H. F. Diaz (1991), An updated global grid point surface air temperature anomaly data set: , NDP-020/R1, Carbon Dioxide Inf. Anal. Cent., Oak Ridge Natl. Lab., Oak Ridge, Tenn. Kalnay, E., and M. Cai (2003), Impact of urbanization and land-use change on climate, Nature, 423(6939), Kalnay, E., et al. (1996), The NCEP/NCAR 40-year reanalysis project, Bull. Am. Meteorol. Soc., 77(3), Liu, X., and B. Chen (2000), Climatic warming in the Tibetan Plateau during recent decades, Int. J. Climatol., 20(14), Vose, R. S., R. L. Schmoyer, P. M. Steurer, T. C. Peterson, R. Heim, T. R. Karl, and J. K. Eischeid (1992), The global historical climatology network: Long-term monthly temperature, precipitation, sea level pressure, and station pressure data, NDP-041, Carbon Dioxide Inf. Anal. Cent., Oak Ridge Natl. Lab., Oak Ridge, Tenn. Zeng, Y., Z. Feng, and G. Cao (2003), Land cover change and its environmental impact in the upper reaches of the Yellow River, northeast Qinghai- Tibetan Plateau, Mt. Res. Dev., 23(4), Zhu, L., and B. Li (2000), Natural hazards and environmental issues, in Mountain Genecology and Sustainable Development of the Tibetan Plateau, edited by D. Zheng, Q. Zhang, and S. Wu, pp , Springer, New York. O. W. Frauenfeld, M. C. Serreze, and T. Zhang, Cooperative Institute for Research in Environmental Sciences (Division of Cryospheric and Polar Processes), National Snow and Ice Data Center, University of Colorado, 449 UCB, Boulder, CO , USA. (oliverf@kryos.colorado.edu) 9of9