|Working Group I: The Scientific Basis|
|Other reports in this collection|
2.2.4 How do Surface and Upper Air Temperature Variations
In Figure 2.12 we display the surface, tropospheric and strato-spheric temperature variations using representative data sets from those described above. Trend values (°C/decade) are shown in Table 2.3 with 95% confidence intervals, which in part represent uncertainties due to temporal sampling, not those due to measurement error (see below). The effect of explosive volcanic events (Agung, 1963; El Chichon, 1982; and Mt. Pinatubo, 1991) is evident in Figure 2.12, as is a relative shift to warmer temperatures in the lower troposphere compared to the surface in the late 1970s, followed by large variations in both due to ENSO (particularly in 1998). After the shift in the late 1970s, the overall tropospheric temperature trend is near zero but the surface has warmed (see Figure 2.12a and Table 2.3).
Global variations and trends in the lower stratosphere are temporally more
coherent than in the troposphere (Figure 2.12b),
though the warming effects due to the volcanic eruptions are clearly evident.
For the period 1958 to 2000, all stratospheric data sets except NCEP 4, which
contains erroneous trends, show significant negative trends (Table
2.3). Note that MSU 4, and simulations of MSU 4 (UKMO 4 and NCEP 4), include
a portion of the upper troposphere below 100 hPa and so are expected to show
less negative trends than those measuring at higher altitudes (e.g., the 100
to 50 hPa layers in Table 2.3 and the SSU in Figure
Between 1979 and 2000, the magnitude of trends between the surface and MSU 2LT is generally most similar in many parts of the Northern Hemisphere extra-tropics (20°N to pole) where deep vertical mixing is often a characteristic of the troposphere. For example, the northern extra-tropical trends for the surface and MSU 2LT were 0.28 and 0.21°C/decade, respectively, and over the North American continent trends were 0.27 ± 0.24 and 0.28 ± 0.23°C/decade, respectively, with an annual correlation of 0.92. Over Europe the rates were 0.38 ± 0.36 and 0.38 ± 0.30°C/decade, respectively. Some additional warming of the surface relative to the lower troposphere would be expected in the winter half year over extra-tropical Asia (whole year warming rates of 0.35 ± 0.20 and 0.18 ± 0.18°C/decade, respectively), consistent with the vertical temperature structure of the increased positive phase of the Arctic Oscillation (Thompson et al., 2000a, Figure 2.30). The vertical structure of the atmosphere in marine environments, however, generally reveals a relatively shallow inversion layer (surface up to 0.7 to 2 km) which is somewhat decoupled from the deep troposphere above (Trenberth et al., 1992; Christy, 1995; Hurrell and Trenberth, 1996). Not only are local surface versus tropospheric correlations often near zero in these regions, but surface and tropospheric trends can be quite different (Chase et al., 2000). This is seen in the different trends for the period 1979 to 2000 in the tropical band, 0.10 ± 0.10 and -0.06 ± 0.16°C/decade, respectively (Table 2.3) and also in the southern extra-tropics where the trends are 0.08 ± 0.06 and -0.05 ± 0.08°C/decade, respectively. Trends calculated for the differences between the surface and the troposphere for 1979 to 2000 are statistically significant globally at 0.13 ± 0.06°C/decade, and even more so in the tropics at 0.17 ± 0.06°C/decade. Statistical significance arises because large interannual variations in the parent time-series are strongly correlated and so largely disappear in the difference time-series (Santer et al., 2000; Christy et al., 2001). However, as implied above, they are not significant over many extra-tropical regions of the Northern Hemisphere such as North America and Europe and they are also insignificant in some Southern Hemisphere areas. The sequence of volcanic eruption, ENSO events, and the trends in the Arctic Oscillation have all been linked to some of this difference in warming rates (Michaels and Knappenburg, 2000; Santer et al., 2000; Thompson et al., 2000a; Wigley, 2000) and do explain a part of the difference in the rates of warming (see Chapter 12).
The linear trend is a simple measure of the overall tendency of a time-series and has several types of uncertainty; temporal sampling uncertainty owing to short data sets, spatial sampling errors owing to incomplete spatial sampling, and various other forms of measurement error, such as instrument or calibration errors. Temporal sampling uncertainties are present even when the data are perfectly known because trends calculated for short periods are unrepresentative of other short periods, or of the longer term, due to large interannual to decadal variations. Thus confidence intervals for estimates of trend since 1979 due to temporal sampling uncertainty can be relatively large, as high as ± 0.2°C/decade below 300 hPa (Table 2.3, Santer et al., 2000). Accordingly, the period from 1979 to 2000 provides limited information on long-term trends, or trends for other 22-year periods.
Uncertainties arising from measurement errors due to the factors discussed
in Section 2.2.3, including incomplete spatial sampling,
can be substantial. One estimate of this uncertainty can be made from comparisons
between the various analyses in Table 2.3. For trends below
300 hPa, this uncertainty may be as large as ± 0.10°C/decade since
1979, though Christy et al. (2000) estimate the 95% confidence interval as ±
0.06°C for the MSU 2LT layer average. For example, Santer et al. (2000)
find that when the satellite observations from MSU 2LT are masked to match the
less than complete global coverage of the surface observations during the past
few decades, the differences in the trends between the surface and the troposphere
are reduced by about one third.
Other reports in this collection