3.2.2.1 Land-Surface Air Temperature

Figure 3.1 shows annual global land-surface air temperatures, relative to the period 1961 to 1990, from the improved analysis (CRU/Hadley Centre gridded land-surface air temperature version 3; CRUTEM3) of Brohan et al. 2006 [JoC, MoS] ). The long-term variations are in general agreement with those from the operational version of the Global Historical Climatology Network (GHCN) data set (National Climatic Data Center (NCDC); Smith and Reynolds, 2005 [JoC] ; Smith et al. 2005 [JoC] ), and with the National Aeronautics and Space Administration’s (NASA) Goddard Institute for Space Studies (GISS; Hansen et al., 2001 [JoC, ARC] and Lugina et al. 2005 [NPR] ) analyses( Figure 3.1 ). Most of the differences arise from the diversity of spatial averaging techniques. The global average for CRUTEM3 is a land-area weighted sum (0.68 × NH + 0.32 × SH). For NCDC it is an area-weighted average of the grid-box anomalies where available worldwide. For GISS it is the average of the anomalies for the zones 90°N to 23.6°N, 23.6°N to 23.6°S and 23.6°S to 90°S with weightings 0.3, 0.4 and 0.3, respectively, proportional to their total areas. For Lugina et al. 2005 [NPR] ) it is (NH + 0.866 × SH) / 1.866 because they excluded latitudes south of 60°S. As a result, the recent global trends are largest in CRUTEM3 and NCDC, which give more weight to the NH where recent trends have been greatest( Table 3.2 ).

Figure 3.1. Annual anomalies of global land-surface air temperature (°C), 1850 to 2005, relative to the 1961 to 1990 mean for CRUTEM3 updated from Brohan et al. 2006 [JoC, MoS] ). The smooth curves show decadal variations (see Appendix 3.A ). The black curve from CRUTEM3 is compared with those from NCDC ( Smith and Reynolds, 2005 [JoC] ; blue), GISS ( Hansen et al., 2001 [JoC, ARC] ; red) and Lugina et al. 2005 [NPR] ; green).

Table 3.2. Linear trends in hemispheric and global land-surface air temperatures, SST (shown in table as HadSST2) and Nighttime Marine Air Temperature (NMAT; shown in table as HadMAT1). Annual averages, with estimates of uncertainties for CRU and HadSST2, were used to estimate trends. Trends with 5 to 95% confidence intervals and levels of significance(bold:<1%; italic,1–5%) were estimated by Restricted Maximum Likelihood (REML; see Appendix 3.A ), which allows for serial correlation (first order autoregression AR1) in the residuals of the data about the linear trend. The Durbin Watson D-statistic (not shown) for the residuals, after allowing for first-order serial correlation, never indicates significant positive serial correlation.

Further, small differences arise from the treatment of gaps in the data. The GISS gridding method favours isolated island and coastal sites, thereby reducing recent trends, and Lugina et al. 2005 [NPR] ) also obtain reduced recent trends owing to their optimal interpolation method that tends to adjust anomalies towards zero where there are few observations nearby (see, e.g., Hurrell and Trenberth, 1999 [PoC, JoC, MoS, SRC] ). The NCDC analysis, which begins in 1880, is higher than CRUTEM3 by between 0.1°C and 0.2°C in the first half of the 20th century and since the late 1990 s. This is probably because its anomalies have been interpolated to be spatially complete: an earlier but very similar version (CRUTEM2v; Jones and Moberg, 2003 [JoC, SRC] ) agreed very closely with NCDC when the global averages were calculated in the same way ( Vose et al., 2005b [JoC, SRC] ). Differences may also arise because the numbers of stations used by CRUTEM3, NCDC and GISS differ (4,349, 7,230 and >7,200 respectively), although many of the basic station data are in common. Differences in station numbers relate principally to CRUTEM3 requiring series to have sufficient data between 1961 and 1990 to allow the calculation of anomalies ( Brohan et al., 2006 [JoC, MoS] ). Further differences may have arisen from differing homogeneity adjustments (see also Appendix 3.B.2).

Trends and low-frequency variability of large-scale surface air temperature from the ERA-40 reanalysis and from CRUTEM2v ( Jones and Moberg, 2003 [JoC, SRC] ) are in general agreement from the late 1970 s onwards ( Simmons et al., 2004 [JoC, SRC] ). When ERA-40 is sub-sampled to match the Jones and Moberg coverage, correlations between monthly hemispheric- and continental-scale averages exceed 0.96, although trends in ERA-40 are then 0.03°C and 0.07°C per decade (NH and SH, respectively) lower than Jones and Moberg 2003 [JoC, SRC] ). The ERA-40 reanalysis is more homogeneous than previous reanalyses (see Section 3.2.1 and Appendix 3.B.5.4) but is not completely independent of the Jones and Moberg data ( Simmons et al., 2004 [JoC, SRC] ). The warming trends continue to be greatest over the continents of the NH (see maps in Section 3.2.2.7 ,Figures 3.9 and 3.10 ), in line with the TAR. Issues of homogeneity of terrestrial air temperatures are discussed in Appendix 3.B.2.

Table 3.2 provides trend estimates from a number of hemispheric and global temperature databases. Warming since 1979 in CRUTEM3 has been 0.27°C per decade for the globe, but 0.33°C and 0.13°C per decade for the NH and SH, respectively. Brohan et al. 2006 [JoC, MoS] and Rayner et al. 2006 [JoC] ) (see Section 3.2.2.3 )provide uncertainties for annual estimates, incorporating the effects of measurement and sampling error, and uncertainties regarding biases due to urbanisation and earlier methods of measuring SST. These factors are taken into account, although ignoring their serial correlation. In Table 3.2 ,the effects of persistence on error bars are accommodated using a red noise approximation, which effectively captures the main influences. For more extensive discussion see Appendix 3.A

From 1950 to 2004, the annual trends in minimum and maximum land-surface air temperature averaged over regions with data were 0.20°C per decade and 0.14°C per decade, respectively, with a trend in diurnal temperature range (DTR) of –0.07°C per decade ( Vose et al., 2005a [JoC, SRC] ; Figure 3.2 ). This is consistent with the TAR where data extended from 1950 to 1993; spatial coverage is now 71% of the terrestrial surface instead of 54% in the TAR, although tropical areas are still under-represented. Prior to 1950, insufficient data are available to develop global-scale maps of maximum and minimum temperature trends. For 1979 to 2004, the corresponding linear trends for the land areas where data are available were 0.29°C per decade for both maximum and minimum temperature with no trend for DTR. Diurnal temperature range is particularly sensitive to observing techniques, and monitoring it requires adherence to GCOS monitoring principles ( GCOS, 2004 [NPR] ). A map of the trend of annual DTR over the period 1979 to 2004 ( Section 3.2.2.7 , Figure 3.11 )is discussed later in the chapter.

Figure 3.2. Annual anomalies of maximum and minimum temperatures and DTR (°C) relative to the 1961 to 1990 mean, averaged for the 71% of global land areas where data are available for 1950 to 2004 . The smooth curves show decadal variations (see Appendix 3.A ). Adapted from Vose et al. 2005a [JoC, SRC] ).

3.2.2.2 Urban Heat Islands and Land Use Effects

The modified land surface in cities affects the storage and radiative and turbulent transfers of heat and its partition into sensible and latent components (see Section 7.2 and Box 7.2 ). The relative warmth of a city compared with surrounding rural areas, known as the urban heat island (UHI) effect, arises from these changes and may also be affected by changes in water runoff, pollution and aerosols. Urban heat island effects are often very localised and depend on local climate factors such as windiness and cloudiness (which in turn depend on season), and on proximity to the sea. Section 3.3.2.4 discusses impacts of urbanisation on precipitation.

Many local studies have demonstrated that the microclimate within cities is on average warmer, with a smaller DTR, than if the city were not there. However, the key issue from a climate change standpoint is whether urban-affected temperature records have significantly biased large-scale temporal trends. Studies that have looked at hemispheric and global scales conclude that any urban-related trend is an order of magnitude smaller than decadal and longer time-scale trends evident in the series (e.g., Jones et al., 1990 [JoC, SRC] ; Peterson et al., 1999 [JoC, ARC] ). This result could partly be attributed to the omission from the gridded data set of a small number of sites (<1%) with clear urban-related warming trends. In a worldwide set of about 270 stations, Parker 2004 [JoC] , 2006 ) noted that warming trends in night minimum temperatures over the period 1950 to 2000 were not enhanced on calm nights, which would be the time most likely to be affected by urban warming. Thus, the global land warming trend discussed is very unlikely to be influenced significantly by increasing urbanisation ( Parker, 2006 [JoC] ). Over the conterminous USA, after adjustment for time-of-observation bias and other changes, rural station trends were almost indistinguishable from series including urban sites ( Peterson, 2003 [JoC, ARC] ; Figure 3.3 ,and similar considerations apply to China from 1951 to 2001 ( (Li et al., 2004 ) ). One possible reason for the patchiness of UHIs is the location of observing stations in parks where urban influences are reduced ( Peterson, 2003 [JoC, ARC] ). In summary, although some individual sites may be affected, including some small rural locations, the UHI effect is not pervasive, as all global-scale studies indicate it is a very small component of large-scale averages. Accordingly, this assessment adds the same level of urban warming uncertainty as in the TAR: 0.006°C per decade since 1900 for land, and 0.002°C per decade since 1900 for blended land with ocean, as ocean UHI is zero. These uncertainties are added to the cool side of the estimated temperatures and trends, as explained by Brohan et al. 2006 [JoC, MoS] ), so that the error bars in Section 3.2.2.4 ,Figures 3.6 and 3.7 and FAQ 3.1 ,Figure 1 are slightly asymmetric. The statistical significances of the trends in Table 3.2 and Section 3.2.2.4 , Table 3.3 take account of this asymmetry.

Figure 3.3. Anomaly (°C) time series relative to the 1961 to 1990 mean of the full US Historical Climatology Network (USHCN) data (red), the USHCN data without the 16% of the stations with populations of over 30,000 within 6 km in the year 2000 (blue), and the 16% of the stations with populations over 30,000 (green). The full USHCN set minus the set without the urban stations is shown in magenta. Both the full data set and the data set without the high-population stations had stations in all of the 2.5° latitude by 3.5° longitude grid boxes during the entire period plotted, but the subset of high-population stations only had data in 56% of these grid boxes. Adapted from Peterson and Owen 2005 [JoC, ARC] ).

( McKitrick and Michaels 2004 ) and de Laat and Maurellis 2006 [JoC] ) attempted to demonstrate that geographical patterns of warming trends over land are strongly correlated with geographical patterns of industrial and socioeconomic development, implying that urbanisation and related land surface changes have caused much of the observed warming. However, the locations of greatest socioeconomic development are also those that have been most warmed by atmospheric circulation changes (Sections 3.2.2.7 and 3.6.4 ), which exhibit large-scale coherence. Hence, the correlation of warming with industrial and socioeconomic development ceases to be statistically significant. In addition, observed warming has been, and transient greenhouse-induced warming is expected to be, greater over land than over the oceans( Chapter 10 ), owing to the smaller thermal capacity of the land.

Comparing surface temperature estimates from the NRA with raw station time series, Kalnay and Cai 2003 [JoC] ) concluded that more than half of the observed decrease in DTR in the eastern USA since 1950 was due to changes in land use, including urbanisation. This conclusion was based on the fact that the reanalysis did not explicitly include these factors, which would affect the observations. However, the reanalysis also did not explicitly include many other natural and anthropogenic effects, such as increasing greenhouse gases and observed changes in clouds or soil moisture ( Trenberth, 2004 [PoC, JoC, SRC] Vose et al. 2004 [JoC, SRC] ) showed that the adjusted station data for the region (for homogeneity issues, see Appendix 3.B.2) do not support Kalnay and Cai’s conclusions. Nor are Kalnay and Cai’s results reproduced in the ERA-40 reanalysis ( Simmons et al., 2004 [JoC, SRC] ). Instead, most of the changes appear related to abrupt changes in the type of data assimilated into the reanalysis, rather than to gradual changes arising from land use and urbanisation changes. Current reanalyses may be reliable for estimating trends since 1979 ( Simmons et al., 2004 [JoC, SRC] ) but are in general unsuited for estimating longer-term global trends, as discussed in Appendix 3.B.5.

Nevertheless, changes in land use can be important for DTR at the local-to-regional scale. For instance, land degradation in northern Mexico resulted in an increase in DTR relative to locations across the border in the USA ( Balling et al., 1998 [JoC] ), and agriculture affects temperatures in the USA ( Bonan, 2001 [JoC, ARC] ; Christy et al., 2006 [NotFound] ). Desiccation of the Aral Sea since 1960 raised DTR locally ( Small et al., 2001 [JoC] ). By processing maximum and minimum temperature data as a function of day of the week, Forster and Solomon 2003 [PoC, JoC, ARC] ) found a distinctive ‘weekend effect’ in DTR at stations examined in the USA, Japan, Mexico and China. The weekly cycle in DTR has a distinctive large-scale pattern with geographically varying sign, and strongly suggests an anthropogenic effect on climate, likely through changes in pollution and aerosols ( Jin et al., 2005 [JoC, SRC] ). Section 7.2 provides fuller discussion of the effects of land use changes.

3.2.2.3 Sea Surface Temperature and Marine Air Temperature

Most analyses of SST estimate the subsurface bulk temperature (i.e., the temperature in the uppermost few metres of the ocean), not the ocean skin temperature measured by satellites. For maximum resolution and data coverage, polar-orbiting infrared satellite data since 1981 can be used so long as the satellite ocean skin temperatures are adjusted to estimate bulk SST values through a calibration procedure (see e.g., Reynolds et al., 2002 [JoC] ; Rayner et al., 2003 [JoC] , 2006; Appendix 3.B.3). But satellite SST data alone have not been used as a major resource for estimating climate change because of their strong time-varying biases which are hard to completely remove, for example, as shown in Reynolds et al. 2002 [JoC] ) for the Pathfinder polar orbiting satellite SST data set ( Kilpatrick et al., 2001 [JoC] ). Figures 3.9 and 3.10 ( Section 3.2.2.7 )do, however, make use of spatial relationships based on adjusted satellite SST estimates after November 1981 to provide nearer-to-global coverage for the 1979 to 2005 period, ( and O’Carroll et al. 2006 ) ) have developed an analysis based on Along-Track Scanning Radiometers (ATSRs) with potential for the future. However, satellite data are unable to fill in estimates of surface temperature over or near sea ice areas.

Recent bulk SSTs estimated using ship and buoy data also have time-varying biases (e.g., Christy et al., 2001 [JoC, SRC] ; Kent and Kaplan, 2006 [MoS] ) that are larger than originally estimated by Folland et al. 1993 [PoC, JoC, SRC] ), but not large enough to prejudice conclusions about recent warming (see Appendix 3.B.3). As reported in the TAR, a combined physical-empirical method ( Folland and Parker, 1995 [PoC, JoC, SRC] ) is mainly used to estimate adjustments to ship SST data obtained up to 1941 to compensate for heat losses from uninsulated (mainly canvas) or partly insulated (mainly wooden) buckets. Details are given in Appendix 3.B.3.

The SST analyses of Rayner et al. 2003 [JoC] and Smith and Reynolds 2004 [JoC] ) are interpolated to fill missing data areas. The main problem for estimating climate variations in the presence of large data gaps is underestimation of change, as most interpolation procedures tend to bias the analysis towards the modern climatologies used in these data sets ( Hurrell and Trenberth, 1999 [PoC, JoC, MoS, SRC] ). To address non-stationary aspects, Rayner et al. 2003 [JoC] ) extracted the leading global covariance pattern, which represents long-term changes, before interpolating using reduced-space optimal interpolation (see Appendix 3.B.1); and Smith and Reynolds removed a smoothed, moving 15-year-average field before interpolating by a related technique.

Figure 3.4 ashows annual and decadally smoothed anomalies of global SST from the new, uninterpolated Hadley Centre SST data set version 2 (HadSST2) analysis ( Rayner et al., 2006 [JoC] ). Figure 3.4 aalso shows NMAT (referred to as HadMAT: Hadley Centre Marine Air Temperature data set), which is used to avoid daytime heating of ship decks ( Bottomley et al., 1990 [NPR] ). The global averages are ocean-area weighted sums (0.44 × NH + 0.56 × SH). The HadMAT analysis includes limited optimal interpolation ( Rayner et al., 2003 [JoC] ) and was chosen because of the demonstration by Folland et al. 2003 [PoC, JoC, SRC] ) of its skill in the sparsely observed South Pacific from the late 19th century onwards, but major gaps (e.g., the Southern Ocean) are not interpolated. Although HadMAT data have been corrected for warm biases during World War II they may still be too warm in the NH and too cool in the SH at that time( Figure 3.4 c,d). However, global HadSST2 and HadMAT generally agree well, especially after the 1880 s. The SST analysis in the TAR is included in Figure 3.4 a. The changes in SST since the TAR are generally fairly small, though the new SST analysis is warmer around 1880 and cooler in the 1950 s. The peak warmth in the early 1940 s is likely to have arisen partly from closely spaced multiple El Niño events ( Brönnimann et al., 2004 [NPR, JoC] ; see also Section 3.6.2 )and also due to the warm phase of the Atlantic Multi-decadal Oscillation (AMO; see Section 3.6.6 ). The HadMAT data generally confirm the hemispheric SST trends in the 20th century( Figure 3.4 c,d and Table 3.2 ). Overall, the SST data should be regarded as more reliable because averaging of fewer samples is needed for SST than for HadMAT to remove synoptic weather noise. However, the changes in SST relative to NMAT since 1991 in the tropical Pacific may be partly real ( Christy et al., 2001 [JoC, SRC] ). As the atmospheric circulation changes, the relationship between SST and surface air temperature anomalies can change along with surface fluxes. Interannual variations in the heat fluxes to the atmosphere can exceed 100 W m-2 locally in individual months, but the main prolonged variations occur with the El Niño-Southern Oscillation (ENSO), where changes in the central tropical Pacific exceed ±50 W m-2 for many months during major ENSO events ( Trenberth et al., 2002a [PoC, JoC, SRC] ).

Figure 3.4 bshows three time series of changes in global SST. Neither the HadSST2 series (as in Figure 3.4 a) nor the NCDC series include polar-orbiting satellite data because of possible time-varying biases that remain difficult to correct fully ( Rayner et al., 2003 [JoC] ). The Japanese series ( Ishii et al., 2005 [JoC, SRC] ; referred to as Centennial in-situ Observation-Based Estimates of SSTs (COBE-SST) from the Japan Meteorological Agency (JMA)) is also in situ except for the specification of sea ice. The warmest year globally in each SST record was 1998 (0.44°C, 0.38°C and 0.37°C above the 1961 to 1990 average for HadSST2, NCDC and COBE-SST, respectively). The five warmest years in all analyses have occurred after 1995 .

Figure 3.4. (a) Annual anomalies of global SST (HadSST2; red bars and blue solid curve), 1850 to 2005, and global NMAT (HadMAT, green curve), 1856 to 2005, relative to the 1961 to 1990 mean (°C) from the UK Meteorological Office (UKMO; Rayner et al., 2006 [JoC] ). The smooth curves show decadal variations (see Appendix 3.A ). The dashed black curve shows equivalent smoothed SST anomalies from the TAR. (b) Smoothed annual global SST anomalies, relative to 1961 to 1990 (°C), from HadSST2 (blue line, as in (a)), from NCDC ( Smith et al., 2005 [JoC] ; red line) and from COBE-SST ( Ishii et al., 2005 [JoC, SRC] ; green line). The latter two series begin later in the 19th century than HadSST2. (c,d) As in (a) but for the NH and SH showing only the UKMO series.

Understanding of the variability and trends in different oceans is still developing, but it is already apparent that they are quite different. The Pacific is dominated by ENSO and modulated by the Pacific Decadal Oscillation (PDO), which may provide ways of moving heat from the tropical ocean to higher latitudes and out of the ocean into the atmosphere ( Trenberth et al., 2002a [PoC, JoC, SRC] ), thereby greatly altering how trends are manifested. In the Atlantic, observations reveal the role of the AMO ( Folland et al., 1999 [NPR, PoC, SRC] ; Delworth and Mann, 2000 [PoC, JoC, MoS, ARC] ; Enfield et al., 2001 [JoC] ; Goldenberg et al., 2001 [JoC] ; Section 3.6.6 and Figure 3.33 ). The AMO is likely to be associated with the Thermohaline Circulation (THC), which transports heat northwards, thereby moderating the tropics and warming the high latitudes. In the Indian Ocean, interannual variability is small compared with the trend. Figure 3.5 presents latitude-time sections from 1900 for SSTs (from HadSST2) for the zonal mean across each ocean, filtered to remove fluctuations of less than about six years, including the ENSO signal. In the Pacific, the long-term warming is clearly evident, but punctuated by cooler episodes centred in the tropics, and no doubt linked to the PDO. The prolonged 1939 – 1942 El Niño shows up as a warm interval. In the Atlantic, the warming from the 1920 s to about 1940 in the NH was focussed on higher latitudes, with the SH remaining cool. This inter-hemispheric contrast is believed to be one signature of the THC ( Zhang and Delworth, 2005 [JoC, MoS, ARC] ). The subsequent relative cooling in the NH extratropics and the more recent intense warming in NH mid-latitudes was predominantly a multi-decadal variation of SST; only in the last decade is an overall warming signal clearly emerging. Therefore, the recent strong warming appears to be related in part to the AMO in addition to a global warming signal( Section 3.6.6 ). The cooling in the northwestern North Atlantic just south of Greenland, reported in the SAR, has now been replaced by strong warming (see also Section 3.2.2.7 ,Figures 3.9 and 3.10 ;also Figures 5.1 and 5.2 for ocean heat content). The Indian Ocean also reveals a poorly observed warm interval in the early 1940 s, and further shows the fairly steady warming in recent years. The multi-decadal variability in the Atlantic has a much longer time scale than that in the Pacific, but it is noteworthy that all oceans exhibit a warm period around the early 1940 s.

Figure 3.5. Latitude-time sections of zonal mean temperature anomalies (°C) from 1900 to 2005, relative to the 1961 to 1990 mean. Left panels: SST annual anomalies across each ocean from HadSST2 ( Rayner et al., 2006 [JoC] ). Right panels: Surface temperature annual anomalies for land (top, CRUTEM3) and land plus ocean (bottom, HadCRUT3). Values are smoothed with the 5-point filter to remove fluctuations of less than about six years (see Appendix 3.A ); and white areas indicate missing data.

3.2.2.4 Land and Sea Combined Temperature: Global, Northern Hemisphere, Southern Hemisphere and Zonal Means

Gridded data sets combining land-surface air temperature and SST anomalies have been developed and maintained by three groups: CRU with the UKMO Hadley Centre in the UK (HadCRUT3; Brohan et al., 2006 [JoC, MoS] ) and NCDC ( Smith and Reynolds, 2005 [JoC] ) and GISS ( Hansen et al., 2001 [JoC, ARC] ) in the USA. Although the component data sets differ slightly (see Sections 3.2.2.1 and 3.2.2.3 )and the combination methods also differ, trends are similar. Table 3.3 provides comparative estimates of linear trends. Overall warming since 1901 has been a little less in the NCDC and GISS analysis than in the HadCRUT3 analysis. All series indicate that the warmest five years have occurred after 1997, although there is slight disagreement about the ordering. The HadCRUT3 data set shows 1998 as warmest, while 2005 is warmest in NCDC and GISS data. Thus the year 2005, with no El Niño, was about as warm globally as 1998 with its major El Niño effects. The GISS analysis of 2005 interpolated the exceptionally warm conditions in the extreme north of Eurasia and North America over the Arctic Ocean (see Figure 3.5 ). If the GISS data for 2005 are averaged only south of 75°N, then 2005 is cooler than 1998 . In addition, there were relatively cool anomalies in 2005 in HadCRUT3 in parts of Antarctica and the Southern Ocean, where sea ice coverage (see Chapter4 )has not declined.

Table 3.3. Linear trends (°C per decade) in hemispheric and global combined land-surface air temperatures and SST. Annual averages, along with estimates of uncertainties for CRU/UKMO (HadCRUT3), were used to estimate trends. For CRU/UKMO, global annual averages are the simple average of the two hemispheres. For NCDC and GISS the hemispheres are weighted as in Section 3.2.2.1 .Trends are estimated and presented as in Table 3.2. R2 is the squared trend correlation (%). The Durbin Watson D-statistic (not shown) for the residuals, after allowing for first-order serial correlation, never indicated significant positive serial correlation, and plots of the residuals showed virtually no long-range persistence.

Hemispheric and global series based on Brohan et al. 2006 [JoC, MoS] ) are shown in Figure 3.6 and tropical and polar series in Figure 3.7 .Owing to the sparsity of SST data, the polar series are for land only. The recent warming is strongest in the NH extratropics, while El Niño events are clearly evident in the tropics, particularly the 1997 – 1998 event that makes 1998 the warmest year in HadCRUT3. The warming over land in the Arctic north of 65°N( Figure 3.7 )is more than double the warming in the global mean from the 19th century to the 21st century and also from about the late 1960 s to the present. In the arctic series, 2005 is the warmest year. A slightly longer warm period, almost as warm as the present, was observed from the late 1920 s to the early 1950 s. Although data coverage was limited in the first half of the 20th century, the spatial pattern of the earlier warm period appears to have been different from that of the current warmth. In particular, the current warmth is partly linked to the Northern Annular Mode (NAM; see Section 3.6.4 )and affects a broader region ( Polyakov et al., 2003 [JoC] ). Temperatures over mainland Antarctica (south of 65°S) have not warmed in recent decades ( Turner et al., 2005 [JoC] ), but it is virtually certain that there has been strong warming over the last 50 years in the Antarctic Peninsula region ( Turner et al., 2005 [JoC] ; see the discussion of changes in the Southern Annular Mode (SAM) and Figure 3.32 in Section 3.6.5 ).

Figure 3.6. Global and hemispheric annual combined land-surface air temperature and SST anomalies (°C) (red) for 1850 to 2006 relative to the 1961 to 1990 mean, along with 5 to 95% error bar ranges, from HadCRUT3 (adapted from Brohan et al., 2006 [JoC, MoS] ). The smooth blue curves show decadal variations (see Appendix 3.A ).

Figure 3.7. Annual temperature anomalies (°C) up to 2005, relative to the 1961 to 1990 mean (red) with 5 to 95% error bars. The tropical series (middle) is combined land-surface air temperature and SST from HADCRUT3 (adapted from Brohan et al., 2006 [JoC, MoS] ). The polar series (top and bottom) are land-only from CRUTEM3, because SST data are sparse and unreliable in sea ice zones. The smooth blue curves show decadal variations (see Appendix 3.A ).

3.2.2.5 Consistency between Land and Ocean Surface Temperature Changes

The course of temperature change over the 20th century, revealed by the independent analysis of land air temperatures, SST and NMAT, is generally consistent( Figure 3.8 ). Warming occurred in two distinct phases ( 1915 – 1945 and since 1975 ), and it has been substantially stronger over land than over the oceans in the later phase, as shown also by the trends in Table 3.2 .The land component has also been more variable from year to year (compare Figures 3.1 and 3.4 a,c,d). Much of the recent difference between global SST (and NMAT) and global land air temperature trends has arisen from accentuated warming over the continents in the mid-latitude NH( Section 3.2.2.7 ,Figures 3.9 and 3.10 ). This is likely to be related to greater evaporation and heat storage in the ocean, and in particular to atmospheric circulation changes in the winter half-year due to the North Atlantic Oscillation (NAO)/NAM (see discussion in Section 3.6.4 ). Accordingly the differences between NH and SH temperatures follow a course similar to the plot of land air temperature minus SST shown in Figure 3.8 .

Figure 3.8. Annual anomalies (°C) of global average SST (blue curve, begins 1850 ), NMAT (green curve, begins 1856 ) and land-surface air temperature (red curve, begins 1850 ) to 2005, relative to their 1961 to 1990 means ( Brohan et al., 2006 [JoC, MoS] ; Rayner et al., 2006 [JoC] ). The smooth curves show decadal variations (see Appendix 3.A ). Inset shows the smoothed differences between the land-surface air temperature and SST anomalies (i.e., red minus blue).

3.2.2.6 Temporal Variability of Global Temperatures and Recent Warming

The standard deviation of the HadCRUT3 annual average temperatures for the globe for 1850 to 2005 shown in Figure 3.6 is 0.24°C. The greatest difference between two consecutive years in the global average since 1901 is 0.29°C between 1976 and 1977, demonstrating the importance of the 0.75°C and 0.74°C temperature increases (the HadCRUT3 linear trend estimates for 1901 to 2005 and 1906 to 2005, respectively) in a centennial time-scale context. However, both trends are small compared with interannual variations at one location, and much smaller than day-to-day variations( Table 3.1 ).

The principal conclusion from the three global analyses is that the global average surface temperature trend has very likely been slightly more than 0.65°C ± 0.2°C over the period from 1901 to 2005 ( Table 3.3 ), a warming greater than any since at least the 11th century (see Chapter6 ). A HadCRUT3 linear trend over the 1906 to 2005 period yields a warming of 0.74°C ± 0.18°C, but this rate almost doubles for the last 50 years (0.64°C ± 0.13°C for 1956 to 2005; see FAQ 3.1 ). Clearly, the changes are not linear and can also be characterised as level prior to about 1915, a warming to about 1945, levelling out or even a slight decrease until the 1970 s, and a fairly linear upward trend since then( Figure 3.6 and FAQ 3.1 ). Considered this way, the overall warming from the average of the first 50-year period ( 1850 – 1899 ) through 2001 to 2005 is 0.76°C ± 0.19°C. Clearly, the world’s surface temperature has continued to increase since the TAR and the trend when computed in the same way as in the TAR remains 0.6°C over the 20th century. In view of Section 3.2.2.2 and the dominance of the globe by ocean, the influence of urbanisation on these estimates is estimated to be very small. The last 12 complete years ( 1995 – 2006 ) now contain 11 of the 12 warmest years since 1850, the earliest year for which comparable records are available. Only 1996 is not in this list – replaced by 1990 . 2002 to 2005 are the 3rd, 4th, 5th and 2nd warmest years in the series, with 1998 the warmest in HadCRUT3 but with 2005 and 1998 switching order in GISS and NCDC. The HadCRUT3 surface warming trend over 1979 to 2005 was more than 0.16°C per decade, that is, a total warming of 0.44°C ± 0.12°C (the error bars overlap those of NCDC and GISS). During 2001 to 2005, the global average temperature anomaly has been 0.44°C above the 1961 – 1990 average. The value for 2006 is close to the 2001 to 2005 average.

3.2.2.7 Spatial Trend Patterns

Figure 3.9 illustrates the spatial patterns of annual surface temperature changes for 1901 to 2005 and 1979 to 2005, and Figure 3.10 shows seasonal trends for 1979 to 2005 . All maps clearly indicate that differences in trends between locations can be large, particularly for shorter time periods. For the century-long period, warming is statistically significant over most of the world’s surface with the exception of an area south of Greenland and three smaller regions over the southeastern USA and parts of Bolivia and the Congo basin. The lack of significant warming at about 20% of the locations ( Karoly and Wu, 2005 [JoC, ARC] ), and the enhanced warming in other places, is likely to be a result of changes in atmospheric circulation (see Section 3.6 ). Warming is strongest over the continental interiors of Asia and northwestern North America and over some mid-latitude ocean regions of the SH as well as southeastern Brazil. In the recent period, some regions have warmed substantially while a few have cooled slightly on an annual basis( Figure 3.9 ). Southwest China has cooled since the mid-20th century ( (Ren et al., 2005 ) ), but most of the cooling locations since 1979 have been oceanic and in the SH, possibly through changes in atmospheric and oceanic circulation related to the PDO and SAM (see discussion in Section 3.6.5 ). Warming dominates most of the seasonal maps for the period 1979 onwards, but weak cooling has affected a few regions, especially the mid-latitudes of the SH oceans, but also over eastern Canada in spring, possibly in relation to the strengthening NAO (see Section 3.6.4 , Figure 3.30 ). Warming in this period was strongest over western North America, northern Europe and China in winter, Europe and northern and eastern Asia in spring, Europe and North Africa in summer and northern North America, Greenland and eastern Asia in autumn( Figure 3.10 ).

Figure 3.9. Linear trend of annual temperatures for 1901 to 2005 (left; °C per century) and 1979 to 2005 (right; °C per decade). Areas in grey have insufficient data to produce reliable trends. The minimum number of years needed to calculate a trend value is 66 years for 1901 to 2005 and 18 years for 1979 to 2005 . An annual value is available if there are 10 valid monthly temperature anomaly values. The data set used was produced by NCDC from Smith and Reynolds 2005 [JoC] ). Trends significant at the 5% level are indicated by white + marks.

Figure 3.10. Linear trend of seasonal MAM, JJA, SON and DJF temperature for 1979 to 2005 (°C per decade). Areas in grey have insufficient data to produce reliable trends. The minimum number of years required to calculate a trend value is 18. A seasonal value is available if there are two valid monthly temperature anomaly values. The dataset used was produced by NCDC from Smith and Reynolds 2005 [JoC] ). Trends significant at the 5% level are indicated by white + marks.

No single location follows the global average, and the only way to monitor the globe with any confidence is to include observations from as many diverse places as possible. The importance of regions without adequate records is determined from complete model reanalysis fields ( Simmons et al., 2004 [JoC, SRC] ). The importance of the missing areas for hemispheric and global averages is incorporated into the errors bars in Figure 3.6 (see Brohan et al., 2006 [JoC, MoS] ). Error bars are generally larger in the more data-sparse SH than in the NH; they are larger before the 1950 s and largest of all in the 19th century.

Figure 3.11 shows annual trends in DTR from 1979 to 2004 . The decline in DTR since 1950 reported in the TAR has now ceased, as confirmed by Figure 3.2 .Since 1979, daily minimum temperature increased in most areas except western Australia and southern Argentina, and parts of the western Pacific Ocean; and daily maximum temperature also increased in most regions except northern Peru, northern Argentina, northwestern Australia, and parts of the North Pacific Ocean ( Vose et al., 2005a [JoC, SRC] ). The changes reported here appear inconsistent with Dai et al. 2006 [Ambiguous] ) who reported decreasing DTR in the USA, but this arises partly because Dai et al. 2006 [Ambiguous] ) included the high DTR years 1976 to 1978 . Furthermore, Figure 3.11 is supported by many other recent regional-scale analyses.

Changes in cloud cover and precipitation explained up to 80% of the variance in historical DTR series for the USA, Australia, mid-latitude Canada and the former Soviet Union during the 20th century ( Dai et al., 1999 [PoC, JoC, SRC] ). Cloud cover accounted for nearly half of the change in the DTR in Fennoscandia during the 20th century ( Tuomenvirta et al., 2000 [JoC] ). Variations in atmospheric circulation also affect DTR. Changes in the frequency of certain synoptic weather types resulted in a decline in DTR during the cold half-year in the Arctic ( Przybylak, 2000 [JoC] ). A positive phase of the NAM (see Section 3.6.4 )is associated with increased DTR in the northeastern USA and Canada ( Wettstein and Mearns, 2002 [JoC, ARC] ). Variations in sea level pressure patterns and associated changes in cloud cover partially accounted for increasing trends in cold-season DTR in the northwestern USA and decreasing trends in the south-central USA ( Durre and Wallace, 2001 [JoC] ). The relationship between DTR and anthropogenic forcings is complex, as these forcings can affect atmospheric circulation, as well as clouds through both greenhouses gases and aerosols.

Figure 3.11. Linear trend in annual mean DTR for 1979 to 2004 (°C per decade). Grey regions indicate incomplete or missing data (after Vose et al., 2005a [JoC, SRC] ).

3.3.2.1 Global Land Areas

Trends in global annual land precipitation were analysed using data from the GHCN, using anomalies with respect to the 1981 to 2000 base period ( Vose et al., 1992 [NPR, SRC] ; Peterson and Vose, 1997 [JoC, SRC] ). The observed GHCN linear trend( Figure 3.12 )over the 106-year period from 1900 to 2005 is statistically insignificant, as is the CRU linear trend up to 2002 ( Table 3.4 b). However, the global mean land changes( Figure 3.12 )are not at all linear, with an overall increase until the 1950 s, a decline until the early 1990 s and then a recovery. Although the global land mean is an indicator of a crucial part of the global hydrological cycle, it is difficult to interpret as it is often made up of large regional anomalies of opposite sign.

Figure 3.12. Time series for 1900 to 2005 of annual global land precipitation anomalies (mm) from GHCN with respect to the 1981 to 2000 base period. The smooth curves show decadal variations (see Appendix 3.A )for the GHCN ( Peterson and Vose, 1997 [JoC, SRC] ), PREC/L ( Chen et al., 2002 [JoC] ), GPCP ( Adler et al., 2003 [JoC, SRC] ), GPCC ( Rudolf et al., 1994 [NPR] ) and CRU ( Mitchell and Jones, 2005 [SRC] ) data sets.

There are several other global land precipitation data sets covering more recent periods: Table 3.4 agives their characteristics, and the linear trends and their significance are given in Table 3.4 b. There are a number of differences in processing, data sources and time periods that lead to the differences in the trend estimates. All but one data set (GHCN) are spatially infilled by either interpolation or the use of satellite estimates of precipitation. The Precipitation Reconstruction over Land (PREC/L) data ( Chen et al., 2002 [JoC] ) include GHCN data, synoptic data from the National Oceanic and Atmospheric Administration (NOAA) Climate Prediction Center’s Climate Anomaly Monitoring System (CAMS), and the Global Precipitation Climatology Project (GPCP) data ( Adler et al., 2003 [JoC, SRC] ), and are a blend of satellite and gauge data. The Global Precipitation Climatology Centre (GPCC; updated from Rudolf et al., 1994 [NPR] ) provides monthly data from surface gauges on several grids constructed using GPCC sources (including data from CRU, GHCN, a Food and Agriculture Organization (FAO) database and many nationally provided data sets). The data set designated GPCC VASClimO ( Beck et al., 2005 [NPR, SRC] ) uses only those quasi-continuous stations whose long-term homogeneity can be assured, while GPCC v.3 has used all available stations to provide more complete spatial coverage. Gridding schemes also vary and include optimal interpolation and grid-box averaging of areally weighted station anomalies. The CRU data set is from Mitchell and Jones 2005 [SRC] ).

Table 3.4. (a) Characteristics and references of the six global land-area precipitation data sets used to calculate trends. (b) Global land precipitation trends (mm per decade). Trends with 5 and 95% confidence intervals and levels of significance (italic, 1–5%) were estimated by REML (see Appendix 3.A ), which allows for serial correlation in the residuals of the data about the linear trend. All trends are based on annual averages without estimates of intrinsic uncertainties. (a)

For 1951 to 2005, trends range from –7 to +2 mm per decade and 5 to 95% error bars range from 3.2 to 5.3 mm per decade. Only the updated PREC/L series ( Chen et al., 2002 [JoC] ) trend and the GPCC v.3 trend appear to be statistically significant, but the uncertainties, as seen in the different estimates, undermine that result. For 1979 to 2005, GPCP data are added and trends range from –16 to +13 mm per decade but none is significant. Nevertheless, the discrepancies in trends are substantial, and highlight the difficulty of monitoring a variable such as precipitation that has large variability in both space and time. On the other hand, Figure 3.12 also suggests that interannual fluctuations have some overall reproducibility for land as a whole. The lag-1 autocorrelation of the residuals from the fitted trend (i.e., the de-trended persistence) is in the range 0.3 to 0.5 for the PREC/L, CRU and GHCN series but 0.5 to 0.7 for the two GPCC and the GPCP series. This suggests that either the limited sampling by in situ gauge data adds noise, or systematic biases lasting a few years (the lifetime of a satellite) are afflicting the GPCP data, or a combination of the two.

3.3.2.2 Spatial Patterns of Precipitation Trends

The spatial patterns of trends in annual precipitation (% per century or per decade) during the periods 1901 to 2005 and 1979 to 2005 are shown in Figure 3.13 .The observed trends over land areas were calculated using GHCN station data interpolated to a 5° × 5° latitude/longitude grid. For most of North America, and especially over high-latitude regions in Canada, annual precipitation has increased during the 105-year period. The primary exception is over the southwest USA, northwest Mexico and the Baja Peninsula, where the trend in annual precipitation has been negative (1 to 2% per decade) as drought has prevailed in recent years. Across South America, increasingly wet conditions were observed over the Amazon Basin and southeastern South America, including Patagonia, while negative trends in annual precipitation were observed over Chile and parts of the western coast of the continent. The largest negative trends in annual precipitation were observed over western Africa and the Sahel. After having concluded that the effect of changing rainfall-gauge networks on Sahel rainfall time series is small, Dai et al. 2004b [JoC, SRC] ) noted that Sahel rainfall in the 1990 s has recovered considerably from the severe dry years in the early 1980 s (see Section 3.7.4 and Figure 3.37 ). A drying trend is also evident over southern Africa since 1901 . Over much of northwestern India the 1901 to 2005 period shows increases of more than 20% per century, but the same area shows a strong decrease in annual precipitation for the 1979 to 2005 period. Northwestern Australia shows areas with moderate to strong increases in annual precipitation over both periods. Over most of Eurasia, increases in precipitation outnumber decreases for both periods.

Figure 3.13. Trend of annual land precipitation amounts for 1901 to 2005 (top, % per century) and 1979 to 2005 (bottom, % per decade), using the GHCN precipitation data set from NCDC. The percentage is based on the means for the 1961 to 1990 period. Areas in grey have insufficient data to produce reliable trends. The minimum number of years required to calculate a trend value is 66 for 1901 to 2005 and 18 for 1979 to 2005 . An annual value is complete for a given year if all 12 monthly percentage anomaly values are present. Note the different colour bars and units in each plot. Trends significant at the 5% level are indicated by black + marks.

To assess the expected large regional variations in precipitation trends, Figure 3.14 presents time series of annual precipitation. The regions are 19 of those defined in Table 11.1 (see Section 11.1) and illustrated in Figure 11.26. The GHCN precipitation data set from NCDC was used, and the CRU decadal values allow the reproducibility to be assessed. Based on this, plots for four additional regions (Greenland, Sahara, Antarctica and the Tibetan Plateau) were not included, as precipitation data for these were not considered sufficiently reliable, nor was the first part of the Alaskan series (prior to 1935 ). Some discrepancies between the decadal variations are still evident at times, mostly owing to different subsets of stations and some stations coming in or dropping out, but overall the confidence in what is presented is quite high. Figure 3.15 presents a latitude-time series of zonal averages over land.

In the tropics, precipitation is highly seasonal, consisting of a dry season and a wet season in association with the summer monsoon. These aspects are discussed in more detail in Section 3.7 .Downward trends are strongest in the Sahel (see Section 3.7.4 )but occur in both western and eastern Africa in the past 50 years, and are reflected in the zonal means. The downward trends in this zone are also found in southern Asia. The linear trends of rainfall decreases for 1900 to 2005 were 7.5% in both the western Africa and southern Asia regions (significant at <1%). The area of the latter region is much greater than India, whose rainfall features strong variability but little in the way of a century-scale trend. Southern Africa also features a strong overall downward trend, although with strong multi-decadal variability present. Often the change in rainfall in these regions occurs fairly abruptly, and in several cases occurs around the same time in association with the 1976 – 1977 climate shift ( Wang and Ding, 2006 [JoC, SRC] ). The timing is not the same everywhere, however, and the downward shift occurred earlier in the Sahel (see also Section 3.7.4 , Figure 3.37 ). The main location with different trends at low latitudes is over Australia, but it is clear that large interannual variability, mostly ENSO-related, is dominant (note also the expanded vertical scales for Australia). The apparent upward trend occurs due to two rather wet spells in northern Australia in the early 1970 s and 1990 s, when it was dry in Southeast Asia (see also Section 3.7.2 ). Also of note in Australia is the marked downward trend in the far southwest characterised by a downward shift around 1975 ( Figure 3.13 ).

Figure 3.14. Precipitation for 1900 to 2005 . The central map shows the annual mean trends (% per century). Areas in grey have insufficient data to produce reliable trends. The surrounding time series of annual precipitation displayed (% of mean, with the mean given at top for 1961 to 1990 ) are for the named regions as indicated by the red arrows. The GHCN precipitation from NCDC was used for the annual green bars and black for decadal variations (see Appendix 3.A ), and for comparison the CRU decadal variations are in magenta. The range is +30 to –30% except for the two Australian panels. The regions are a subset of those defined in Table 11.1 (Section 11.1) and include: Central North America, Western North America, Alaska, Central America, Eastern North America, Mediterranean, Northern Europe, North Asia, East Asia, Central Asia, Southeast Asia, Southern Asia, Northern Australia, Southern Australia, Eastern Africa, Western Africa, Southern Africa, Southern South America, and the Amazon.

At higher latitudes, especially from 30°N to 85°N, quite distinct upward trends are evident in many regions and these are reflected in the zonal means( Figure 3.15 ). Central North America, eastern North America, northern Europe, northern Asia and central Asia (east of the Caspian Sea) all experienced upward linear trends of between 6 and 8% from 1900 to 2005 (all significant at <5%). These regions all experience snowfall (see also Section 3.3.2.3 )and part of the upward trend may arise from changes in efficiency of catching snow, especially in northern Asia. However, there is ample evidence that these trends are real (see Section 3.3.4 ), and they extend from North America to Europe across the North Atlantic as evidenced by ocean freshening, documented in Sections 5.2.3 and 5.3.2 .Western North America shows longer time-scale variability, principally due to the severe drought in the 1930 s and lesser events more recently. Note the tendency for inverse variations between northern Europe and the Mediterranean, associated with changes in the NAO (see Section 3.6.4 ). Southern Europe and parts of central Europe, as well as North Africa, are characterised by a drier winter (DJF) during the positive phase of the NAO, while the reverse is true in the British Isles, Fennoscandia and northwestern Russia.

Figure 3.15. Latitude-time section of zonal average annual anomalies for precipitation (%) over land from 1900 to 2005, relative to their 1961 to 1990 means. Values are smoothed with the 5-point filter to remove fluctuations of less than about six years (see Appendix 3.A ). The colour scale is nonlinear and grey areas indicate missing data.

In the SH, Amazonia and southern South America feature opposite changes, as the South American monsoon features shifted southwards (see Section 3.7.3 ). This movement was in association with changes in ENSO and the 1976 – 1977 climate shift. The result is a pronounced upward trend in Argentina and the La Plata River Basin, but not in Chile (where the main declines in precipitation are evident in the austral summer (DJF) and autumn (MAM; Figure 3.13 ). Decadal-scale variations over Amazonia are also out of phase with the Central American region to the north, which in turn has out-of-phase variations with western North America, again suggestive of latitudinal changes in monsoon features. East and Southeast Asia show hardly any long-term changes, with both having plentiful rains in the 1950 s. At interannual time scales there are a number of surprisingly strong correlations: Amazonia is correlated with northern Australia (0.44, significant at <1%) and also Southeast Asia (0.55, <1%), while southern South America is inversely correlated with western Africa (−0.51, <1%). The correlations are surprising because they are based on high-frequency relationships and barely change when the smoothed series are used.

3.3.2.3 Changes in Snowfall

Winter precipitation has increased at high latitudes, although uncertainties exist because of changes in undercatch, especially as snow changes to rain. Snow cover changes are discussed in Section 4.2 .Annual precipitation for the circumpolar region north of 50°N has increased during the past 50 years (not shown) by approximately 4% but this increase has not been homogeneous in time and space ( Groisman et al., 2003 [NPR, SRC] , 2005 ). Statistically significant increases were documented over Fennoscandia, coastal regions of northern North America ( Groisman et al., 2005 [SRC] ), most of Canada (particularly northern regions) up until at least 1995 when the analysis ended ( Stone et al., 2000 [ARC] ), the permafrost-free zone of Russia ( Groisman and Rankova, 2001 [JoC, SRC] ) and the entire Great Russian Plain ( Groisman et al., 2005 [SRC] , 2007 ). However, there were no discernible changes in summer and annual precipitation totals over northern Eurasia east of the Ural Mountains ( Gruza et al., 1999 [JoC] ; Sun and Groisman, 2000 [JoC, SRC] ; Groisman et al., 2005 [SRC] , 2007 ). Rainfall (liquid precipitation) has increased during the past 50 years over western portions of North America and Eurasia north of 50°N by about 6%. Rising temperatures have generally resulted in rain rather than snow in locations and seasons where climatological average ( 1961 – 1990 ) temperatures were close to 0°C. The liquid precipitation season has become longer by up to three weeks in some regions of the boreal high latitudes over the last 50 years ( Cayan et al., 2001 [JoC, SRC] ; Groisman et al., 2001 [Ambiguous] ; Easterling, 2002 [JoC, ARC] ; Groisman et al., 2005 [SRC] , 2007 ) owing, in particular, to an earlier onset of spring. Therefore, in some regions (southern Canada and western Russia), snow has provided a declining fraction of total annual precipitation ( Groisman et al., 2003 [NPR, SRC] , 2005, 2007 ). In other regions, particularly north of 55°N, the fraction of annual precipitation falling as snow in winter has changed little.

( Berger et al. 2002 ) ) found a trend towards fewer snowfall events during winter across the lower Missouri River Basin from 1948 to 2002, but little or no trend in snowfall occurrences within the plains region to the south. In New England, there has been a decrease in the proportion of precipitation occurring as snow at many locations, caused predominantly by a decrease in snowfall, with a lesser contribution from increased rainfall ( Huntington et al., 2004 [JoC] ). By contrast, Burnett et al. 2003 [JoC] ) found large increases in lake-effect snowfall since 1951 for locations near the North American Great Lakes, consistent with the observed decrease in ice cover for most of the Great Lakes since the early 1980 s ( Assell et al., 2003 [JoC] ). In addition to snow data, Burnett et al. 2003 [JoC] ) used lake sediment reconstructions for locations south of Lake Ontario to indicate that these increases have been ongoing since the beginning of the 20th century. Ellis and Johnson 2004 [JoC] ) found that the increases in snowfall across the regions to the lee of Lakes Erie and Ontario are due to increases in the frequency of snowfall at the expense of rainfall events, an increase in the intensity of snowfall events, and to a lesser extent an increase in the water equivalent of the snow. In Canada, the frequency of heavy snowfall events has decreased since the 1970 s in the south and increased in the north ( Zhang et al., 2001a [JoC, ARC] ).

3.3.2.4 Urban Areas

As noted in Section 3.2.2.2 (see also Box 7.2 ), the microclimates in cities are clearly different than in neighbouring rural areas. The presence of a city affects runoff, moisture availability and precipitation. ( Crutzen 2004 ) ) pointed out that while human energy production is relatively small globally compared with the Sun, it is locally important in cities, where it can reach 20 to 70 Wm^–2.Urban effects can lead to increased precipitation (5 to 25% over background values) during the summer months within and 50 to 75 km downwind of the city ( Changnon et al., 1981 [NPR] ). More frequent or intense storms have been linked to city growth in Phoenix, Arizona ( (Balling and Brazel, 1987 ) ) and Mexico City ( (Jauregui and Romales, 1996 ) ). More recent observational studies ( (Bornstein and Lin, 2000; ) Changnon and Westcott, 2002 [MoS] ; Shepherd et al., 2002 [JoC, SRC] ;( Diem and Brown, 2003; ) Dixon and Mote, 2003 [JoC] ;( Fujibe, 2003; ) Shepherd and Burian, 2003 [SRC] ; Inoue and Kimura, 2004 [JoC] ; Shepherd et al., 2004 [SRC] ; Burian and Shepherd, 2005 [NPR, SRC] ) have continued to link urban-induced dynamic processes to precipitation anomalies. Nor is land use change confined to urban areas (see Section 7.2 ). Other changes in land use also affect precipitation. A notable example arises from deforestation in the Amazon, where Chagnon and Bras 2005 [JoC] ) found large changes in local rainfall with increases in deforested areas, associated with local atmospheric circulations that are changed by gradients in vegetation, and also found changes in seasonality.

Suggested mechanisms for urban-induced rainfall include: (1) enhanced convergence due to increased surface roughness in the urban environment (e.g., Changnon et al., 1981 [NPR] ;( Bornstein and Lin, 2000; ) ( Thielen et al., 2000 ) ); (2) destabilisation due to UHI thermal perturbation of the boundary layer and resulting downstream translation of the UHI circulation or UHI-generated convective clouds (e.g., Shepherd et al., 2002 [JoC, SRC] ; Shepherd and Burian, 2003 [SRC] ); (3) enhanced aerosols in the urban environment for cloud condensation nuclei sources (e.g., ( Diem and Brown, 2003; ) Molders and Olson, 2004 [JoC] ); or (4) bifurcating or diverting of precipitating systems by the urban canopy or related processes (e.g., ( Bornstein and Lin, 2000 ) ). The ‘weekend effect’ noted in Section 3.2.2.2 likely arises from some of these mechanisms. The diurnal cycle in precipitation, which varies over the USA from late afternoon maxima in the Southeast to nocturnal maxima in the Great Plains ( Dai and Trenberth, 2004 [PoC, JoC, MoS, SRC] ), may be modified in some regions by urban environments. Dixon and Mote 2003 [JoC] ) found that a growing UHI effect in Atlanta, Georgia (USA) enhanced and possibly initiated thunderstorms, especially in July (summer) just after midnight. Low-level moisture was found to be a key factor.

3.3.2.5 Ocean Precipitation

Remotely sensed precipitation measurements over the ocean are based on several different sensors in the microwave and infrared that are combined in different ways. Many experimental products exist. Operational merged products seem to perform best in replicating island-observed monthly amounts ( Adler et al., 2001 [JoC, SRC] ). This does not mean they are best for trends or low-frequency variability, because of the changing mixes of input data. The main global data sets available for precipitation, and which therefore include ocean coverage, have been the GPCP ( Huffman et al., 1997 [JoC] ; Adler et al., 2003 [JoC, SRC] ) and the NOAA Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP; Xie and Arkin, 1997 [JoC, MoS] ). Comparisons of these data sets and others ( Adler et al., 2001 [JoC, SRC] ; Yin et al., 2004 [JoC] ) reveal large discrepancies over the ocean; however, there is better agreement among the passive microwave products even using different algorithms. Over the tropical oceans, mean amounts in CMAP and GPCP differ by 10 to 15%. Calibration using observed rainfall from small atolls in CMAP was extended throughout the tropics in ways that are now recognised as incorrect. However, evaluation of GPCP reveals that it is biased low by 16% at such atolls ( Adler et al., 2003 [JoC, SRC] ), also raising questions about the ocean GPCP values. Differences arise due to sampling and algorithms. Polar-orbiting satellites each obtain only two instantaneous rates per day over any given location, and thus suffer from temporal sampling deficiencies that are offset by using geostationary satellites. However, only less-accurate infrared sensors are available with the latter. Model-based (including reanalysis) products perform poorly in the evaluation of Adler et al. 2001 [JoC, SRC] ) and are not currently suitable for climate monitoring. Robertson et al. 2001b [JoC, SRC] ) examined monthly anomalies from several satellite-derived precipitation data sets (using different algorithms) over the tropical oceans. The expectation in the TAR was that measurements from the Tropical Rainfall Measuring Mission (TRMM) Precipitation Radar (PR) and passive TRMM microwave imager (TMI) would clarify the reasons for the discrepancies, but this has not yet been the case. Robertson et al. 2003 [JoC] ) documented poorly correlated behaviour (correlation 0.12) between the monthly, tropical ocean-averaged precipitation anomalies from the PR and TMI sensors. Although the TRMM PR responds directly to precipitation size hydrometeors, it operates with a single attenuating frequency (13.8 GHz) that necessitates significant microphysical assumptions regarding drop size distributions for relating reflectivity, signal attenuation and rainfall, and uncertainties in microphysical assumptions for the primary TRMM algorithm (2A25) remain problematic.

The large regional signals from monsoons and ENSO that emphasise large-scale shifts in precipitation are reasonably well captured in GPCP and CMAP (see Section 3.6.2 ), but cancel out when area-averaged over the tropics, and the trends and variability of the tropical average are quite different in the two products. Global precipitation from GPCP (updated from Adler et al., 2003 [JoC, SRC] , but not shown) has monthly variability with a standard deviation of about 2% of the mean. The variability in the ocean and land areas when examined separately is larger, about 3%, and with variations related to ENSO events ( Curtis and Adler, 2003 [JoC, SRC] ). During El Niño events, area-averaged precipitation increases over the oceans but decreases over land.

Although the trend over 25 years in global total precipitation in the GPCP data set ( Adler et al., 2003 [JoC, SRC] ) is very small, there is a small increase (about 4% over the 25 years) over the oceans in the latitude range 25°S to 25°N, with a partially compensating decrease over land (2%) in the same latitude belt. Northern mid-latitudes show a decrease over land and ocean. Over a slightly longer time frame, precipitation increased over the North Atlantic between 1960 to 1974 and 1975 to 1989 ( Josey and Marsh, 2005 [JoC, SRC] ) and is reflected in changes in salinity in the oceans( Section 5.2.3 ). The inhomogeneous nature of the data sets and the large ENSO variability limit what can be said about the validity of changes, both globally and regionally.

3.4.1.1 Radiosondes

Since the TAR, considerable effort has been devoted to assessing and improving the quality of the radiosonde temperature record (see Appendix 3.B.5.1). A particular aim has been to reduce artificial changes arising from instrumental and procedural developments during the seven decades ( 1940 s– 2000 s) of the radiosonde record ( Free and Seidel, 2005 [JoC, SRC] ; Thorne et al., 2005a [PoC, JoC, SRC] ; Karl et al., 2006 [NPR, PoC, SRC] ). Comparisons of several adjustment methods showed that they gave disparate results when applied to a common set of radiosonde station data ( Free et al., 2002 [JoC, SRC] ). One approach, based on the physics of heat transfer within the radiosonde, performed poorly when evaluated against satellite temperature records ( Durre et al., 2002 [JoC, SRC] ). Another method, comparison with satellite data (HadRT (Hadley Centre Radiosonde Temperature Data Set); Parker et al., 1997 [JoC] ), is limited to the satellite era and to events with available metadata, and causes a reduction in spatial consistency of the data. A comprehensive intercomparison ( Seidel et al., 2004 [JoC] ) showed that five radiosonde data sets yielded consistent signals for higher-frequency events such as ENSO, the Quasi-Biennial Oscillation (QBO) and volcanic eruptions, but inconsistent signals for long-term trends.

Several approaches have been used to create new adjusted data sets since the TAR. The Lanzante-Klein-Seidel (LKS; Lanzante et al., 2003a [JoC, SRC] ,b) data set, using 87 carefully selected stations, has subjectively derived bias adjustments throughout the length of its record but terminates in 1997 . It has been updated using the Integrated Global Radiosonde Archive (IGRA; Durre et al., 2006 [JoC, SRC] ) by applying a different bias adjustment technique ( Free et al., 2004b [JoC, SRC] ) after 1997, creating a new archive (Radiosonde Atmospheric Temperature Products for Assessing Climate; RATPAC). Another new radiosonde record, HadAT2 (Hadley Centre Atmospheric Temperature Data Set Version 2, successor to HadRT), uses a neighbour comparison approach to build spatial as well as temporal consistency. A third approach ( Haimberger, 2005 [NPR, MoS] ) uses the bias adjustments estimated during data assimilation into model-based reanalyses to identify and reduce inhomogeneities in radiosonde data. Despite the risk of contamination by other biased data or by model bias, the resulting adjustments agree with those estimated by other methods. Rather than adjusting the data, Angell 2003 [JoC] ) tried to reduce data quality problems by removing several tropical stations from his radiosonde network.

Despite these efforts to produce homogeneous data sets, recent analyses of radiosonde data indicate that significant problems remain. Sherwood et al. 2005 [JoC, SRC] ) found substantial changes in the diurnal cycle in unadjusted radiosonde data. These changes are probably a consequence of improved sensors and radiation error adjustments. Relative to nighttime values, they found a daytime warming of sonde temperatures prior to 1971 that is likely to be spurious and then a spurious daytime cooling from 1979 to 1997 . They estimated that there was probably a spurious overall downward trend in sonde temperature records during the satellite era (since 1978 ) throughout the atmosphere of order 0.1°C per decade globally. The assessed spurious cooling is greatest in the tropics (0.16°C per decade for the 850 to 300 hPa layer) and least in the NH extratropics (0.04°C per decade). Randel and Wu 2006 [JoC] ) used collocated MSU data to show that cooling biases remain in some of the LKS and RATPAC radiosonde data for the tropical stratosphere and upper troposphere due to changes in instruments and radiation correction adjustments. They also identified problems in night data as well as day, indicating that negative biases are not limited to daytime observations. However, a few stations may have positive biases ( Christy and Spencer, 2005 [NPR, SRC] ).

The radiosonde data set is limited to land areas, and coverage is poor over the tropics and SH. Accordingly, when global estimates based solely on radiosondes are presented, there are considerable uncertainties ( Hurrell et al., 2000 [JoC, SRC] ; Agudelo and Curry, 2004 [JoC] ) and denser networks – which perforce still omit oceanic areas – may not yield more reliable ‘global’ trends ( Free and Seidel, 2005 [JoC, SRC] ). Radiosonde records have an advantage of starting in the 1940 s regionally, and near-globally from about 1958 . They monitor the troposphere and lower stratosphere; the layers analysed are described below and in Figure 3.16 .Radiosonde-based global mean temperature estimates are given in Figure 3.17 ,presented later.

Figure 3.16. Vertical weighting functions (grey) depicting the layers sampled by satellite MSU measurements and their derivatives, and used also for radiosonde and reanalysis records. The right panel schematically depicts the variation in the tropopause (that separates the stratosphere and troposphere) from the tropics (left) to the high latitudes (right). The fourth panel depicts T4 in the lower stratosphere, the third panel shows T2, the second panel shows the troposphere as a combination of T2 and T4 ( Fu et al., 2004a [JoC] ) and the first panel shows T2_LT from the UAH for the low troposphere. Adapted from Karl et al. 2006 [NPR, PoC, SRC] ).

3.4.1.2 The Satellite Microwave Sounding Unit Record

3.4.1.2.2 Progress since the TAR

Since the TAR, several important developments and advances have occurred in the analysis of satellite measurements of atmospheric temperatures. Existing data sets have been scrutinised and problems identified, leading to new versions as described below. A number of new data records have been constructed from the MSU measurements, as well as from global reanalyses (see Section 3.4.1.3 ). Further, new insights have come from statistical combinations of the MSU records from different channels that have minimised the influence of the stratosphere on the tropospheric records ( Fu et al., 2004a [JoC] ,b; Fu and Johanson, 2004 [JoC] , 2005 ). These new data sets and analyses are very important because the differences highlight assumptions and it becomes possible to estimate the uncertainty in satellite-derived temperature trends that arises from different methods and approaches to the construction of temporally consistent records.

Analyses of MSU channels 2 and 4 have been conducted by the University of Alabama in Huntsville (UAH; Christy et al., 2000 [SRC] , 2003 ) and by Remote Sensing Systems (RSS; Mears et al., 2003 [JoC, SRC] ; Mears and Wentz, 2005 [NPR, SRC] ). Another analysis of channel 2 is that of Vinnikov and Grody 2003 [JoC] ; version 1 – VG1), now superseded by Grody et al. 2004 [JoC] and Vinnikov et al. 2006 [JoC] ; version 2 – VG2). MSU channel 2 (T2) measures a thick layer of the atmosphere, with approximately 75 to 80% of the signal coming from the troposphere, 15% from the lower stratosphere, and the remaining 5 to 10% from the surface. MSU channel 4 (T4) is primarily sensitive to temperature in the lower stratosphere( Figure 3.16 ).

Global time series from each of the MSU records are shown in Figure 3.17 and calculated global trends are depicted in Figure 3.18 .These show a global cooling of the stratosphere (T4) of –0.32°C to –0.47 °C per decade and a global warming of the troposphere (T2) of 0.04°C to 0.20°C per decade for the period 1979 to 2004 . The large spread in T2 trends stems from differences in the inter-satellite calibration and merging technique, and differences in the corrections for orbital drift, diurnal cycle change and the hot-point calibration temperature ( Christy et al., 2003 [MoS, SRC] ; Mears et al., 2003 [JoC, SRC] ; Christy and Norris, 2004 [JoC, SRC] ; Grody et al., 2004 [JoC] ; Fu and Johanson, 2005 [JoC] ; Mears and Wentz, 2005 [NPR, SRC] ; Vinnikov et al., 2006 [JoC] ; see also Appendix 3.B.5.3)

The RSS results for T2 indicate nearly 0.1°C per decade more warming in the troposphere than UAH (see Figure 3.18 )and most of the difference arises from the use of different amounts of data to determine the parameters of the calibration target effect (Appendix 3.B.5.3). The UAH analysis yields parameters for the NOAA-9 satellite ( 1985 – 1987 ) outside of the physical bounds expected by Mears et al. 2003 [JoC, SRC] ). Hence, the large difference in the calibration parameters for the single instrument mounted on the NOAA-9 satellite accounted for a substantial part of the difference between the UAH and RSS T2 trends. The rest arises from differences in merging parameters for other satellites; differences in the correction for the drift in measurement time, especially for the NOAA-11 satellite ( Mears et al., 2003 [JoC, SRC] ; Christy and Norris, 2004 [JoC, SRC] ); and differences in the ways the hot-point temperature is corrected for ( Grody et al., 2004 [JoC] ; Fu and Johanson, 2005 [JoC] ). In the tropics, these accounted for differences in T2 trends of about 0.07°C per decade after 1987 and discontinuities were also present in 1992 and 1995 at times of satellite transitions ( Fu and Johanson, 2005 [JoC] ). The T2 data record of Grody et al. 2004 [JoC] and Vinnikov et al. 2006 [JoC] ) (VG2) shows slightly more warming in the troposphere than the RSS data record( Figure 3.18 ). See also Appendix 3.B.5.3 for discussion of the VG2 analysis.

Although the T4 from RSS has about 0.1°C per decade less cooling than the UAH product( Figure 3.18 ), both data sets support the conclusions that the stratosphere has undergone strong cooling since 1979 . Because about 15% of the signal for T2 comes from the lower stratosphere, the observed cooling causes the reported T2 trends to underestimate tropospheric warming. By creating a weighted combination of T2 and T4, this effect has been greatly reduced ( Fu et al., 2004a [JoC] ; see Figure 3.16 ). This technique for estimating the global mean temperature implies small negative weights at some stratospheric levels, but because of vertical coherence these merely compensate for other positive weights nearby and it is the integral that matters ( Fu and Johanson, 2004 [JoC] ). From 1979 to 2001 the stratospheric contribution to the trend in T2 is about –0.08°C per decade. Questions about this technique ( Tett and Thorne, 2004 [NPR, PoC, JoC, SRC] ) have led to clearer interpretation of its application to the tropics ( Fu et al., 2004b [NPR, JoC] ). The technique has also been successfully applied to model results ( Gillett et al., 2004 [NPR, PoC, JoC, ARC] ; Kiehl et al., 2005 [JoC, MoS] ), although model biases in depicting stratospheric cooling can affect results. In a further development, weighted combinations of T2, MSU channel 3 (T3) and T4 since 1987 have formed tropical series for the upper, lower and whole troposphere ( Fu and Johanson, 2005 [JoC] ).

Figure 3.17. Observed surface and upper-air temperature anomalies (°C). (A) Lower stratospheric T4, (B) Tropospheric T2, (C) Lower tropospheric T2_LT,from UAH, RSS and VG2 MSU satellite analyses and UKMO HadAT2 and NOAA RATPAC radiosonde observations; and (D) Surface records from NOAA, NASA/GISS and UKMO/CRU (HadCRUT2v). All time series are monthly mean anomalies relative to the period 1979 to 1997 smoothed with a seven-month running mean filter. Major volcanic eruptions are indicated by vertical blue dashed lines. Adapted from Karl et al. 2006 [NPR, PoC, SRC] ).

By differencing T2 measurements made at different slant angles, the UAH group produced an updated data record weighted for the lower and mid troposphere, T2_LT ( Christy et al., 2003 [MoS, SRC] ). This retrieval also has the effect of removing the stratospheric influence on long-term trends, but its uncertainties are augmented by the need to compensate for orbital decay and by computing a small residual from two large values ( Wentz and Schabel, 1998 [JoC] ). T2_LT retrievals include a large signal from the surface and so are adversely affected by changes in surface emissivity, including changes in sea ice cover ( Swanson, 2003 [JoC] Fu and Johanson 2005 [JoC] ) found that the T2_LT trends were physically inconsistent compared with those of the surface, T2 and T4, even if taken from the UAH record. They also showed that the large trend bias is mainly attributed to the periods when a satellite had substantial drifts in local equator crossing time that caused large changes in calibration target temperatures and large diurnal drifts. Mears and Wentz 2005 [NPR, SRC] ) further found that the adjustments for diurnal cycle required from satellite drift had the wrong sign in the UAH record in the tropics. Corrections have been made (version 5.2; Christy and Spencer, 2005 [NPR, SRC] ) and are reflected in Figure 3.18 ,but the trend in the tropics is still smaller for most periods than both those in the troposphere (using T2 and T4) and those at the surface. Mears and Wentz 2005 [NPR, SRC] ) computed their own alternative T2_LT record and found a T2_LT trend nearly 0.1°C per decade larger than the revised UAH trend. After 1987, when MSU channel 3 became available, Fu and Johanson 2005 [JoC] ), using RSS data, found a systematic trend of increasing temperature with altitude throughout the tropics.

Figure 3.18. Linear temperature trends (°C per decade) for 1979 to 2004 for the globe (top) and tropics (20°N to 20°S; bottom) for the MSU channels T4 (top panel) and T2 (second panel) or equivalent for radiosondes and reanalyses; for the troposphere (third panel) based on T2 with T4 used to statistically remove stratospheric influences ( Fu et al., 2004a [JoC] ); for the lower troposphere (fourth panel) based on the UAH retrieval profile; and for the surface (bottom panel). Surface records are from NOAA/NCDC (green), NASA/GISS (blue) and HadCRUT2v (light blue). Satellite records are from UAH (orange), RSS (dark red) and VG2 (magenta); radiosonde-based records are from NOAA RATPAC (brown) and HadAT2 (light green); and atmospheric reanalyses are from NRA (red) and ERA-40 (cyan). The error bars are 5 to 95% confidence limits associated with sampling a finite record with an allowance for autocorrelation. Where the confidence limits exceed –1, the values are truncated. ERA-40 trends are only for 1979 to August 2002 . Data from Karl et al. 2006 [NPR, PoC, SRC] ; D. Seidel courtesy of J. Lanzante; and J. Christy).

Comparisons of tropospheric radiosonde station data with collocated satellite data ( Christy and Norris, 2004 [JoC, SRC] ) show considerable scatter, and root mean square differences between UAH satellite data and radiosondes are substantial ( Hurrell et al., 2000 [JoC, SRC] ). Although Christy and Norris 2004 [JoC, SRC] ) found good agreement between median radiosonde temperature trends and UAH trends, comparisons are more likely to be biased by spurious cooling than by spurious warming in unhomogenised ( Sherwood et al., 2005 [JoC, SRC] ) and even homogenised ( Randel and Wu, 2006 [JoC] ) radiosonde data (see Section 3.4.1.1 and Appendix 3.B.5.1). In the stratosphere, radiosonde trends are more negative than both MSU retrievals, especially when compared with RSS, and this is very likely due to changes in sondes and their processing for radiation corrections ( Randel and Wu, 2006 [JoC] ).

Geographical patterns of the linear trend in tropospheric temperature from 1979 to 2004 ( Figure 3.19 )are qualitatively similar in the RSS and UAH MSU data sets. Both show coherent warming over most of the NH, but UAH shows cooling over parts of the tropical Pacific and tropospheric temperature trends differ south of 45°S where UAH indicate more cooling than RSS.

Figure 3.19. Linear tropospheric temperature trends (°C per decade) for 1979 to 2005 from RSS (based on T2 and T4 adjusted as in Fu et al., 2004a [JoC] ). Courtesy Q. Fu.

3.4.1.3 Reanalyses

Acomprehensive global reanalysis completed since the TAR, ERA-40 ( Uppala et al., 2005 [JoC, SRC] ), extends from September 1957 to August 2002 . Reanalysis is designed to prevent changes in the analysis system from contaminating the climate record, as occurs with global analyses from operational numerical weather prediction, and it compensates for some but not all of the effects of changes in the observing system (see Appendix 3.B.5.4). Unlike the earlier NRA that assimilated satellite retrievals, ERA-40 assimilated bias-adjusted radiances including MSU data ( Harris and Kelly, 2001 [JoC] ; Uppala et al., 2005 [JoC, SRC] ), and the assimilation procedure itself accounts for orbital drift and change in satellite height – factors that have to be addressed in direct processing of MSU radiances for climate studies (e.g., Christy et al., 2003 [MoS, SRC] ; Mears et al., 2003 [JoC, SRC] ; Mears and Wentz, 2005 [NPR, SRC] ). Onboard calibration biases are treated indirectly via the influence of other data sets. Nonetheless, the veracity of low-frequency variability in atmospheric temperatures is compromised in ERA-40 by residual problems in bias corrections.

Trends and low-frequency variability in large-scale surface air temperature from ERA-40 and from the monthly climate station data analysed by Jones and Moberg 2003 [JoC, SRC] ) are in generally good agreement from the late 1970 s onwards (see also Section 3.2.2.1 ). Temperatures from ERA-40 vary quite coherently throughout the planetary boundary layer over this period, and earlier for regions with consistently good coverage from both surface and upper-air observations ( Simmons et al., 2004 [JoC, SRC] ).

Processed MSU records of layer temperature have been compared with equivalents derived from the ERA-40 analyses ( Santer et al., 2004 [PoC, JoC, ARC] ). The use of deep layers conceals disparate trends at adjacent tropospheric levels in ERA-40. Relatively cold tropospheric values before the satellite era arose from a combination of scarcity of radiosonde data over the extratropical SH and a cold bias of the assimilating model, giving a tropospheric warming trend that is clearly too large when taken over the full period of the reanalysis ( Bengtsson et al., 2004 [JoC, ARC] ; Simmons et al., 2004 [JoC, SRC] ; Karl et al., 2006 [NPR, PoC, SRC] ). ERA-40 also exhibits a middle-tropospheric cooling over most of the tropics and subtropics since the 1970 s that is certainly too strong owing to a warm bias in the analyses for the early satellite years.

Tropospheric patterns of trends from ERA-40 are similar to Figure 3.19 ,with coherent warming over the NH, although with discrepancies south of 45°S. These differences are not fully understood, although the treatment of surface emissivity anomalies over snow- and ice-covered surfaces may contribute ( Swanson, 2003 [JoC] ). At high southern latitudes, ERA-40 shows strong positive temperature trends in JJA in the period 1979 to 2001, in good accord with antarctic radiosonde data ( Turner et al., 2006 [JoC] ). The large-scale patterns of stratospheric cooling are similar in ERA-40 and the MSU data sets ( Santer et al., 2004 [PoC, JoC, ARC] ). However, the ERA-40 analyses in the lower stratosphere are biased cold relative to radiosonde data in the early satellite years, reducing downward trends. Section 3.5 relates the trends to atmospheric circulation changes.

3.4.1.4 The Tropopause

The tropopause marks the division between the troposphere and stratosphere and generally a minimum in the vertical profile of temperature. The height of the tropopause is affected by the heat balance of both the troposphere and the stratosphere. For example, when the stratosphere warms owing to absorption of radiation by volcanic aerosol, the tropopause is lowered. Conversely, a warming of the troposphere raises the tropopause, as does a cooling of the stratosphere. The latter is expected as a result of increasing greenhouse gas concentrations and stratospheric ozone depletion. Accordingly, changes in the height of the tropopause provide a sensitive indicator of human effects on climate. Inaccuracies and spurious trends in NRA preclude their use in determining tropopause trends ( Randel et al., 2000 [JoC] ) although they were found useful for interannual variability. Over 1979 to 2001, tropopause height increased by nearly 200 m (as a global average) in ERA-40, partly due to tropospheric warming plus stratospheric cooling ( Santer et al., 2004 [PoC, JoC, ARC] ). Atmospheric temperature changes in the UAH and RSS satellite MSU data sets (see Section 3.4.1.2 )were found to be more highly correlated with changes in ERA-40 than with those in NRA, illustrating the improved quality of ERA-40 and satellite data. The Santer et al. 2004 [PoC, JoC, ARC] ) results provide support for warming of the troposphere and cooling of the lower stratosphere over the last four decades of the 20th century, and indicate that both of these changes in atmospheric temperature have contributed to an overall increase in tropopause height. The radiosonde-based analyses of Randel et al. 2000 [JoC] Seidel et al. 2001 [JoC, ARC] and Highwood et al. 2000 [JoC, SRC] ) also show increases in tropical tropopause height.

3.4.1.5 Synthesis and Comparison with the Surface Temperatures

Figure 3.17 presents the radiosonde and satellite global time series and Figure 3.18 gives a summary of the linear trends for 1979 to 2004 for global and tropical (20°N to 20°S) averages. Values at the surface are from NOAA (NCDC), NASA (GISS), UKMO/CRU (HadCRUT2v) and the NRA and ERA-40 reanalyses. Trends aloft are for the lower troposphere corresponding to T2_LT,T2, T4 and also the linear combination of T2 and T4 to better depict the entire troposphere as given by Fu et al. 2004a [JoC] ). In addition to the reanalyses, the results from the satellite-based methods from UAH, RSS and VG2 are given along with radiosonde estimates from HadAT2 and RATPAC. The ERA-40 trends only extend through August 2002, and VG2 is available only for T2. The error bars plotted here are 5 to 95% confidence limits associated with sampling a finite record where an allowance has been made for temporal autocorrelation in computing degrees of freedom( Appendix 3.A ). However, the error bars do not include spatial sampling uncertainty, which increases the noise variance. Noise typically cuts down on temporal autocorrelation and reduces the temporal sampling error bars, which is why the RATPAC error bars are often smaller than the rest. Other sources of ‘structural’ and ‘internal’ errors of order 0.08°C for 5 to 95% levels ( Mears and Wentz, 2005 [NPR, SRC] ; see Appendix 3.B.5) are also not explicitly accounted for here. Structural uncertainties and parametric errors ( Thorne et al., 2005b [PoC, JoC, SRC] ) reflect divergence between different data sets after the common climate variability has been accounted for and are better illustrated by use of difference time series, as seen for instance in T2 for RSS vs. UAH in Fu and Johanson 2005 [JoC] ; see also Karl et al., 2006 [NPR, PoC, SRC] ).

From Figure 3.17 the first dominant impression is that overall, the records agree remarkably well, especially in the timing and amplitude of interannual variations. This is especially true at the surface, and even the tropospheric records from the two radiosonde data sets agree reasonably well, although HadAT2 has lower values in the 1970 s. In the lower stratosphere, all records replicate the dominant variations and the pulses of warming following the volcanic eruptions indicated in the figure. The sonde records differ prior to 1963 in the lower stratosphere when fewer observations were available, and differences also emerge among all data sets after about 1992, with the sonde values lower than the satellite temperatures. The focus on linear trends tends to emphasize these relatively small differences.

Alinear trend over the long term is often not a very good approximation of what has occurred ( Seidel and Lanzante, 2004 [JoC, SRC] ; Thorne et al., 2005a [PoC, JoC, SRC] ,b); alternative interpretations are to factor in the abrupt 1976 – 1977 climate regime shift ( Trenberth, 1990 [PoC, JoC, SRC] ) and episodic stratospheric warming and tropospheric cooling for the two years following major volcanic eruptions. Hence, the confidence limits for linear trends( Figure 3.18 )are very large in the lower stratosphere owing to the presence of the large warming perturbations from volcanic eruptions. In the troposphere, the confidence limits are much wider in the tropics than globally, reflecting the strong interannual variability associated with ENSO.

Radiosonde, satellite observations and reanalyses agree that there has been global stratospheric cooling since 1979 (Figures 3.17 and 3.18 ), although radiosondes almost certainly still overestimate the cooling owing to residual effects of changes in instruments and processing (such as for radiation corrections; Lanzante et al., 2003b [JoC, SRC] ; Sherwood et al., 2005 [JoC, SRC] ; Randel and Wu, 2006 [JoC] ) and possibly increased sampling of cold conditions owing to stronger balloons ( Parker and Cox, 1995 [JoC] ). As the stratosphere is cooling and T2 has a 15% signal from there, it is virtually certain that the troposphere must be warming at a significantly greater rate than indicated by T2 alone. Thus, the tropospheric record adjusted for the stratospheric contribution to T2 has warmed more than T2 in every case. The differences range from 0.06°C per decade for ERA-40 to 0.09°C per decade for both radiosonde and NRA data sets. For UAH and RSS the difference is 0.07°C per decade.

The weakest tropospheric trends occur for NRA. However, unlike ERA-40, the NRA did not allow for changes in greenhouse gas increases over the record ( Trenberth, 2004 [PoC, JoC, SRC] ), resulting in errors in radiative forcing and in satellite retrievals in the infrared and making trends unreliable ( Randel et al., 2000 [JoC] ); indeed, upward trends at high surface mountain stations are stronger than NRA free-atmosphere temperatures at nearby locations ( Pepin and Seidel, 2005 [JoC] ). The records suggest that since 1979, the global and tropical tropospheric trends are similar to those at the surface although RSS, and by inference VG2, indicate greater tropospheric than surface warming. The reverse is indicated by the UAH and the radiosonde record although these data are subject to significant imperfections discussed above. Amplification occurs in the tropics for the RSS fields, especially after 1987 when there are increasing trends with altitude throughout the troposphere based on T2, T3 and T4 ( Fu and Johanson, 2005 [JoC] ). In the tropics, the theoretically expected amplification of temperature perturbations with height is borne out by interannual fluctuations (ENSO) in radiosonde, RSS, UAH and model data ( Santer et al., 2005 [PoC, JoC, ARC] ), but it is not borne out in the trends of the radiosonde records and UAH data.

The global mean trends since 1979 disguise many regional differences. In particular, in winter much larger temperature trends are present at the surface over northern continents than at higher levels ( Karl et al., 2006 [NPR, PoC, SRC] ) (see Figures 3.9 and 3.10 ; FAQ 3.1 ,Figure 1). These are associated with weakening of shallow winter temperature inversions and the strong stable surface layers that have little signature in the main troposphere. Such changes are related to changes in surface winds and atmospheric circulation (see Section 3.6.4 ).

In summary, for the period since 1958, overall global and tropical tropospheric warming estimated from radiosondes has slightly exceeded surface warming( Figure 3.17 and Karl et al., 2006 [NPR, PoC, SRC] ). The climate shift of 1976 appeared to yield greater tropospheric than surface warming( Figure 3.17 ); such climate variations make differences between the surface and tropospheric temperature trends since 1979 unsurprising. After 1979, there has also been global and tropical tropospheric warming; however, it is uncertain whether tropospheric warming has exceeded that at the surface because the spread of trends among tropospheric data sets encompasses the surface warming trend. The range (due to different data sets, but not including the reanalyses) of global surface warming since 1979 from Figure 3.18 is 0.16°C to 0.18°C per decade compared to 0.12°C to 0.19°C per decade for MSU estimates of tropospheric temperatures. A further complexity is that surface trends have been greater over land than over ocean. Substantial cooling has occurred in the lower stratosphere. Compensation for the effects of stratospheric cooling trends on the T2 record (a cooling of about 0.08°C per decade) has been an important development. However, a linear trend is a poor fit to the data in the stratosphere and the tropics at all levels. The overall global variability is well replicated by all records, although small relative trends exacerbate the differences between records. Inadequacies in the observations and analytical methods result in structural uncertainties that still contribute to the differences between surface and tropospheric temperature trends, and revisions continue to be made. Changes in the height of the tropopause since 1979 are consistent with overall tropospheric warming as well as stratospheric cooling.

3.4.2.1 Surface and Lower-Tropospheric Water Vapour

Boundary layer moisture strongly determines the longwave (LW) radiative flux from the atmosphere to the surface. It also accounts for a significant proportion of the direct absorption of solar radiation by the atmosphere. The TAR reported widespread increases in surface water vapour in the NH. The overall sign of these trends has been confirmed from analysis of specific humidity over the USA ( Robinson, 2000 [JoC] ) and over China from 1951 to 1994 ( (Wang and Gaffen, 2001 ) ), particularly for observations made at night. Differences in the spatial, seasonal and diurnal patterns of these changes were found with strong sensitivity of the results to the network choice. Philipona et al. 2004 [JoC, MoS] ) inferred rapid increases in surface water vapour over central Europe from cloud-cleared LW radiative flux measured over the period 1995 to 2003 . Subsequent analyses ( Philipona et al., 2005 [JoC, MoS] ) confirmed that changes in integrated water vapour for this region are strongly coupled to the surface temperature, with regions of warming experiencing increasing moisture and regions of cooling experiencing decreasing moisture. For central Europe, Auer et al. 2007 [JoC, ARC, 2007] ) demonstrated increasing moisture trends. Their vapour pressure series from the Greater Alpine Region closely follow the decadal- to centennial-scale warming at both urban lowland and rural summit sites. In Canada, van Wijngaarden and Vincent 2005 [JoC] ) found a decrease in relative humidity of several percent in the spring for 75 stations, after correcting for instrumentation changes, but little change in relative humidity elsewhere or for other seasons. Ishii et al. 2005 [JoC, SRC] ) reported that globally averaged dew points over the ocean have risen by about 0.25°C between 1950 and 2000 . Increasing extremes in summer dew points, and increased humidity during summer heat waves, were found at three stations in northeastern Illinois ( Sparks et al., 2002 [JoC] ;( Changnon et al., 2003 ) ) and attributed in part to changes in agricultural practices in the region.

Dai 2006 [Ambiguous] ) analysed near-global (60°S–75°N) synoptic data for 1976 to 2005 from ships and buoys and more than 15,000 land stations for specific humidity, temperature and relative humidity. Nighttime relative humidity was found to be greater than daytime by 2 to 15% over most land areas, as temperatures undergo a diurnal cycle, while moisture does not change much. The global trends of near-surface relative humidity are very small. Trends in specific humidity tend to follow surface temperature trends with a global average increase of 0.06 g kg^–1 per decade ( 1976 – 2004 ). The rise in specific humidity corresponds to about 4.9% per 1°C warming over the globe. Over the ocean, the observed surface specific humidity increases at 5.7% per 1°C warming, which is consistent with a constant relative humidity. Over land, the rate of increase is slightly smaller (4.3% per 1°C), suggesting a modest reduction in relative humidity as temperatures increase, as expected in water-limited regions.

For the lower troposphere, water vapour information has been available from the TOVS since 1979 and from the Scanning Multichannel Microwave Radiometer (SMMR) from 1979 to 1984 . However, the main improvement occurred with the introduction of the Special Sensor Microwave/Imager (SSM/I) in mid- 1987 ( Wentz and Schabel, 2000 [JoC] ). Retrievals of column-integrated water vapour from SSM/I are generally regarded as providing the most reliable measurements of lower-tropospheric water vapour over the oceans, although issues pertaining to the merging of records from successive satellites do arise ( Trenberth et al., 2005a [PoC, JoC, SRC] ; Sohn and Smith, 2003 [JoC] ).

Significant interannual variability of column-integrated water vapour has been observed using TOVS, SMMR and SSM/I data. In particular, column water vapour over the tropical oceans increased by 1 to 2 mm during the 1982 – 1983, 1986 – 1987 and 1997 – 1998 El Niño events ( Soden and Schroeder, 2000 [JoC, MoS, ARC] ; Allan et al., 2003 [JoC, MoS] ; Trenberth et al., 2005a [PoC, JoC, SRC] ) and decreased by a smaller magnitude in response to global cooling following the eruption of Mt. Pinatubo in 1991 ( Soden et al., 2002 [JoC, ARC] ; Trenberth and Smith, 2005 [PoC, JoC, SRC] ; see also Section 8.6.3.1 ). The linear trend based on monthly SSM/I data over the oceans was 1.2% per decade (0.40 ± 0.09 mm per decade) for 1988 to 2004 ( Figure 3.20 ). Since the trends are similar in magnitude to the interannual variability, it is likely that the latter affects the magnitude of the linear trends. The trends are overwhelmingly positive in spatial structure, but also suggestive of an ENSO influence. As noted by Trenberth et al. 2005a [PoC, JoC, SRC] ), most of the patterns associated with the interannual variability and linear trends can be reproduced from the observed SST changes over this period by assuming a constant relative humidity increase in water vapour mixing ratio. Given observed SST increases, this implies an overall increase in water vapour of order 5% over the 20th century and about 4% since 1970 .

Figure 3.20. Linear trends in precipitable water (total column water vapour) in % per decade (top) and monthly time series of anomalies relative to the 1988 to 2004 period in % over the global ocean plus linear trend (bottom), from RSS SSM/I (updated from Trenberth et al., 2005a [PoC, JoC, SRC] ).

An independent check on globally vertically integrated water vapour amounts is whether the change in water vapour mass is reflected in the surface pressure field, as this is the only significant influence on the global atmospheric mass to within measurement accuracies. As Trenberth and Smith 2005 [PoC, JoC, SRC] ) showed, such checks indicate considerable problems prior to 1979 in reanalyses, but results are in better agreement thereafter for ERA-40. Evaluations of column integrated water vapour from the NASA Water Vapor Project (NVAP; Randel et al., 1996 [JoC] ), and reanalysis data sets from NRA, NCEP-2 and ERA-15/ERA-40 (see Appendix 3.B.5.4) reveal several deficiencies and spurious trends, which limit their utility for climate monitoring ( Zveryaev and Chu, 2003 [JoC, MoS] ; Trenberth et al., 2005a [PoC, JoC, SRC] ; Uppala et al., 2005 [JoC, SRC] ). The spatial distributions, trends and interannual variability of water vapour over the tropical oceans are not always well reproduced by reanalyses, even after the 1970 s ( Allan et al., 2002 [Ambiguous] , 2004; Trenberth et al., 2005a [PoC, JoC, SRC] ).

To summarise, global, local and regional studies all indicate increases in moisture in the atmosphere near the surface, but highlight differences between regions and between day and night. Satellite observations of oceanic lower-tropospheric water vapour reveal substantial variability during the last two decades. This variability is closely tied to changes in surface temperatures, with the water vapour mass changing at roughly the same rate at which the saturated vapour pressure does. A significant upward trend is observed over the global oceans and some NH land areas, although the calculated trend is likely influenced by large interannual variability in the record.

3.4.2.2 Upper-Tropospheric Water Vapour

Water vapour in the middle and upper troposphere accounts for a large part of the atmospheric greenhouse effect and is believed to be an important amplifier of climate change ( Held and Soden, 2000 [ARC] ). Changes in upper-tropospheric water vapour in response to a warming climate have been the subject of significant debate.

Due to instrumental limitations, long-term changes in water vapour in the upper troposphere are difficult to assess. ( Wang et al. 2001 ) ) found an increasing trend of 1 to 5% per decade in relative humidity during 1976 to 1995, with the largest increases in the upper troposphere, using 17 radiosonde stations in the tropical west Pacific. Conversely, a combination of Microwave Limb Sounder (MLS) and Halogen Occultation Experiment (HALOE) measurements at 215 hPa suggested increases in water vapour with increasing temperature ( Minschwaner and Dessler, 2004 [JoC, MoS] ) on interannual time scales, but at a rate smaller than expected from constant relative humidity.

( Maistrova et al. 2003 ) ) reported an increase in specific humidity at 850 hPa and a decrease from 700 to 300 hPa for 1959 to 2000 in the Arctic, based on data from ships and temporary stations as well as permanent stations. In general, the radiosonde trends are highly suspect owing to the poor quality of, and changes over time in, the humidity sensors (e.g., ( Wang et al., 2002a ) ). Comparisons of water vapour sensors during recent intensive field campaigns have produced a renewed appreciation of random and systematic errors in radiosonde measurements of upper-tropospheric water vapour and of the difficulty in developing accurate corrections for these measurements ( Guichard et al., 2000 [JoC] ; Revercombe et al., 2003 [JoC, MoS] ;( Turner et al., 2003; ) Wang et al., 2003 [JoC] ;( Miloshevich et al., 2004; ) Soden et al., 2004 [JoC, ARC] ).

Information on the decadal variability of upper-tropospheric relative humidity is now provided by 6.7 µm thermal radiance measurements from Meteosat ( Picon et al., 2003 [JoC] ) and the High-resolution Infrared Radiation Sounder (HIRS) series of instruments flying on NOAA operational polar-orbiting satellites ( (Bates and Jackson, 2001; ) Soden et al., 2005 [JoC, ARC] ). These products rely on the merging of many different satellites to ensure uniform calibration. The HIRS channel 12 (T12) data have been most extensively analysed for variability and show linear trends in relative humidity of order ±1% per decade at various latitudes ( (Bates and Jackson, 2001 ) ), but these trends are difficult to separate from larger interannual fluctuations due to ENSO ( McCarthy and Toumi, 2004 [JoC] ) and are negligible when averaging over the tropical oceans ( Allan et al., 2003 [JoC, MoS] ).

In the absence of large changes in relative humidity, the observed warming of the troposphere (see Section 3.4.1 )implies that the specific humidity in the upper troposphere should have increased. As the upper troposphere moistens, the emission level for T12 increases due to the increasing opacity of water vapour along the satellite line of sight. In contrast, the emission level for the MSU T2 remains constant because it depends primarily on the concentration of oxygen, which does not vary by any appreciable amount. Therefore, if the atmosphere moistens, the brightness temperature difference (T2 − T12) will increase over time due to the divergence of their emission levels ( Soden et al., 2005 [JoC, ARC] ). This radiative signature of upper-tropospheric moistening is evident in the positive trends of T2 − T12 for the period 1982 to 2004 ( Figure 3.21 ). If the specific humidity in the upper troposphere had not increased over this period, the emission level for T12 would have remained unchanged and T2 − T12 would show little trend over this period (dashed line in Figure 3.21 ).

Figure 3.21. The radiative signature of upper-tropospheric moistening is given by upward linear trends in T2−T12 for 1982 to 2004 (0.1 ºC per decade; top) and monthly time series of the global-mean (80°N to 80°S) anomalies relative to 1982 to 2004 (ºC) and linear trend (dashed; bottom). Data are from the RSS T2 and HIRS T12 ( Soden et al., 2005 [JoC, ARC] ). The map is smoothed to spectral truncation T31 resolution.

Clear-sky outgoing longwave radiation (OLR) is also highly sensitive to upper-tropospheric water vapour and a number of scanning instruments have made well-calibrated but non-overlapping measurements since 1985 (see Section 3.4.3 ). Over this period, the small changes in clear-sky OLR can be explained by the observed temperature changes while maintaining a constant relative humidity ( Wong et al., 2000 [JoC, SRC] ; Allan and Slingo, 2002 [JoC, MoS] ) and changes in well-mixed greenhouse gases ( Allan et al., 2003 [JoC, MoS] ). This again implies a positive relationship between specific humidity and temperature in the upper troposphere.

To summarise, the available data do not indicate a detectable trend in upper-tropospheric relative humidity. However, there is now evidence for global increases in upper-tropospheric specific humidity over the past two decades, which is consistent with the observed increases in tropospheric temperatures and the absence of any change in relative humidity.

3.4.2.4 Stratospheric Water Vapour

The TAR noted an apparent increase of roughly 1% yr^–1 in stratospheric water vapour content (~0.05 ppm yr^–1)during the last half of the 20th century ( Kley et al., 2000 [NPR] ; Rosenlof et al., 2001 [JoC, SRC] ). This was based on data taken at mid-latitudes, and from multiple instruments. However, the longest series of data come from just two locations in North America with no temporal overlap. The combination of measurement uncertainties and relatively large variability on time scales from months to years warrants some caution when interpreting the longer-term trends ( Kley et al., 2000 [NPR] ; Fueglistaler and Haynes, 2005 [JoC] ). The moistening is more convincingly documented during the 1980 s and most of the 1990 s than earlier, due to a longer continuous record (the NOAA Climate Monitoring and Diagnostics Laboratory (CMDL) frost-point balloon record from Boulder, Colorado; Oltmans et al., 2000 [JoC] ) and the availability of satellite observations during much of this period. However, discrepancies between satellite- and balloon-measured variations are apparent at decadal time scales, largely over the latter half of the 1990 s ( Randel et al., 2004a [JoC] ).

An increase in stratospheric water vapour has important radiative and chemical consequences (see also Section 2.3.8 ). These may include a contribution to the recent observed cooling of the lower stratosphere and/or warming of the surface ( Forster and Shine, 1999 [JoC] , 2002; Smith et al., 2001 [JoC, MoS, ARC] ), although the exact magnitude is difficult to quantify ( Oinas et al., 2001 [JoC] ; Forster and Shine, 2002 [JoC] ). Some efforts to reconcile observed rates of cooling in the stratosphere with those expected based on observed changes in ozone and carbon dioxide (CO₂)since 1979 ( Langematz et al., 2003 [JoC] ; Shine et al., 2003 [JoC, MoS] ) have found discrepancies in the lower stratosphere consistent with an additional cooling effect of a stratospheric water vapour increase. However, Shine et al. 2003 [JoC, MoS] ) noted that because the water vapour observations over the period of consideration are not global in extent, significant uncertainties remain as to whether radiative effects of a water vapour change are a significant contributor to the stratospheric temperature changes. Moreover, other studies which account for uncertainties in the ozone profiles and temperature trends, and natural variability, can reconcile the observed stratospheric temperature changes without the need for sizable water vapour changes ( Ramaswamy and Schwarzkopf, 2002 [JoC, ARC] ; Schwarzkopf and Ramaswamy, 2002 [JoC, ARC] ).

Although methane oxidation is a major source of water in the stratosphere, and has been increasing over the industrial period, the noted stratospheric trend in water vapour is too large to attribute to methane oxidation alone ( Kley et al., 2000 [NPR] ; Oltmans et al., 2000 [JoC] ). Therefore, other contributors to an increase in stratospheric water vapour are under active investigation. It is likely that different mechanisms are affecting water vapour trends at different altitudes. Aviation emits a small but potentially significant amount of water vapour directly into the stratosphere ( IPCC, 1999 [NPR] ). Several indirect mechanisms have also been considered including: a) volcanic eruptions ( Considine et al., 2001 [JoC, MoS] ; Joshi and Shine, 2003 [JoC] ); b) biomass-burning aerosol ( Sherwood, 2002 [JoC, SRC] ; Andreae et al., 2004 [JoC] ); c) tropospheric sulphur dioxide ( Notholt et al., 2005 [JoC] ); and d) changes to methane oxidation rates from changes in stratospheric chlorine, ozone and the hydroxyl radical ( (Röckmann et al., 2004 ) ). Other proposed mechanisms relate to changes in tropopause temperatures or circulation ( Stuber et al., 2001 [JoC, ARC] ; Zhou et al., 2001 [JoC, ARC] ; Rosenlof, 2002 [SRC] ; Nedoluha et al., 2003 [JoC] ; Dessler and Sherwood, 2004 [JoC, SRC] ; Fueglistaler et al., 2004 [JoC] ;( Roscoe, 2004 ) ).

It has been assumed that temperatures near the tropical tropopause control stratospheric water vapour according to equilibrium thermodynamics, importing more water vapour into the stratosphere when temperatures are warmer. However, tropical tropopause temperatures have cooled slightly over the period of the stratospheric water vapour increase (see Section 3.4.1 ;Seidel et al., 2001 [JoC, ARC] ; Zhou et al, 2001 [JoC, ARC] ). This makes the mid-latitude lower-stratospheric increases harder to explain ( Fueglistaler and Haynes, 2005 [JoC] ). Satellite observations ( Read et al., 2004 [JoC] ) show water vapour injected above the tropical tropopause by deep convective clouds, bypassing the traditional control point. Changes in the amount of condensate sublimating in this layer may have contributed to the upward trend, but to what degree is uncertain ( Sherwood, 2002 [JoC, SRC] ). Another suggested source for temperature-independent variability is changes in the efficiency with which air is circulated through the coldest regions before entering the stratosphere ( Hatsushika and Yamazaki, 2003 [JoC] ; Bonnazola and Haynes, 2004 [JoC] ; Dessler and Sherwood, 2004 [JoC, SRC] ; Fueglistaler et al., 2004 [JoC] ). However, it is not yet clear that a circulation-based mechanism can explain the observed trend ( Fueglistaler and Haynes, 2005 [JoC] ).

The TAR noted a stalling of the upward trend in water vapour during the last few years of observations available at that time. This change in behaviour has persisted, with a near-zero trend in stratospheric water vapour between 1996 and 2000 ( Nedoluha et al., 2003 [JoC] ; Randel et al., 2004a [JoC] ). The upward trend of methane is also smaller and is currently close to zero (see Section 2.3.2 ). Further, at the end of 2000 there was a dramatic drop in water vapour in the tropical lower stratosphere as observed by both satellite and CMDL balloon data ( Randel et al., 2004a [JoC] ). Temperatures observed near the tropical tropopause also dropped, but the processes producing the tropical tropopause cooling itself are currently not fully understood. The propagation of this recent decrease through the stratosphere should ensure flat or decreasing stratospheric moisture for at least the next few years.

To summarise, water vapour in the stratosphere has shown significant long-term variability and an apparent upward trend over the last half of the 20th century but with no further increases since 1996 . It does not appear that this behaviour is a straightforward consequence of known climate changes. Although ideas have been put forward, there is no consensus as to what caused either the upward trend or its recent disappearance.

3.4.3.1 Surface Cloud Observations

As noted in the TAR and extended with more recent studies, surface observations suggest increased total cloud cover since the middle of the last century over many continental regions including the USA ( Sun, 2003 [NPR, JoC] ; Groisman et al., 2004 [JoC, SRC] ; Dai et al., 2006 [Ambiguous] ), the former USSR ( Sun and Groisman, 2000 [JoC, SRC] ; Sun et al., 2001 [JoC, SRC] ), Western Europe, mid-latitude Canada, and Australia ( Henderson-Sellers, 1992 [ARC] ). This increasing cloudiness since 1950 is consistent with an increase in precipitation and a reduction in DTR ( Dai et al., 2006 [Ambiguous] ). However, decreasing cloudiness over this period has been reported over China ( Kaiser, 1998 [JoC] ), Italy ( Maugeri et al., 2001 [JoC] ) and over central Europe ( Auer et al., 2007 [JoC, ARC, 2007] ). If the analyses are restricted to after about 1971, changes in continental cloud cover become less coherent. For example, using a worldwide analysis of cloud data ( Hahn and Warren, 2003 [NPR, SRC] ; Minnis et al., 2004 [JoC] ) regional reductions were found since the early 1970 s over western Asia and Europe but increases over the USA.

Changes in total cloud cover along with an estimate of precipitation over global and hemispheric land (excluding North America) from 1976 to 2003 are shown in Figure 3.22 .During this period, secular trends over land are small. The small variability evident in land cloudiness appears to be correlated with precipitation changes, particularly in the SH( Figure 3.22 ). Note that surface observations from North America are excluded from this figure due to the declining number of human cloud observations since the early 1990 s over the USA and Canada, as human observers have been replaced with Automated Surface Observation Systems (ASOS) from which cloud amounts are less reliable and incompatible with previous records ( Dai et al., 2006 [Ambiguous] ). However, independent human observations from military stations suggest an increasing trend (~1.4% of sky per decade) in total cloud cover over the USA.

Figure 3.22. Annual total land (excluding the USA and Canada) cloud cover (black) and precipitation (red) anomalies from 1976 to 2003 over global (60°S–75°N), NH and SH regions, with the correlation coefficient (r) shown at the top. The cloud cover is derived by gridding and area-averaging synoptic observations and the precipitation is updated from Chen et al. 2002 [JoC] ). Typical 5 to 95% error bars for each decade are estimates using inter-grid-box variations (from Dai et al., 2006 [Ambiguous] ).

The TAR also noted multi-decadal trends in cloud cover over the ocean. An updated analysis of this information ( Norris, 2005a [JoC, MoS, SRC] ) documented substantial decadal variability and decreasing trends in upper-level cloud cover over mid- and low-latitude oceans since 1952 . However, there are no direct observations of upper-level clouds from the surface and instead Norris 2005a [JoC, MoS, SRC] ) infers them from reported total and low cloud cover assuming a random overlap. These results partially reverse the finding of increasing trends in mid-level cloud amount in the northern mid-latitude oceans that was reported in the TAR, although the new study does not distinguish between high and middle clouds. Norris 2005b [JoC, SRC] ) found that upper-level cloud cover had increased over the equatorial South Pacific between 1952 and 1997 and decreased over the adjacent subtropical regions, the tropical Western Pacific, and the equatorial Indian Ocean. This pattern is consistent with decadal changes in precipitation and atmospheric circulation over these regions noted in the TAR, which further supports their validity. Deser et al. 2004 [JoC, SRC] ) found similar spatial patterns in inter-decadal variations in total cloud cover, SST and precipitation over the tropical Pacific and Indian Oceans during 1900 to 1995 . In contrast, low-cloud cover increased over almost all of the tropical Indian and Pacific Oceans, but this increase bears little resemblance to changes in atmospheric circulation over this period, suggesting that it may be spurious ( Norris, 2005b [JoC, SRC] ). When averaged globally, oceanic cloud cover appears to have increased over the last 30 years or more (e.g., Ishii et al., 2005 [JoC, SRC] ).

During El Niño events, cloud cover generally decreases over land throughout much of the tropics and subtropics, but increases over the ocean in association with precipitation changes ( Curtis and Adler, 2003 [JoC, SRC] ). Multi-decadal variations are affected by the 1976 – 1977 climate shift ( Deser et al., 2004 [JoC, SRC] ), and these dominate the low-latitude trends from 1971 to 1996 found in Hahn and Warren 2003 [NPR, SRC] ).

3.4.3.2 Satellite Cloud Observations

Since the TAR, there has been considerable effort in the development and analysis of satellite data sets for documenting changes in global cloud cover over the past few decades. The most comprehensive cloud climatology is that of the International Satellite Cloud Climatology Project (ISCCP), begun in July 1983 . The ISCCP shows an increase in globally averaged total cloud cover of about 2% from 1983 to 1987, followed by a decline of about 4% from 1987 to 2001 ( Rossow and Dueñas, 2004 [JoC] Cess and Udelhofen 2003 [JoC] ) documented decreasing ISCCP total cloud cover in all latitude zones between 40°S and 40°N. Norris 2005a [JoC, MoS, SRC] ) found that both ISCCP and ship synoptic reports show consistent reductions in middle- or high-level cloud cover from the 1980 s to the 1990 s over low- and mid-latitude oceans. Minnis et al. 2004 [JoC] ) also found consistent trends in high-level cloud cover between ISCCP and surface observations over most areas, except for the North Pacific where they differed by almost 2% per decade. In addition, an analysis of Stratospheric Aerosol and Gas Experiment II (SAGE II) data revealed a decline in cloud frequency above 12 km between 1985 and 1998 ( Wang et al., 2002b [JoC] ) that is consistent with the decrease in upper-level cloud cover noted in ISCCP and ocean surface observations. The decline in upper-level cloud cover since 1987 may also be consistent with a decrease in reflected shortwave (SW) radiation during this period as measured by the Earth Radiation Budget Satellite (ERBS; see Section 3.4.4 ). Radiative transfer calculations, which use the ISCCP cloud properties as input, are able to independently reproduce the decadal changes in outgoing LW and reflected SW radiation reported by ERBS ( Zhang et al., 2004c [JoC, MoS] ).

Analyses of the spatial trends in ISCCP cloud cover reveal changing biases arising from changes in satellite view angle and coverage that affect the global mean anomaly time series ( Norris, 2000 [SRC] ; Dai et al., 2006 [Ambiguous] ). The ISCCP spurious variability may occur primarily in low-level clouds with the least optical thickness (the ISCCP ‘cumulus’ category; Norris, 2005a [JoC, MoS, SRC] ), due to discontinuities in satellite view angles associated with changes in satellites. Such biases likely contribute to ISCCP’s negative cloud cover trend, although their magnitude and impact on radiative flux calculations using ISCCP cloud data are not yet known. Additional artefacts, including radiometric noise, navigation and rectification errors are present in the ISCCP data ( Norris, 2000 [SRC] ), but the effects of known and unknown artefacts on ISCCP cloud and flux data have not yet been quantified.

Other satellite data sets show conflicting decadal changes in total cloud cover. For example, analysis of cloud cover changes from the HIRS shows a slight increase in cloud cover between 1985 and 2001 ( Wylie et al., 2005 [JoC] ). However, spurious changes have also been identified in the HIRS data set, which may affect its estimates of decadal variability. One important source of uncertainty results from the drift in Equatorial Crossing Time (ECT) of polar-orbiting satellite measurements (e.g., HIRS and the Advanced Very High Resolution Radiometer; AVHRR), which aliases the large diurnal cycle of clouds into spurious lower-frequency variations. After correcting for ECT drift and other small calibration errors in AVHRR measurements of cloudiness, Jacobowitz et al. 2003 [JoC] ) found essentially no trend in cloud cover for the tropics from 1981 to 2000 .

While the variability in surface-observed upper-level cloud cover has been shown to be consistent with that observed by ISCCP ( Norris, 2005a [JoC, MoS, SRC] ), the variability in total cloud cover is not, implying differences between ISCCP and surface-observed low cloud cover. Norris 2005a [JoC, MoS, SRC] ) shows that even after taking into account the difference between surface and satellite views of low-level clouds, the decadal changes between the ISCCP and surface data sets still disagree. The extent to which this results from differences in spatial and temporal sampling or differences in viewing perspective is unclear.

In summary, while there is some consistency between ISCCP, ERBS, SAGE II and surface observations of a reduction in high cloud cover during the 1990 s relative to the 1980 s, there are substantial uncertainties in decadal trends in all data sets and at present there is no clear consensus on changes in total cloudiness over decadal time scales.

3.4.4.1 Top-of-Atmosphere Radiation

One important development since the TAR is the apparent unexpectedly large changes in tropical mean radiation flux reported by ERBS ( Wielicki et al., 2002a [JoC, SRC] ,b). It appears to be related in part to changes in the nature of tropical clouds ( Wielicki et al., 2002a [JoC, SRC] ), based on the smaller changes in the clear-sky component of the radiative fluxes ( Wong et al., 2000 [JoC, SRC] ; Allan and Slingo, 2002 [JoC, MoS] ), and appears to be statistically distinct from the spatial signals associated with ENSO ( Allan and Slingo, 2002 [JoC, MoS] ; Chen et al., 2002 [JoC] ). A recent reanalysis of the ERBS active-cavity broadband data corrects for a 20 km change in satellite altitude between 1985 and 1999 and changes in the SW filter dome ( Wong et al., 2006 [JoC, SRC] ). Based upon the revised (Edition 3_Rev1) ERBS record( Figure 3.23 ), outgoing LW radiation over the tropics appears to have increased by about 0.7 Wm^–2 while the reflected SW radiation decreased by roughly 2.1 Wm^–2 from the 1980 s to 1990 s( Table 3.5 ).

Figure 3.23. Tropical mean (20°S to 20°N) TOA flux anomalies from 1985 to 1999 (Wm^–2)for LW, SW, and NET radiative fluxes [NET = −(LW + SW)]. Coloured lines are observations from ERBS Edition 3_Rev1 data from Wong et al. 2006 [JoC, SRC] ) updated from Wielicki et al. 2002a [JoC, SRC] ), including spacecraft altitude and SW dome transmission corrections.

These conclusions depend upon the calibration stability of the ERBS non-scanner record, which is affected by diurnal sampling issues, satellite altitude drifts and changes in calibration following a three-month period when the sensor was powered off ( Trenberth, 2002 [PoC, JoC, SRC] ). Moreover, rather than a trend, the reflected SW radiation change may stem mainly from a jump in late 1992 in the ERBS record that is also observed in the ISCCP (version FD) record ( Zhang et al., 2004c [JoC, MoS] ) but not in the AVHRR Pathfinder record ( Jacobowitz et al., 2003 [JoC] ). However, careful inspection of the sensor calibration revealed no known issues that can explain the decadal shift in the fluxes despite corrections to the ERBS time series relating to diurnal aliasing and satellite altitude changes ( Wielicki et al., 2002b [JoC, SRC] ; Wong et al., 2006 [JoC, SRC] ).

As noted in Section 3.4.3 ,the low-latitude changes in the radiation budget appear consistent with reduced cloud fraction from ISCCP. Detailed radiative transfer computations, using ISCCP cloud products along with additional global data sets, show broad agreement with the ERBS record of tropical radiative fluxes ( (Hatzianastassiou et al., 2004; ) Zhang et al., 2004c [JoC, MoS] ; Wong et al., 2006 [JoC, SRC] ). However, the decrease in reflected SW radiation from the 1980 s to the 1990 s may be inconsistent with the increase in total and low cloud cover over oceans reported by surface observations ( Norris, 2005a [JoC, MoS, SRC] ), which show increased low cloud occurrence. The degree of inconsistency, however, is difficult to ascertain without information on possible changes in low-level cloud albedo.

While the ERBS satellite provides the only continuous long-term top-of-atmosphere (TOA) flux record from broadband active-cavity instruments, narrow spectral band radiometers have made estimates of both reflected SW and outgoing LW radiation trends using regressions to broadband data, or using radiative transfer theory to estimate unmeasured portions of the spectrum of radiation. Table 3.5 shows the 1980 s to 1990 s TOA tropical mean flux changes for the ERBS Edition 3 data ( Wong et al., 2006 [JoC, SRC] ), the HIRS Pathfinder data ( Mehta and Susskind, 1999 [JoC] ), the AVHRR Pathfinder data ( Jacobowitz et al., 2003 [JoC] ) and the ISCCP FD data ( Zhang et al., 2004c [JoC, MoS] ).

The most accurate of the data sets in Table 3.5 is believed to be the ERBS Edition 3 Rev 1 active-cavity wide field of view data ( Wielicki et al., 2005 [JoC, SRC] ). The ERBS stability is estimated as better than 0.5 Wm^–2 over the 1985 to 1999 period and the spatial and temporal sampling noise is less than 0.5 Wm^–2 on annual time scales ( Wong et al., 2006 [JoC, SRC] ). The outgoing LW radiation changes from ERBS are similar to the decadal changes in the HIRS Pathfinder and ISCCP FD records, but disagree with the AVHRR Pathfinder data ( Wong et al., 2006 [JoC, SRC] ). The AVHRR Pathfinder data also do not support the TOA SW radiation trends. However, calibration issues, conversion from narrow to broadband, and satellite orbit changes are thought to render the AVHRR record less reliable for decadal changes compared to ERBS ( Wong et al., 2006 [JoC, SRC] ). Estimates of the stability of the ISCCP time series for long-term TOA flux records are 3 to 5 Wm^–2 for SW radiative flux and 1 to 2 Wm^–2 for LW radiative flux ( (Brest et al., 1997 ) ), although the time series agreement of the ISCCP and ERBS records are much closer than these estimated calibration drift uncertainties ( Zhang et al., 2004c [JoC, MoS] ).

Table 3.5. Top-of-atmosphere (TOA) radiative flux changes from the 1980 s to 1990 s (Wm^–2). Values are given as tropical means (20°S to 20°N) for the 1994 to 1997 period minus the 1985 to 1989 period. Dashes are shown where no data are available. From Wong et al. 2006 [JoC, SRC] ).

The changes in SW radiation measured by ERBS Edition 3 Rev 1 are larger than the clear-sky flux changes due to humidity variations ( Wong et al., 2000 [JoC, SRC] ) or anthropogenic radiative forcing (see Chapter2 ). If correct, the large decrease in reflected SW radiation with little change in outgoing LW radiation implies a reduction in tropical low cloud cover over this period. However, specific information on cloud radiative forcing is not available from ERBS after 1989 and, as noted in Section 3.4.3 ,surface data sets suggest an increase in low cloud cover over this period.

Since most of the net tropical heating of 1.4 Wm^–2 is a decrease in reflected SW radiative flux, the change implies a similar increase in solar insolation at the surface that, if unbalanced by other changes in surface fluxes, would increase the amount of ocean heat storage. Wong et al. 2006 [JoC, SRC] ) showed that the changes in global net radiation are consistent with a new ocean heat-storage data set from Willis et al. 2004 [JoC, ARC] ; see Chapter5 and Figure 5.1 ). Differences between the two data sets are roughly 0.4 Wm^–2,in agreement with the estimated annual sampling noise in the ocean heat-storage data.

Using astronomical observations of visible wavelength solar photons reflected from parts of the Earth to the moon and then back to the Earth at a surface-based observatory, Pallé et al. 2004 [JoC] ) estimated a dramatic increase of Earth-reflected SW radiative flux of 5.5 Wm^–2 over three years. This is unlikely to be real, as over the same time period ( 2000 – 2003 ), the Clouds and the Earth’s Radiant Energy System (CERES) broadband data indicate a decrease in SW radiative flux of almost 1 Wm^–2,which is much smaller and the opposite sign ( Wielicki et al., 2005 [JoC, SRC] ). In addition, changes in ocean heat storage are more consistent with the CERES data than with the Earthshine indirect observation.

The only long-term time series ( 1979 – 2001 ) of energy divergence in the atmosphere ( Trenberth and Stepaniak, 2003b [PoC, JoC, SRC] ) are based on NRA, which, although not reliable for depicting trends, are reliable on interannual times scales for which they show substantial variability associated with ENSO. Analyses by Trenberth and Stepaniak 2003b [PoC, JoC, SRC] ) reveal more divergence of energy out of the deep tropics in the 1990 s compared with the 1980 s due to differences in ENSO, which may account for at least some of the changes discussed above.

In summary, although there is independent evidence for decadal changes in TOA radiative fluxes over the last two decades, the evidence is equivocal. Changes in the planetary and tropical TOA radiative fluxes are consistent with independent global ocean heat-storage data, and are expected to be dominated by changes in cloud radiative forcing. To the extent that they are real, they may simply reflect natural low-frequency variability of the climate system.

3.4.4.2 Surface Radiation

The energy balance at the surface requires net radiative heating to be balanced by turbulent energy fluxes and thus determines the evolution of surface temperature and the cycling of water, which are key parameters of climate change (see Box 7.1 ). In recent years, several studies have focused on observational evidence of changing surface radiative heating. Reliable SW radiative measurement networks have existed since the 1957 – 1958 International Geophysical Year.

Areduction in downward solar radiation (‘dimming’) of about 1.3% per decade or about 7 Wm^–2 was observed from 1961 to 1990 at land stations around the world ( Gilgen et al., 1998 [JoC, MoS, ARC] ; Liepert, 2002 [JoC, SRC] ). Additional studies also found declines in surface solar radiation in the Arctic and Antarctic ( (Stanhill and Cohen, 2001 ) ) as well as at sites in the former Soviet Union ( Russak, 1990 [JoC] ; Abakumova et al., 1996 [JoC] ), around the Mediterranean Sea ( (Aksoy, 1997; ) ( Omran, 2000 ) ), China ( (Ren et al., 2005 ) ), the USA ( Liepert, 2002 [JoC, SRC] ) and southern Africa ( Power and Mills, 2005 [JoC] ( Stanhill and Cohen 2001 ) ) claim an overall globally averaged reduction of 2.7% per decade but used only 30 records. However, the stations where these analyses took place are quite limited in domain and dominated by large urban areas, and the dimming is much less at rural sites ( Alpert et al., 2005 [JoC] ) or even missing altogether over remote areas, except for identifiable effects of volcanic eruptions, such as Mt. Pinatubo in 1991 ( Schwartz, 2005 [JoC] ). At the majority of 421 analysed sites, the decline in surface solar radiation ended around 1990 and a recovery of about 6 Wm^–2 occurred afterwards ( Wild et al., 2004 [JoC, ARC] ; 2005 ). The increase in surface solar radiation (‘brightening’) agrees with satellite and surface observations of reduced cloud cover ( Wang et al., 2002b [JoC] ; Wielicki et al., 2002a [JoC, SRC] ; Rossow and Dueñas, 2004 [JoC] ; Norris, 2005b [JoC, SRC] ; Pinker et al., 2005 [JoC] ), although there is evidence that some of these changes are spurious (see Section 3.4.3 ). In addition, the satellite-observed increase in surface radiation noted by Pinker et al. 2005 [JoC] ) occured primarily over ocean, whereas the increase observed by Wild et al. 2005 [JoC, ARC] ) was restricted to land stations.

From 1981 to 2003 over central Europe, Philipona and Dürr 2004 [JoC, MoS] ) showed that decreases in surface solar radiation from increases in clouds were cancelled by opposite changes in LW radiation and that increases in net radiative flux were dominated by the clear-sky LW radiation component relating to an enhanced water vapour greenhouse effect. Alpert et al. 2005 [JoC] ) provided evidence that a significant component of the reductions may relate to increased urbanisation and anthropogenic aerosol concentrations over the period (see also Section 7.5 ). This has been detected in solar radiation reductions for polluted regions (e.g., China; Luo et al., 2001 [JoC, ARC] ), but cloudiness changes must also play a major role, as shown for European sites and the USA ( Liepert, 2002 [JoC, SRC] ; Dai et al., 2006 [Ambiguous] ). In the USA increasing cloud optical thickness and a shift from cloud-free to more cloudy skies are the dominating factors compared to the aerosol direct effects. Possible causes of the 1990 s reversal are reduced cloudiness and increased cloud-free atmospheric transparency due to the reduction of anthropogenic aerosol concentrations and recovery from the effects of the 1991 eruption of Mt. Pinatubo. See Box 3.2 for more discussion and a likely explanation of these aspects.

3.6.2.1 El Niño-Southern Oscillation

El Niño-Southern Oscillation events are a coupled ocean-atmosphere phenomenon. El Niño involves warming of tropical Pacific surface waters from near the International Date Line to the west coast of South America, weakening the usually strong SST gradient across the equatorial Pacific, with associated changes in ocean circulation. Its closely linked atmospheric counterpart, the Southern Oscillation (SO), involves changes in trade winds, tropical circulation and precipitation. Historically, El Niño events occur about every 3 to 7 years and alternate with the opposite phases of below-average temperatures in the eastern tropical Pacific (La Niña). Changes in the trade winds, atmospheric circulation, precipitation and associated atmospheric heating set up extratropical responses. Wavelike extratropical teleconnections are accompanied by changes in the jet streams and storm tracks in mid-latitudes ( Chang and Fu, 2002 [JoC, SRC] ).

The El Niño-Southern Oscillation has global impacts, manifested most strongly in the northern winter months (November–March). Anomalies in MSLP are much greater in the extratropics while the tropics feature large precipitation variations. Associated patterns of surface temperature and precipitation anomalies around the globe are given in Figure 3.27 ( Trenberth and Caron, 2000 [PoC, JoC, SRC] ), and the evolution of these patterns and links to global mean temperature perturbations are given by Trenberth et al. 2002b [PoC, JoC, SRC] ).

Figure 3.27. Correlations with the SOI, based on normalised Tahiti minus Darwin sea level pressures, for annual (May to April) means for sea level pressure (top left) and surface temperature (top right) for 1958 to 2004, and GPCP precipitation for 1979 to 2003 (bottom left), updated from Trenberth and Caron 2000 [PoC, JoC, SRC] ). The Darwin-based SOI, in normalized units of standard deviation, from 1866 to 2005 ( Können et al., 1998 [JoC] ; lower right) features monthly values with an 11-point low-pass filter, which effectively removes fluctuations with periods of less than eight months ( Trenberth, 1984 [PoC, JoC, SRC] ). The smooth black curve shows decadal variations (see Appendix 3.A ). Red values indicate positive sea level pressure anomalies at Darwin and thus El Niño conditions.

The nature of ENSO has varied considerably over time. Strong ENSO events occurred from the late 19th century through the first 25 years of the 20th century and again after about 1950, but there were few events of note from 1925 to 1950 with the exception of the major 1939 – 1941 event( Figure 3.27 ). The 1976 – 1977 climate shift ( Trenberth, 1990 [PoC, JoC, SRC] ; see Figure 3.27 and Section 3.6.3 , Figure 3.28 )was associated with marked changes in El Niño evolution ( Trenberth and Stepaniak, 2001 [PoC, JoC, SRC] ), a shift to generally above-normal SSTs in the eastern and central equatorial Pacific and a tendency towards more prolonged and stronger El Niños. Since the TAR, there has been considerable work on decadal and longer-term variability of ENSO and Pacific climate. Such decadal atmospheric and oceanic variations( Section 3.6.3 )are more pronounced in the North Pacific and across North America than in the tropics but are also present in the South Pacific, with evidence suggesting they are at least in part forced from the tropics ( Deser et al., 2004 [JoC, SRC] ).

Figure 3.28. Pacific Decadal Oscillation: (top) SST based on the leading EOF SST pattern for the Pacific basin north of 20°N for 1901 to 2004 (updated; see Mantua et al., 1997 [JoC] ; Power et al., 1999b [JoC, ARC] ) and projected for the global ocean (units are nondimensional); and (bottom) annual time series (updated from Mantua et al., 1997 [JoC] ). The smooth black curve shows decadal variations (see Appendix 3.A ).

El Niño-Southern Oscillation events involve large exchanges of heat between the ocean and atmosphere and affect global mean temperatures. The 1997 – 1998 event was the largest on record in terms of SST anomalies and the global mean temperature in 1998 was the highest on record (at least until 2005 Trenberth et al. 2002b [PoC, JoC, SRC] ) estimated that global mean surface air temperatures were 0.17°C higher for the year centred on March 1998 owing to the El Niño. Extremes of the hydrological cycle such as floods and droughts are common with ENSO and are apt to be enhanced with global warming ( Trenberth et al., 2003 [PoC, JoC, SRC] ). For example, the modest 2002 – 2003 El Niño was associated with a drought in Australia, made much worse by record-breaking heat ( Nicholls, 2004 [JoC, SRC] ; and see Section 3.8.4 , Box 3.6 ). Thus, whether observed changes in ENSO behaviour are physically linked to global climate change is a research question of great importance.

3.6.2.2 Tropical-Extratropical Teleconnections: PNA and PSA

Circulation variability over the extratropical Pacific features wave-like patterns emanating from the subtropical western Pacific, characteristic of Rossby wave propagation associated with anomalous tropical heating ( Horel and Wallace, 1981 [JoC] ; Hoskins and Karoly, 1981 [JoC, MoS, SRC] ). These are known as the PNA and Pacific-South American (PSA) patterns and can arise naturally through atmospheric dynamics as well as in response to heating. Over the NH in winter, the PNA pattern lies across North America from the subtropical Pacific, with four centres of action( Figure 3.26 ). While the PNA pattern can be illustrated by taking a single point correlation, this is not so easy for the PSA pattern (not shown), as its spatial centres of action are not fixed. However, the PSA pattern can be present at all times of year, lying from Australasia over the southern Pacific and Atlantic ( Mo and Higgins, 1998 [JoC] ; Kidson, 1999 [JoC] ; Mo, 2000 [JoC] ).

The PNA, or a variant of it ( Straus and Shukla, 2002 [JoC, ARC] ), is associated with modulation of the Aleutian Low, the Asian jet, and the Pacific storm track, affecting precipitation in western North America and the frequency of Alaskan blocking events and associated cold air outbreaks over the western USA in winter ( Compo and Sardeshmukh, 2004 [JoC, MoS] ). The PSA is associated with modulation of the westerlies over the South Pacific, effects of which include significant rainfall variations over New Zealand, changes in the nature and frequency of blocking events across the high-latitude South Pacific, and interannual variations in antarctic sea ice across the Pacific and Atlantic sectors ( Renwick and Revell, 1999 [JoC] ; Kwok and Comiso, 2002a [JoC, ARC] ; Renwick, 2002 [JoC] ). While both PNA and PSA activity have varied with decadal modulation of ENSO, no systematic changes in their behaviour have been reported.

3.6.7.1 Antarctic Circumpolar Wave

The Antarctic Circumpolar Wave (ACW) is described as a pattern of variability with an approximately four-year period in the southern high-latitude ocean-atmosphere system, characterised by the eastward propagation of anomalies in antarctic sea ice extent, and coupled to anomalies in SST, sea surface height, MSLP and wind ( Jacobs and Mitchell, 1996 [JoC] ; White and Peterson, 1996 [JoC] ; White and Annis, 2004 [JoC] ). Since its initial formulation ( White and Peterson, 1996 [JoC] ), questions have arisen concerning many aspects of the ACW: the robustness of the ACW on inter-decadal time scales ( Carril and Navarra, 2001 [JoC] ; Connolley, 2003 [JoC] ; Simmonds, 2003 [JoC, SRC] ), its generating mechanisms ( Cai and Baines, 2001 [JoC, MoS, ARC] ; Venegas, 2003 [JoC] ; White et al., 2004 [Ambiguous] ; White and Simmonds, 2006 [JoC, SRC] ) and even its very existence ( Park et al., 2004 [JoC] ).

3.6.7.2 Indian Ocean Dipole

Large interannual variability of SST in the Indian Ocean has been associated with the Indian Ocean Dipole (IOD), also referred to as the Indian Ocean Zonal Mode (IOZM; Saji et al., 1999 [JoC] ; Webster et al., 1999 [JoC] ). This pattern manifests through a zonal gradient of tropical SST, which in one extreme phase in boreal autumn shows cooling off Sumatra and warming off Somalia in the west, combined with anomalous easterlies along the equator. The magnitude of the secondary rainfall maximum from October to December in East Africa is strongly correlated with positive IOD events ( Xie et al., 2002 [JoC, ARC] ). Several recent IOD events have occurred simultaneously with ENSO events and there is a significant debate on whether the IOD is an Indian Ocean pattern or whether it is triggered by ENSO in the Pacific Ocean ( (Allan et al., 2001 ) ). The strongest IOD episode ever observed occurred in 1997 to 1998 and was associated with catastrophic flooding in East Africa. Trenberth et al. 2002b [PoC, JoC, SRC] ) showed that Indian Ocean SSTs tend to rise about five months after the peak of ENSO in the Pacific. Monsoon variability and the SAM ( Lau and Nath, 2004 [JoC, MoS, ARC] ) are also likely to play a role in triggering or intensifying IOD events. One argument for an independent IOD was the large episode in 1961 when no ENSO event occurred ( Saji et al., 1999 [JoC] Saji and Yamagata 2003 [JoC] ), analysing observations from 1958 to 1997, concluded that 11 out of the 19 episodes identified as moderate to strong IOD events occurred independently of ENSO. However, this was disputed by ( Allan et al. 2001 ) ), who found that accounting for varying lag correlations removes the apparent independence from ENSO. Decadal variability in correlations between SST-based indices of the IOD and ENSO has been documented ( Clark et al., 2003 [JoC, ARC] ). At inter-decadal time scales, the SST patterns associated with the inter-decadal variability of ENSO indices are very similar to the SST patterns associated with the Indian monsoon rainfall ( Krishnamurthy and Goswami, 2000 [JoC] ) and with the North Pacific inter-decadal variability ( Deser et al., 2004 [JoC, SRC] ), raising the issue of coupled mechanisms modulating both ENSO-monsoon system and IOD variability (e.g., Terray et al., 2005 [JoC] ).

ENSO has exhibited considerable inter-decadal variability in the past century, in association with the PDO (or IPO). Systematic changes in ENSO behaviour have also been observed, in particular the different evolution of ENSO events and enhanced El Niño activity since the 1976 – 1977 climate shift. Over North America, ENSO- and PNA-related changes appear to have led to contrasting changes across the continent, as the west has warmed more than the east, while the latter has become cloudier and wetter. Over the Indian Ocean, ENSO, monsoon and SAM variability are related to a zonal gradient of tropical SST associated with anomalous easterlies along the equator, and opposite precipitation and thermal anomalies in East Africa and over the Maritime Continent. The tropical Pacific variability is influenced by interactions with the tropical Atlantic and Indian Oceans, and by the extratropical North and South Pacific. Responses of the extratropical ocean become more important as the time scale is extended, and processes such as subduction, gyre changes and the THC come into play.

3.8.2.1 Temperature

For temperature extremes in the 20th century, the TAR highlighted the lengthening of the growing or frost-free season in most mid- and high-latitude regions, a reduction in the frequency of extreme low monthly and seasonal average temperatures and smaller increases in the frequency of extreme high average temperatures. In addition, there was evidence to suggest a decrease in the intra-annual temperature variability with consistent reductions in frost days and increases in warm nighttime temperatures across much of the globe.

Evidence for changes in observed interannual variability (such as standard deviations of seasonal averages) is still sparse. Scherrer et al. 2005 [JoC, ARC] ) investigated standardised distribution changes for seasonal mean temperature in central Europe and found that temperature variability showed a weak increase (decrease) in summer (winter) for 1961 to 2004, but these changes are not statistically significant at the 10% level. On the daily time scale, regional studies have been completed for southern South America ( Vincent et al., 2005 [JoC] ), Central America and northern South America ( Aguilar et al., 2005 [JoC] ), the Caribbean ( Peterson et al., 2002 [JoC, ARC] ), North America ( Kunkel et al., 2004 [JoC, SRC] ;( Vincent and Mekis, 2006 ) ), the Arctic ( Groisman et al., 2003 [NPR, SRC] ), central and northern Africa ( Easterling et al., 2003 [JoC, SRC] ), southern and western Africa ( New et al., 2006 [JoC, ARC] ), the Middle East ( Zhang et al., 2005 [JoC, ARC] ), Western Europe and east Asia ( Kiktev et al., 2003 [PoC, JoC, MoS, SRC] ), Australasia and southeast Asia ( Griffiths et al., 2005 [JoC, MoS] ), China ( Zhai and Pan, 2003 [JoC] ) and central and southern Asia ( Klein Tank et al., 2006 [JoC] ). They all show patterns of changes in extremes consistent with a general warming, although the observed changes of the tails of the temperature distributions are often more complicated than a simple shift of the entire distribution would suggest (see Figure 3.38 ). In addition, uneven trends are observed for nighttime and daytime temperature extremes. In southern South America, significant increasing trends were found in the occurrence of warm nights and decreasing trends in the occurrence of cold nights, but no consistent changes in the indices based on daily maximum temperature. In Central America and northern South America, high extremes of both minimum and maximum temperature have increased. Warming of both the nighttime and daytime extremes was also found for the other regions where data have been analysed. For Australasia and Southeast Asia, the dominant distribution change at rural stations for both maximum and minimum temperature involved a change in the mean, affecting either one or both distribution tails, with no significant change in standard deviation ( Griffiths et al., 2005 [JoC, MoS] ). For urbanised stations, however, the dominant change also involved a change in the standard deviation. This result was particularly evident for minimum temperature.

Figure 3.38. Annual probability distribution functions for temperature indices for 202 global stations with at least 80% complete data between 1901 and 2003 for three time periods: 1901 to 1950 (black), 1951 to 1978 (blue) and 1979 to 2003 (red). The x-axis represents the percentage of time during the year when the indicators were below the 10th percentile for cold nights (left) or above the 90th percentile for warm nights (right). From Alexander et al. 2006 [JoC, SRC] ).

Few other studies have considered mutual changes in both the high and low tail of the same daily (minimum, maximum or mean) temperature distribution. Klein Tank and Können 2003 [JoC] ) analysed such changes over Europe using standard indices, and found that the annual number of warm extremes (above the 90th percentile for 1961 to 1990 ) of the daily minimum and maximum temperature distributions increased twice as fast during the last 25 years than expected from the corresponding decrease in the number of cold extremes (lowest 10%). Moberg and Jones 2005 [JoC, SRC] ) found that both the high and the low tail (defined by the 90th and 10th percentile) of the daily minimum and maximum temperature distribution over Europe in winter increased over the 20th century as a whole, with the low tail of minimum temperature warming significantly in summer. For an even longer period, Yan et al. 2002 [JoC] ) found decreasing warm extremes in Europe and China up to the late 19th century, decreasing cold extremes since then, and increasing warm extremes only since 1961, especially in summer (JJA). Brunet et al. 2006 [JoC] ) analysed 22 Spanish records for the period 1894 to 2003 and found greater reductions in the number of cold days than increases in hot days. However, since 1973 warm days have been rising dramatically, particularly near the Mediterranean coast. Beniston and Stephenson 2004 [JoC, SRC] ) showed that changes in extremes of daily temperature in Switzerland were due to changes in both the mean and the variance of the daily temperatures. ( Vincent and Mekis 2006 ) ) found progressively fewer extreme cold nights and cool days but more extreme warm nights and hot days for Canada from 1900 to 2003 and Robeson 2004 [JoC] ) found intense warming of the lowest daily minimum temperatures over western and central North America. In Argentina, the strong positive changes in minimum temperature seen during 1959 to 1998 were associated with significant increases in the frequency of warm nights; there were also decreases in cold days ( Rusticucci and Barrucand, 2004 [JoC, ARC] ).

Alexander et al. 2006 [JoC, SRC] and Caesar et al. 2006 [JoC, SRC] ) have brought all these and other regional results together, gridding the common indices or data for the period since 1946 . Over 74% of the global land area sampled showed a significant decrease in the annual occurrence of cold nights; a significant increase in the annual occurrence of warm nights took place over 73% of the area( Table 3.6 , Figure 3.38 and FAQ 3.3 ). This implies a positive shift in the distribution of daily minimum temperatureT_min throughout the globe. Changes in the occurrence of cold and warm days show warming as well, but generally less marked. This is consistent withT_min increasing more than maximum temperatureT_max,leading to a reduction in DTR since 1951 (see Sections 3.2.2.1 and 3.2.2.7 ). The change in the four extremes indices( Table 3.6 )also show that the distribution ofT_min andT_max have not only shifted, but also changed in shape. The indices for the number of cold and warm events have changed almost equally, which for a near-Gaussian distributed quantity indicates that the cold tails of the distributions have warmed considerably more than the warm tails over the last 50 years. Considering the last 25 years only, such a change in shape is not seen( Table 3.6 ).

3.8.2.2 Precipitation

The conceptual basis for changes in precipitation has been given by Allen and Ingram 2002 [JoC, MoS, ARC] and Trenberth et al. 2003 [PoC, JoC, SRC] ; see Section 3.3 and FAQ 3.2 ). Issues relate to changes in type, amount, frequency, intensity and duration of precipitation. Observed increases in atmospheric water vapour (see Section 3.4.2 )imply increases in intensity, but this will lead to reduced frequency or duration if the total evaporation rate from the Earth’s surface (land and ocean) is unchanged. The TAR states that it is likely that there has been a statistically significant 2 to 4% increase in the frequency of heavy and extreme precipitation events when averaged across the middle and high latitudes. Since then a more refined understanding of the observed changes in precipitation extremes has been achieved.

Many analyses indicate that the evolution of rainfall statistics through the second half of the 20th century is dominated by variations on the interannual to inter-decadal time scale and that trend estimates are spatially incoherent ( Manton et al., 2001 [JoC] ; Peterson et al., 2002 [JoC, ARC] ; Griffiths et al., 2003 [JoC] ; Herath and Ratnayake, 2004 [JoC, MoS] ). In Europe, there is a clear majority of stations with increasing trends in the number of moderately and very wet days (defined as wet days (≥1 mm of rain) that exceed the 75th and 95th percentiles, respectively) during the second half of the 20th century ( Klein Tank and Können, 2003 [JoC] ; Haylock and Goodess, 2004 [JoC] ). Similarly, for the contiguous USA, Kunkel et al. 2003 [JoC, SRC] and Groisman et al. 2004 [JoC, SRC] ) confirmed earlier results and found statistically significant increases in heavy (upper 5%) and very heavy (upper 1%) precipitation of 14 and 20%, respectively. Much of this increase occurred during the last three decades of the 20th century and is most apparent over the eastern parts of the country. In addition, there is new evidence from Europe and the USA that the relative increase in precipitation extremes is larger than the increase in mean precipitation, and this is manifested as an increasing contribution of heavy events to total precipitation ( Klein Tank and Können, 2003 [JoC] ; Groisman et al., 2004 [JoC, SRC] ).

Despite a decrease in mean annual rainfall, an increase in the fraction from heavy events was inferred for large parts of the Mediterranean ( Alpert et al., 2002 [JoC] ; Brunetti et al., 2004 [JoC] ; Maheras et al., 2004 [JoC] ). Further, Kostopoulou and Jones 2005 [SRC] ) noted contrasting trends of heavy rainfall events between an increase in the central Mediterranean (Italy) and a decrease over the Balkans. In South Africa, Siberia, central Mexico, Japan and the northeastern part of the USA, an increase in heavy precipitation was also observed, while total precipitation and/or the frequency of days with an appreciable amount of precipitation (wet days) was either unchanged or decreasing ( Easterling et al., 2000 [ARC] ;( Fauchereau et al., 2003; ) Sun and Groisman, 2004 [JoC, SRC] ; Groisman et al., 2005 [SRC] ).

Anumber of recent regional studies have been completed for southern South America ( Haylock et al., 2006 [JoC] ), Central America and northern South America ( Aguilar et al., 2005 [JoC] ), southern and western Africa ( New et al., 2006 [JoC, ARC] ), the Middle East ( Zhang et al., 2005 [JoC, ARC] ) and central and southern Asia ( Klein Tank et al., 2006 [JoC] ). For southern South America, the pattern of trends for extremes between 1960 and 2000 was generally the same as that for total annual rainfall ( Haylock et al., 2006 [JoC] ). A majority of stations showed a change to wetter conditions, related to the generally lower value of the SOI since 1976 / 1977, with the exception of southern Peru and southern Chile, where a decrease was observed in many precipitation indices. In the latter region, the change in ENSO has led to a weakening of the continental trough resulting in a southward shift in storm tracks and an important effect on the observed rainfall trends. No significant increases in total precipitation amounts were found over Central America and northern South America (see also Figure 3.14 ), but rainfall intensities have increased related to changes in SST of tropical Atlantic waters. Over southern and western Africa, and the Middle East, there are no spatially coherent patterns of statistically significant trends in precipitation indices. Averaged over central and southern Asia, a slight indication of disproportionate changes in the precipitation extremes compared with the total amounts is seen. In the Indian sub-continent Sen Roy and Balling 2004 [JoC] ) found that about two- thirds of all considered time series exhibit increasing trends in indices of precipitation extremes and that there are coherent regions with increases and decreases.

Alexander et al. 2006 [JoC, SRC] ) also gridded the extreme indices for precipitation (as for temperature in Section 3.8.2.1 ). Changes in precipitation extremes are much less coherent than for temperature, but globally averaged over the land area with sufficient data, the percentage contribution to total annual precipitation from very wet days (upper 5%) is greater in recent decades than earlier decades( Figure 3.39 ,top panel, and Table 3.6 ,last line). Observed changes in intense precipitation (with geographically varying thresholds between the 90th and 99.9th percentile of daily precipitation events) for more than one half of the global land area indicate an increasing probability of intense precipitation events beyond that expected from changes in the mean for many extratropical regions ( Groisman et al., 2005 [SRC] ; Figure 3.39 ,bottom panel). This finding supports the disproportionate changes in the precipitation extremes described in the majority of regional studies above, in particular for the mid-latitudes since about 1950 . It is still difficult to draw a consistent picture of changes in the tropics and the subtropics, where many areas are not analysed and data are not readily available.

Table 3.6. Global trends in extremes of temperature and precipitation as measured by the 10th and 90th percentiles (for 1961 – 1990 ). Trends with 5 and 95% confidence intervals and levels of significance (bold: <1%) were estimated by REML (see Appendix 3.A ), which allows for serial correlation in the residuals of the data about the linear trend. All trends are based on annual averages. Values are % per decade. Based on Alexander et al. 2006 [JoC, SRC] ).

As well as confirming previous findings, the new analyses provide seasonal detail and insight into longer-term variations for the mid-latitudes. While the increase in the USA is found primarily in the warm season ( Groisman et al., 2004 [JoC, SRC] ), central and northern Europe exhibited changes primarily in winter (DJF) and changes were insignificant in summer (JJA), but the studies did not include the extreme European summers of 2002 (very wet) and 2003 (very dry) ( Osborn and Hulme, 2002 [PoC, JoC, ARC] ; Haylock and Goodess, 2004 [JoC] ; Schmidli and Frei, 2005 [JoC, SRC] ). Although data are not as good, the frequencies of precipitation extremes in the USA were at comparable levels from 1895 into the early 1900 s and during the 1980 s to 1990 s ( Kunkel et al., 2003 [JoC, SRC] ). For Canada (excluding the high-latitude Arctic), Zhang et al. 2001a [JoC, ARC] ( and Vincent and Mekis 2006 ) ) found that the frequency of precipitation days significantly increased during the 20th century, but averaged for the country as a whole, there is no identifiable trend in precipitation extremes. Nevertheless, Groisman et al. 2005 [SRC] ) found significant increases in the frequency of heavy and very heavy (between the 95th and 99.7th percentile of daily precipitation events) precipitation in British Columbia south of 55°N for 1910 to 2001, and in other areas( Figure 3.39 ,bottom panel).

Figure 3.39. (Top) Observed trends (% per decade) for 1951 to 2003 in the contribution to total annual precipitation from very wet days (95th percentile). Trends were only calculated for grid boxes where both the total and the 95th percentile had at least 40 years of data during this period and had data until at least 1999 . (Middle) Anomalies (%) of the global annual time series (with respect to 1961 to 1990 ) defined as the percentage change of contributions of very wet days from the base period average (22.5%). The smooth red curve shows decadal variations (see Appendix 3.A ). From Alexander et al. 2006 [JoC, SRC] ). (Bottom) Regions where disproportionate changes in heavy and very heavy precipitation during the past decades were documented as either an increase (+) or decrease (–) compared to the change in the annual and/or seasonal precipitation (updated from Groisman et al., 2005 [SRC] ). Thresholds used to define “heavy” and “very heavy” precipitation vary by season and region. However, changes in heavy precipitation frequencies are always greater than changes in precipitation totals and, in some regions, an increase in heavy and/or very heavy precipitation occurred while no change or even a decrease in precipitation totals was observed.

Since the TAR, several regional analyses have been undertaken for statistics with return periods much longer than in the previous studies. For the UK, Fowler and Kilsby 2003a [JoC] ,b), using extreme value statistics, estimated that the recurrence of 10-day precipitation totals with a 50-year return period (based on data for 1961 to 1990 ) had increased by a factor of two to five by the 1990 s in northern England and Scotland. Their results for long return periods are qualitatively similar to changes obtained for traditional (moderate) statistics ( Osborn et al., 2000 [PoC, JoC, ARC] ; Osborn and Hulme, 2002 [PoC, JoC, ARC] ), but there are differences in the relative magnitude of the change between seasons ( Fowler and Kilsby, 2003b [JoC] ). For the contiguous USA, Kunkel et al. 2003 [JoC, SRC] and Groisman et al. 2004 [JoC, SRC] ) analysed return periods of 1 to 20 years, and interannual to inter-decadal variations during the 20th century exhibit a high correlation between all return periods. Similar results were obtained for several extratropical regions ( Groisman et al., 2005 [SRC] ), including the central USA, the northwestern coast of North America, southern Brazil, Fennoscandia, the East European Plain, South Africa, southeastern Australia and Siberia. In summary, from the available analyses there is evidence that the changes at the extreme tail of the distribution (several-decade return periods) are consistent with changes inferred for more robust statistics based on percentiles between the 75th and 95th levels, but practically no regions have sufficient data to assess such trends reliably.

3.8.3.1 Western North Pacific

In the western North Pacific, long-term trends are masked by strong inter-decadal variability for 1960 to 2004 ( Chan and Liu, 2004 [JoC] ; Chan, 2006 [JoC] ), but results also depend on the statistics used and there are uncertainties in the data prior to the mid- 1980 s (Klotzbach, 2006 ). Further increases in activity have occurred in the last few years after Chan and Liu 2004 [JoC] ) was completed( Figure 3.40 ). Tropical cyclones making landfall in China are a small fraction of the total storms, and no obvious long-term trend can be discerned ( (He et al., 2003; ) Liu and Chan, 2003 [JoC, MoS] ; Chan and Liu, 2004 [JoC] ). However, Emanuel 2005a [JoC] and Webster et al. 2005 [JoC] , 2006 ) indicated that the typhoons have become more intense in this region, with almost a doubling of PDI values since the 1950 s and an increase of about 30% in the number of category 4 and 5 storms from 1990 to 2004 compared with 1975 to 1989 . The post- 1985 record analysed by Klotzbach ( 2006 ) is too short to provide reliable trends.

The main modulating influence on tropical cyclone activity in the western North Pacific appears to be the changes in atmospheric circulation associated with ENSO, rather than local SSTs ( Liu and Chan, 2003 [JoC, MoS] ; Chan and Liu, 2004 [JoC] ). In El Niño years, tropical cyclones tend to be more intense and longer-lived than in la Niña years ( Camargo and Sobel, 2004 [NPR, MoS] ) and occur in different locations. In the summer (JJA) and autumn (SON) of strong El Niño years, tropical cyclone numbers increase markedly in the southeastern quadrant of the western North Pacific (0°N–17°N, 140°E–180°E) and decrease in the northwestern quadrant (17°N–30°N, 120°E–140°E; Wang and Chan, 2002 [JoC, SRC] ). In SON of El Niño years from 1961 to 2000, significantly fewer tropical cyclones made landfall in the western North Pacific compared with neutral years, although in Japan and the Korean Peninsula no statistically significant change was detected. In contrast, in SON of la Niña years, significantly more landfalls were reported in China ( Wu et al., 2004 [JoC] ). Overall in 2004, the number of tropical depressions, tropical storms and typhoons was slightly above the 1971 to 2000 median but the number of typhoons (21) was well above the median (17.5) and second highest to 1997, when 23 developed. Moreover, a record number of 10 tropical cyclones or typhoons made landfall in Japan; the previous record was 6 ( Levinson, 2005 [JoC, SRC] ). The ACE index was very close to normal for the 2005 season( Figure 3.40 ).

3.8.3.2 North Atlantic

The North Atlantic hurricane record begins in 1851 and is the longest among cyclone series. Values are considered fairly reliable after about 1950 when measurements from reconnaissance aircraft began. Methods of estimating wind speed from aircraft have evolved over time and, unfortunately, changes were not always well documented. The record is most reliable after the early 1970 s ( Landsea, 2005 [JoC] ). The North Atlantic record shows a fairly active period from the 1930 s to the 1960 s followed by a less active period in the 1970 s and 1980 s, similar to the fluctuations of the AMO( Figure 3.33 ).

Beginning with 1995, all but two Atlantic hurricane seasons have been above normal (relative to the 1981 to 2000 base period). The exceptions are the two El Niño years of 1997 and 2002 . As noted in Section 3.8.3 ,El Niño acts to reduce activity and la Niña acts to increase activity in the North Atlantic. The increased activity after 1995 contrasts sharply with the generally below-normal seasons observed during the previous 25-year period ( 1970 – 1994 ). These multi-decadal fluctuations in hurricane activity result nearly entirely from differences in the number of hurricanes and major hurricanes forming from tropical storms first named in the tropical Atlantic and Caribbean Sea. The change from the negative phase of the AMO in the 1970 s and 1980 s (see Section 3.6.6 )to the post- 1995 period has been a contributing factor to the increased hurricane activity ( Goldenberg et al., 2001 [JoC] ) and is well depicted in Atlantic SSTs( Figure 3.33 ), including those in the tropics. Nevertheless, it appears likely that most of the warming since the 1970 s can be associated with global SST increases rather than the AMO ( Trenberth and Shea, 2006 [PoC, JoC, SRC] ; see Section 3.6.6 ).

During 1995 to 2004, hurricane seasons averaged 13.6 tropical storms, 7.8 hurricanes and 3.8 major hurricanes, and have an average ACE index of 159% of the median. The record-breaking 2005 season is documented in more detail in Section 3.8.4 , Box 3.6 .In contrast, during the preceding 1970 to 1994 period, hurricane seasons averaged 8.6 tropical storms, 5 hurricanes and 1.5 major hurricanes, and had an average ACE index of only 70% of the median. NOAA classifies 12 (almost one-half) of these 25 seasons as being below normal, and only three as being above normal ( 1980, 1988, 1989 ), with the remainder as normal. The positive phase of the AMO was also present during the above-normal hurricane decades of the 1950 s and 1960 s, as indicated by comparing Atlantic SSTs( Figure 3.33 )and seasonal ACE values( Figure 3.40 ). In 2004, there were 15 named storms, of which 9 were hurricanes, and an unprecedented 4 hit Florida, causing extensive damage ( Levinson, 2005 [JoC, SRC] ). In 2005, record-high SSTs( Figure 3.33 )and favourable atmospheric conditions enabled the most active season on record (by many measures), but this was not fully reflected in the ACE index (see also Section 3.8.4 , Box 3.6 ). In 2005, the North Atlantic ACE was the third highest since 1948, while the PDI was the highest on record, exceeding the previous high reached in 2004 .

Key factors in the recent increase in Atlantic activity ( Chelliah and Bell, 2004 [JoC] ) include: (1) warmer SSTs across the tropical Atlantic; (2) an amplified subtropical ridge at upper levels across the central and eastern North Atlantic; (3) reduced vertical wind shear in the deep tropics over the central North Atlantic, which results from an expanded area of easterly winds in the upper atmosphere and weaker easterly trade winds in the lower atmosphere; and (4) a configuration of the African easterly jet that favours hurricane development from tropical disturbances moving westward from the African coast. The vertical shear in the main development region where most Atlantic hurricanes form ( Aiyyer and Thorncroft, 2006 [JoC, SRC] ) fluctuates interannually with ENSO, and with a multi-decadal variation that is correlated with Sahel precipitation. The latter switched sign around 1970 and remained in that phase until the early 1990 s, consistent with the AMO variability. It has been argued that the QBO is also a factor in interannual variability ( Gray, 1984 [JoC] ). The most recent decade has the highest SSTs on record in the tropical North Atlantic( Figure 3.33 ), apparently as part of global warming and a favourable phase of the AMO. In the Atlantic generally, the changing environmental conditions( Box 3.5 )have been more favourable in the past decade for tropical storms to develop.

3.8.3.3 Eastern North Pacific

Tropical cyclone activity (both frequency and intensity) in this region is related especially to SSTs, the phase of ENSO and the phase of the QBO in the tropical lower stratosphere. Above-normal tropical cyclone activity during El Niño years and the lowest activity typically associated with la Niña years is the opposite of the North Atlantic Basin ( Landsea et al., 1998 [JoC, MoS] ). Tropical cyclones tend to attain a higher intensity when the QBO is in its westerly phase at 30 hPa in the tropical lower stratosphere. A well-defined peak in the seasonal ACE occurred in early 1990 s, with the largest annual value in 1992 ( Figure 3.40 ), but values are unreliable prior to 1970 in the pre-satellite era. In general, seasonal hurricane activity, including the ACE index, has been below average since 1995, with the exception of the El Niño year of 1997, and is inversely related to the observed increase in activity in the North Atlantic basin over the same time period. This pattern is associated with the AMO ( Levinson, 2005 [JoC, SRC] ) and ENSO. Nevertheless, there has been an increase in category 4 and 5 storms ( Webster et al., 2005 [JoC] ).

The tropical cyclone season in the South Indian Ocean is normally active from December through April and thus the data are summarised by season in Figure 3.40 ,rather than by calendar year. The basin extends from the African coastline, where tropical cyclones affect Madagascar, Mozambique and the Mascarene Islands, including Mauritius, to 110°E (tropical cyclones east of 110°E are included in the Australian summary), and from the equator southward, although most cyclones develop south of 10°S. The intensity of tropical cyclones in the South Indian Ocean is reduced during El Niño events( Figure 3.40 ;Levinson, 2005 [JoC, SRC] ). Lack of historical record keeping severely hinders trend analysis.

3.8.3.4 Indian Ocean

The North Indian Ocean tropical cyclone season extends from May to December, with peaks in activity during May to June and November when the monsoon trough lies over tropical waters in the basin. Tropical cyclones are usually short-lived and weak, quickly moving into the sub-continent. Tropical storm activity in the northern Indian Ocean has been near normal in recent years( Figure 3.40 ).

3.8.3.5 Australia and the South Pacific

The tropical cyclone season in the South Pacific-Australia region typically extends over the period November through April, with peak activity from December through March. Tropical cyclone activity in the Australian region (105°E–160°E) apparently declined somewhat over the past decade( Figure 3.40 ), although this may be partly due to improved analysis and discrimination of weak cyclones that previously were estimated at minimum tropical storm strength ( Plummer et al., 1999 [JoC] ). Increased cyclone activity in the Australian region has been associated with la Niña years, while below-normal activity has occurred during El Niño years ( Plummer et al., 1999 [JoC] ;( Kuleshov and de Hoedt, 2003 ) ). In contrast, in the South Pacific east of 160°E, the opposite signal has been observed, and the most active years have been associated with El Niño events, especially during the strong 1982 – 1983 and 1997 – 1998 events ( Levinson, 2005 [JoC, SRC] ), and maximum ACE values occurred from January through March 1998 ( Figure 3.40 Webster et al. 2005 [JoC] ) found more than a doubling in the numbers of category 4 and 5 hurricanes in the southwest Pacific region between 1975 to 1989 and 1990 to 2004 . In the 2005 – 2006 season, la Niña influences shifted tropical storm activity away from the South Pacific to the Australian region and in March and April 2006, four category 5 typhoons (Floyd, Glenda, Larry and Monica) occurred.

3.8.3.6 South Atlantic

In late March 2004 in the South Atlantic, off the coast of Brazil, the first and only documented hurricane in that region occurred ( Pezza and Simmonds, 2005 [JoC, SRC] ). It came ashore in the Brazilian state of Santa Catarina on 28 March 2004 with winds, estimated by the U.S. National Hurricane Center, of near 40 ms^–1,causing much damage to property and some loss of life (see Levinson, 2005 [JoC, SRC] ). The Brazilian meteorologists dubbed it ‘Catarina’. This event appears to be unprecedented although records are poor before the satellite era. Pezza and Simmonds 2005 [JoC, SRC] ) suggest that a key factor in the hurricane development was the more favourable atmospheric circulation regime associated with the positive trend in the SAM (see Section 3.6 )

Box 3.6: Recent Extreme Events

Single extreme events cannot be simply and directly attributed to anthropogenic climate change, as there is always a finite chance the event in question might have occurred naturally. However, when a pattern of extreme weather persists for some time, it may be classed as an extreme climate event, perhaps associated with anomalies in SSTs (such as El Niño). This box provides examples of some recent (post-TAR) notable extreme climate events. A lack of long and homogeneous observational data often makes it difficult to place some of these events in a longer-term context. The odds may have shifted to make some of them more likely than in an unchanging climate, but attribution of the change in odds typically requires extensive model experiments, a topic taken up in Chapter9 .It may be possible, however, to say that the occurrence of recent events is consistent with physically based expectations arising from climate change. Some examples of these recent events are described below (in response to the questions posed to IPCC by the governments) and placed in a long-term perspective.

Drought in Central and Southwest Asia, 1998–2003

Between 1999 and 2003 a severe drought hit much of southwest Asia, including Afghanistan, Kyrgyzstan, Iran, Iraq, Pakistan, Tajikistan, Turkmenistan, Uzbekistan and parts of Kazakhstan ( Waple and Lawrimore, 2003 [JoC, SRC] ; Levinson and Waple, 2004 [JoC, SRC] ). Most of the area is a semiarid steppe, receiving precipitation only during winter and early spring through orographic capture of eastward- propagating mid-latitude cyclones from the Atlantic Ocean and the Mediterranean Sea ( Martyn, 1992 [NPR] ). Precipitation between 1998 and 2001 was on average less than 55% of the long-term average, making the drought conditions in 2000 the worst in 50 years ( Waple et al., 2002 [JoC] ). By June 2000, some parts of Iran had reported no measurable rainfall for 30 consecutive months. In December 2001 and January 2002, snowfall at higher altitudes brought relief for some areas, although the combination of above-average temperatures and early snowmelt, substantial rainfall and hardened ground desiccated by prolonged drought resulted in flash flooding during spring in parts of central and southern Iran, northern Afghanistan and Tajikistan. Other regions in the area continued to experience drought through 2004 ( Levinson, 2005 [JoC, SRC] ). In these years, an anomalous ridge in the upper-level circulation was a persistent feature during the cold season in central and southern Asia. The pattern served to both inhibit the development of baroclinic storm systems and deflect eastward-propagating storms to the north of the drought-affected area. Hoerling and Kumar 2003 [JoC] ) linked the drought in certain areas of the mid-latitudes to common global oceanic influences. Both the prolonged duration of the 1998 – 2002 cold phase ENSO (La Niña) event and the unusually warm ocean waters in the western Pacific and eastern Indian Oceans appear to contribute to the severity of the drought ( Nazemosadat and Cordery, 2000 [JoC] ; Barlow et al., 2002 [JoC] ; Nazemosadat and Ghasemi, 2004 [JoC] ).

Drought in Australia, 2002–2003

Asevere drought affected Australia during 2002, associated with a moderate El Niño event ( (Watkins, 2002 ) ). However, droughts in 1994 and 1982 were about as dry as the 2002 drought. Earlier droughts in the first half of the 20th century may well have been even drier. The 2002 drought came after several years of good rainfall (averaged across the country), rather than during an extended period of low rainfall such as occurred in the 1940 s. If only rainfall is considered, the 2002 drought alone does not provide evidence of Australian droughts becoming more extreme. However, daytime temperatures during the 2002 drought were much higher than during previous droughts. The mean annual maximum temperature for 2002 was 0.5°C warmer than in 1994 and 1.0°C warmer than in 1982 . So in this sense, the 2002 drought and associated heat waves were more extreme than the earlier droughts, because the impact of the low rainfall was exacerbated by high potential evaporation ( Karoly et al., 2003 [Ambiguous] ; Nicholls, 2004 [JoC, SRC] ). The very high maximum temperatures during 2002 could not simply be attributed to the low rainfall, although there is a strong negative correlation between rainfall and maximum temperature. Severe long-term drought, stemming from at least three years of rainfall deficits, continued during 2005, especially in the eastern third of Australia, although above-normal rainfall in winter and spring 2005 brought some relief. These conditions also have been accompanied by record high maximum temperatures over Australia during 2005 (a comparable national series is only available since 1951 ).

Drought in Western North America, 1999–2004

The western USA, southern Canada and northwest Mexico experienced a recent pervasive drought ( Lawrimore et al., 2002 [JoC, SRC] ), with dry conditions commencing as early as 1999 and persisting through the end of 2004 (Box 3.6, Figure 1). Drought conditions were recorded by several hydrologic measures, including precipitation, streamflow, lake and reservoir levels and soil moisture ( Piechota et al., 2004 [JoC] ). The period 2000 through 2004 was the first instance of five consecutive years of below-average flow in the Colorado River since the beginning of modern records in 1922 ( Pagano et al., 2004 [JoC] Cook et al. 2004 [PoC, JoC, ARC] ) provided a longer-term context for this drought. In the western conterminous USA, the area under moderate to extreme drought, as given by the PDSI, rose above 20% in November 1999 and stayed above this level persistently until October 2004 . At its peak (August 2002 ), this drought affected 87% of the West (Rocky Mountains westward), making it the second most extensive and one of the longest droughts in the last 105 years. The impacts of this drought have been exacerbated by depleted or earlier than average melting of the mountain snowpack, due to warm springs, as observed changes in timing from 1948 to 2000 trended earlier by one to two weeks in many parts of the West ( Cayan et al., 2001 [JoC, SRC] ; Regonda et al., 2005 [JoC] ; Stewart et al., 2005 [JoC, SRC] ). Within this episode, the spring of 2004 was unusually warm and dry, resulting in record early snowmelt in several western watersheds ( Pagano et al., 2004 [JoC] ).

Hoerling and Kumar 2003 [JoC] ) attributed the drought to changes in atmospheric circulation associated with warming of the western tropical Pacific and Indian oceans, while McCabe et al. 2004 [JoC] ) have produced evidence suggesting that the confluence of both Pacific decadal and Atlantic multi-decadal fluctuations is involved. In the northern winter of 2004 to 2005, the weak El Niño was part of a radical change in atmospheric circulation and storm track across the USA, ameliorating the drought in the Southwest, although lakes remain low.

Box 3.6, Figure 1. Percentage of the USA west of the Rocky Mountains (the 11 states west of and including Montana to New Mexico) that was dry (top) or wet (bottom), based on the Palmer Drought Severity Index for classes of moderate to extreme drought or wet. From NOAA, NCDC.

Floods in Europe, Summer 2002

Acatastrophic flood occurred along several central European rivers in August 2002 . The floods resulting from extraordinarily high precipitation were enhanced by the fact that the soils were completely saturated and the river water levels were already high because of previous rain (Rudolf

and Rapp, 2003 [NPR] ;( Ulbrich et al., 2003a, ) b). Hence, it was part of a pattern of weather over an extended period. In the flood, the water levels of the Elbe at Dresden reached a maximum mark of 9.4 m, which is the highest level since records began in 1275 ( (Ulbrich et al., 2003a ) ). Some small villages in the Ore Mountains (on tributaries of the Elbe) were hit by extraordinary flash floods. The river Vltava inundated the city of Prague before contributing to the Elbe flood. A return period of 500 years was estimated for the flood levels at Prague ( (Grollmann and Simon, 2002 ) ). The central European floods were caused by two heavy precipitation episodes. The first, on 6–7 August, was situated mainly over Lower Austria, the southwestern part of the Czech Republic and southeastern Germany. The second took place on 11–13 August 2002 and most severely affected the Ore Mountains and western parts of the Czech Republic. A persistent low-pressure system moved slowly from the Mediterranean Sea to central Europe on a path over or near the eastern Alps and led to large-scale, strong and quasi-stationary frontal lifting of air with very high liquid water content. Additional to this advective rain were convective precipitation processes (showers and thunderstorms) and a significant orographic lifting (mainly over the Ore Mountains). A maximum 24-hour precipitation total of 353 mm was observed at the German station Zinnwald-Georgenfeld, a new record for Germany. The synoptic situation leading to floods is well known to meteorologists of the region. Similar situations led to the summer floods of the River Oder in 1997 and the River Vistula in 2001 ( (Ulbrich et al., 2003b ) ). Average summer precipitation trends in the region are negative but barely significant ( Schönwiese and Rapp, 1997 [NPR] ) and there is no significant trend in flood occurrences of the Elbe within the last 500 years ( Mudelsee et al., 2003 [JoC] ). However, the observed increase in precipitation variability at a majority of German precipitation stations during the last century ( (Trömel and Schönwiese, 2005 ) ) is indicative of an enhancement of the probability of both floods and droughts.

Heat Wave in Europe, Summer 2003

The heat wave that affected many parts of Europe during the course of summer 2003 produced record-breaking temperatures particularly during June and August ( Beniston, 2004 [JoC, SRC] ; Schär et al., 2004 [JoC, ARC] ; see Box 3.6, Figure 2). Absolute maximum temperatures exceeded the record highest temperatures observed in the 1940 s and early 1950 s in many locations in France, Germany, Switzerland, Spain, Italy and the UK, according to the information supplied by national weather agencies ( WMO, 2004 [NPR] ). Gridded instrumental temperatures (from CRUTEM2v for the region 35°N–50°N, 0–20°E) show that the summer was the hottest since comparable records began in 1780 : 3.8°C above the 1961 to 1990 average and 1.4°C hotter than any other summer in this period (the second hottest was 1807 ). Based on early documentary records, Luterbacher et al. 2004 [JoC, ARC] ) estimated that 2003 is very likely to have been the hottest summer since at least 1500 . The 2003 heat wave was associated with a very robust and persistent blocking high-pressure system that may be a manifestation of an exceptional northward extension of the Hadley Cell ( (Black et al., 2004; ) ( Fink et al., 2004 ) ). Already a record month in terms of maximum temperatures, June exhibited high geopotential values that penetrated northwards towards the British Isles, with thegreatest northward extension and longest persistence of record-high temperatures observed in August. An exacerbating factor for the temperature extremes was the lack of precipitation in many parts of western and central Europe, leading to much-reduced soil moisture and surface evaporation and evapotranspiration, and thus to a strong positive feedback effect ( Beniston and Diaz, 2004 [JoC, MoS, SRC] ).

Box 3.6, Figure 2. Long time series of JJA temperature anomalies in Central Europe relative to the 1961 to 1990 mean (top). The smooth curve shows decadal variations (see Appendix 3.A ). In the summer of 2003, the value of 3.8°C far exceeded the next largest anomaly of 2.4°C in 1807, and the highly smoothed Gaussian distribution (bottom) of maximum temperatures (red) compared with normal (blue) at Basel, Switzerland ( Beniston and Diaz, 2004 [JoC, MoS, SRC] ) shows how the whole distribution shifted.

The 2005 Tropical Storm Season in the North Atlantic

The 2005 North Atlantic hurricane season (1 June to 30 November) was the most active on record by several measures, surpassing the very active season of 2004 (e.g., Levinson, 2005 [JoC, SRC] ) and causing an unprecedented level of damage. Even before the peak in the seasonal activity, the seven tropical storms in June and July were the most ever, and hurricane Dennis was the strongest on record in July and the earliest ever fourth-named storm. The record 2005 North Atlantic hurricane season featured the largest number of named storms (28; sustained winds greater than 17 ms^–1)and is the only time names have ventured into the Greek alphabet. It had the largest number of hurricanes (15; sustained winds greater than 33 ms^–1)recorded, and is the only time there have been four category 5 storms (maximum sustained winds greater than 67 ms^–1). These included the most intense Atlantic storm on record (Wilma, with recorded surface pressure in the eye of 882 hPa), the most intense storm in the Gulf of Mexico (Rita, 897 hPa), and Katrina. Tropical storm Vince was the first ever to make landfall in Portugal and Spain. In spite of these metrics, the ACE index, although very high and surpassing the 2004 value( Figure 3.40 ), was not the highest on record, as several storms were quite short lived. Six of the eight most damaging storms on record for the USA occurred from August 2004 to September 2005 (Charlie, Ivan, Francis, Katrina, Rita, Wilma) while another storm in 2005 (Stan) caused severe flooding and mudslides as well as about 2,000 fatalities in central America (Guatemala, El Salvador and southern Mexico).

SSTs in the tropical North Atlantic region critical for hurricanes (10°N to 20°N) were at record-high levels (0.9°C above the 1901 to 1970 normal ) in the extended summer (June to October) of 2005 ( Figure 3.33 ), and these high values were a major reason for the very active hurricane season along with favourable atmospheric conditions (see Box 3.5 ). A substantial component of this warming was the global mean SST increase ( Trenberth and Shea, 2006 [PoC, JoC, SRC] ; see Sections 3.2 and 3.6.6 ).

3.8.4.1 Extratropical Cyclones

Intense extratropical cyclones are often associated with extreme weather, particularly with severe windstorms. Significant increases in the number or strength of intense extratropical cyclone systems have been documented in a number of studies (e.g., Lambert, 1996 [JoC] ;( Gustafsson, 1997; ) McCabe et al., 2001 [JoC] ; Wang et al., 2006a [JoC, ARC] ) with associated changes in the preferred tracks of storms as described in Section 3.5.3 .As with tropical cyclones, detection of long-term changes in cyclone measures is hampered by incomplete and changing observing systems. Some earlier results have been questioned because of changes in the observation system (e.g., Graham and Diaz, 2001 [JoC] ).

Results from NRA and ERA-40 show that an increase in the number of deep cyclones is apparent over the North Pacific and North Atlantic ( Graham and Diaz, 2001 [JoC] ; Gulev et al., 2001 [JoC, SRC] ), with statistically significant winter increases over both ocean basins ( Simmonds and Keay, 2002 [SRC] ; Wang et al., 2006a [JoC, ARC] Geng and Sugi 2001 [JoC] ) found that cyclone density, deepening rate, central pressure gradient and translation speed have all been increasing in the winter North Atlantic. Caires and Sterl 2005 [JoC, MoS] ) compared global estimates of 100-year return values of wind speed and SWH in ERA-40, with linear bias corrections based on buoy data, for three different 10-year periods. They showed that the differences in the storm tracks can be attributed to decadal variability in the NH, linked to changes in global circulation patterns, most notably to the NAO (see also Section 3.5.6 ).

Using NCEP-2 reanalysis data, Lim and Simmonds 2002 [JoC, SRC] ) showed that for 1979 to 1999, increasing trends in the annual number of explosively developing (deepening by 1 hPa per hour or more) extratropical cyclones are significant in the SH and over the globe (0.56 and 0.78 more systems per year, respectively), while the positive trend did not achieve significance in the NH. Simmonds and Keay 2002 [SRC] ) obtained similar results for the change in the number of cyclones in the decile for deepest cyclones averaged over the North Pacific and over the North Atlantic in winter over the period 1958 to 1997 .

As noted in Sections 3.5.3 and 3.5.7 ,the time-dependent biases in the reanalysis cause uncertainties in the trends reported above. Besides reanalyses, station data may also be used to indicate evidence for changes in extratropical cyclone activity. Instead of direct station wind measurements, which may suffer from a lack of consistency of instrumentation, methodology and exposure, values based on pressure gradients have been derived that are more reliable for discerning long-term changes. Alexandersson et al. 2000 [SRC] ) used station pressure observations for 21 stations over northwestern Europe back to 1881, from which geostrophic winds were calculated using ‘pressure-triangle’ methods. They found a decline of storminess expressed by the 95th and 99th percentiles from high levels during the late 19th century to a minimum around 1960 and then a quite rapid increase to a maximum around 1990, followed again by a decline( Figure 3.41 ). Positive NAO winters are typically associated with more intense and frequent storms (see Section 3.6.4 ). Similar results were obtained by Schmith et al. 1998 [JoC] ) using simpler indices based on pressure tendency. Bärring and von Storch 2004 [JoC] ), using both pressure tendencies and the number of very low pressure values, confirmed these results on the basis of two especially long station series in southern Sweden dating back to about 1800 . Studies of rapid pressure changes at stations indicate an increase in the frequency, duration and intensity of winter cyclone activity over the lower Canadian Arctic and in the number and intensity of severe storms over the southern UK since the 1950 s, but a decrease over southern Canada and Iceland ( Wang et al., 2006b [JoC] ; Alexander et al., 2005 [JoC, SRC] ). Besides a northward shift of the storm track (see 3.5.3), the station pressure data for parts of the North Atlantic region show a modest increase in severe storms in recent decades. However, decadal-scale fluctuations of similar magnitude have occurred earlier in the 19th and 20th centuries.

Figure 3.41. Storm index for the British Isles, North Sea and Norwegian Sea, 1881 to 2004 . Blue circles are 95th percentiles and red crosses 99th percentiles of standardised geostrophic winds averaged over 10 sets of triangles of stations. The smoothed curves are a decadal filter (updated from Alexandersson et al., 2000 [SRC] ).

Direct surface wind measurements have, however, been used in a few studies. An analysis of extreme pressure differences and surface winds ( Salinger et al., 2005 [JoC] ) showed a significant increasing trend over the last 40 years in westerly wind extremes over the southern part of New Zealand and the oceans to the south. The trends are consistent with the increased frequency of El Niño events in recent decades, associated with Pacific decadal variability (see Section 3.6.3 ). While the zonal pressure gradient and extreme westerly wind frequency have both increased over southern New Zealand, the frequency of extreme easterly winds has also increased there, suggesting more variability in the circulation generally. However, trends in pressure differences (based on the ERA-40, NRA and station data) are not always consistent with changes in surface windiness (e.g., ( Smits et al., 2005 ) ). Based on observed winds at 10 m height over the Netherlands, ( Smits et al. 2005 ) ) found a decline in strong (greater than about 8 on the Beaufort scale) wind events over the last 40 years. Differences cannot entirely be explained by changes in surface aerodynamic roughness, and they concluded that inhomogeneities in the reanalyses are the cause. However, local differences can be important and intensity and severity of storms may not always be synonymous with local extreme surface winds and gusts.

3.8.4.2 Tornadoes, Hail, Thunderstorms, Dust Storms and Other Severe Local Weather

Evidence for changes in the number or intensity of tornadoes relies entirely on local reports. In the USA, databases for tornado reporting are well established, although changes in procedures for evaluating the intensity of tornadoes introduced significant discontinuities in the record. In particular, the apparent decrease in strong tornadoes in the USA from the early period of the official record ( 1950 s– 1970 s) to the more recent period is, in large part, a result of the way damage from the earlier events was evaluated. ( Trapp et al. 2005 ) ) also questioned the completeness of the tornado record and argued that about 12% of squall-line tornadoes remain unreported. In many European countries, the number of tornado reports has increased considerably over the last decade ( (Snow, 2003 ) ), leading to a much higher estimate of tornado activity ( Dotzek, 2003 [MoS, SRC] Bissolli et al. 2007 [NPR, JoC, SRC, 2007] ) showed that the increase in Germany between 1950 and 2003 mainly concerns weak tornadoes (F0 and F1 on the Fujita scale), thus paralleling the evolution of tornado reports in the USA after 1950 (see, e.g., Dotzek et al., 2005 [JoC, SRC] ) and making it likely that the increase in reports in Europe is at least dominated (if not solely caused) by enhanced detection and reporting efficiency. Doswell et al. 2005 [MoS] ) highlighted the difficulties encountered when trying to find observational evidence for changes in extreme events at local scales connected to severe thunderstorms. In light of the very strong spatial variability of small-scale severe weather phenomena, the density of surface meteorological observing stations is too coarse to measure all such events. Moreover, homogeneity of existing station series is questionable. While remote sensing techniques allow detection of thunderstorms even in remote areas, they do not always uniquely identify severe weather events from these storms. Another approach links severe thunderstorm occurrence to larger-scale environmental conditions in places where the observations of events are fairly good and then consider the changes in the distribution of those environments ( (Brooks et al., 2003; ) Bissolli et al., 2007 [NPR, JoC, SRC, 2007] ).

Although a decreasing trend in dust storms was observed from the mid- 1950 s to the mid- 1990 s in northern China, the number of dust storm days increased from 1997 to 2002 ( (Li and Zhai, 2003; ) ( Zhou and Zhang, 2003 ) ). The decreasing trend appears linked to the reduced cyclone frequency and increasing winter (DJF) temperatures ( Qian et al., 2002 [JoC] ). The recent increase is associated with vegetation degradation and drought, plus increased surface wind speed ( (Wang and Zhai, 2004; ) Zou and Zhai, 2004 [JoC] ).