SEARCH

SEARCH BY CITATION

Keywords:

  • reanalysis;
  • surface air temperature;
  • trend

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Analysis Method
  5. 3. Global Trends
  6. 4. Spatial Patterns
  7. 5. Uncertainties in CRUTEM and in 20CR
  8. 6. Conclusion
  9. Appendix A
  10. Acknowledgments
  11. References
  12. Supporting Information

[1] Land surface air temperature trends observed during 1979–2008 are compared with those simulated by the 20th Century Reanalysis that is driven only by observed sea surface temperatures and sea ice, atmospheric CO2 concentrations, solar and volcanic forcings, and surface pressure data. On a global annual average, the 20th Century Reanalysis simulates a little more than 80% of the observed trend, but with substantial regional and seasonal variations. The remainder of the trend may be ascribed tentatively to land use changes, aerosol increases and decreases, and changes in minor greenhouse gases not accounted for in the 20th Century Reanalysis.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Analysis Method
  5. 3. Global Trends
  6. 4. Spatial Patterns
  7. 5. Uncertainties in CRUTEM and in 20CR
  8. 6. Conclusion
  9. Appendix A
  10. Acknowledgments
  11. References
  12. Supporting Information

[2] The Fourth Assessment Report of the Intergovernmental Panel on Climate Change [Intergovernmental Panel on Climate Change, 2007] concluded that most of the observed global warming since the mid-twentieth century is very likely due to the observed increase in anthropogenic greenhouse gas concentrations. This statement was based on formal attribution studies, but was also supported by observational studies showing that there were no major urban warming influences on the observed global temperature series because urban sites had either been avoided [Brohan et al., 2006] or appropriate adjustments had been applied [Hansen et al., 2001]. A subsequent review [Parker, 2010] maintained these conclusions, and Hansen et al. [2010] found virtually no change in estimated global land surface air warming in 1900–2009 when temperatures from night-lit areas, detected by satellite, were adjusted so that their trends conformed with those from unlit, rural areas nearby. Efthymiadis and Jones [2010] estimated an upper limit on urban influence of only 0.02°C per decade, or ∼15% of the total land air temperature trends, in 1951–2009. In some regions, however, urban influences are greater, and Yang et al. [2011] summarize studies for China that are generally consistent with their own estimate that about a quarter of the observed warming in east China since the early 1980s has been urban.

[3] Rural land use change also influences surface air temperature. Fall et al. [2010] used the North American Regional Reanalysis (NARR), which does not assimilate observed surface air temperatures, to assess the effects of changes to the land surface on patterns of surface air temperature trend over the USA. The NARR generated overall surface air temperature trends for 1979–2003 similar to the U.S. Historical Climatology Network over the USA; the geographical pattern of observations-minus-reanalysis trends was in qualitative agreement with Hansen et al.'s [2001] nightlights-based adjustments. Land use transitions to urban or barren caused observations to warm relative to NARR, whereas transitions to agriculture caused relative cooling, as expected from changes in surface radiative, sensible, and latent heat fluxes. Simmons et al. [2010] compared observed surface air temperatures for 1973–2008 from Brohan et al. [2006] with collocated temperatures from the European Centre for Medium-Range Weather Forecast 40-Year Re-analysis Project (ERA-40) and ERA-interim reanalyses, which only use observed surface air temperatures indirectly through the soil thermodynamics, and found that observations-minus-reanalysis trends on continental scales were typically about 10% of the observed trends.

[4] However, McKitrick and Michaels [2007] analyzed surface air temperature trend fields statistically in terms of national socioeconomic indicators and geographical data, including tropospheric temperatures, rather than making physical comparisons through a reanalysis, and they concluded that about half the trend in global-average land surface air temperature in 1980–2002 arose from urbanization and other land surface changes, and from faults in the observations owing to, for example, station moves, instrumental changes, and observation schedule changes. McKitrick and Nierenberg [2010] and Schmidt [2009] obtained generally weaker relationships with the socioeconomic indicators when they used the Mears et al. [2003] microwave-sounder tropospheric temperature data rather than the equivalent University of Alabama data [Christy et al., 2003] used by McKitrick and Michaels [2007]: this suggests that the statistical relationships may not be robust. Schmidt [2009] also applied the technique of McKitrick and Michaels [2007] to suites of numerical model simulations and found instability, traceable to the spatial autocorrelation of the socioeconomic indicators. Nonetheless McKitrick [2010] defended this aspect of the robustness of the relationships using analytical spatial structure functions. He also included indices of four major atmospheric circulation patterns among the explanatory variables and found that this did not affect the overall conclusion that nearly half the warming trend over land arose from local to regional socioeconomic effects plus faulty data. The circulation indices were monthly, but they were applied without seasonal discrimination.

[5] A physically based and seasonally resolved estimate of the influence of atmospheric circulation on surface air temperature trends is a prerequisite for unraveling complex socioeconomic influences such as changes to the land surface. Four monthly indices, applied without seasonal discrimination, cannot fully specify the influence of atmospheric circulation on surface air temperature. Neither can tropospheric temperatures, because, for example, strong near-surface temperature inversions occur in calm, anticyclonic conditions, but not in the presence of strong winds or thick clouds.

[6] The purpose of this paper is to estimate physically the effects of atmospheric circulation and global forcings on land surface air temperature trends, and thereby to estimate the combined land surface and urban and data-inhomogeneity influences as a residual. This is done using the 20th Century Reanalysis (20CR [Compo et al., 2006, 2011]). 20CR covers 1871–2008 and is driven only by observed sea surface temperatures and sea ice [Rayner et al., 2003], atmospheric CO2 concentrations, solar and volcanic forcings, and observations of surface pressure over both land and ocean. No land surface changes are specified in 20CR. The analyzed period, 1979–2008, covers the satellite era and the greatest development of the Earth's land surface. The 20CR is more clearly independent of the surface temperature observations than are the ERA-40 and ERA-interim reanalyses used by Simmons et al. [2010]. The atmospheric circulation in 20CR is controlled by the pressure data with some influence from the sea surface temperatures that also partly mediate the influences of global forcings. The reliable reproduction by 20CR of the influence of atmospheric circulation on samples of daily land surface air temperatures is shown in Appendix A.

2. Analysis Method

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Analysis Method
  5. 3. Global Trends
  6. 4. Spatial Patterns
  7. 5. Uncertainties in CRUTEM and in 20CR
  8. 6. Conclusion
  9. Appendix A
  10. Acknowledgments
  11. References
  12. Supporting Information

[7] Annual temperature trends for 1979–2008 were calculated from the monthly CRUTEM3 data set [Brohan et al., 2006] on its 5° latitude × 5° longitude grid (Figure 1a). For an annual trend value to be plotted, a grid box required data for at least 10 months in each year. The data in CRUTEM3 are derived from observing station data that had first been converted to anomalies relative to 1961–1990. Annual trends for 1979–2008 were also calculated from 20CR monthly ensemble-mean 2 m air temperature fields after converting these to anomalies relative to 1961–1990 and then interpolating from their ∼1.9° latitude × 1.875° longitude Gaussian grid [Compo et al., 2011] to the CRUTEM3 grid (Figure 1b). The regridding was done by averaging the nine 20CR grid anomalies closest to the center of the CRUTEM3 grid, with weighting of 1.0 if the distance from this center was up to 1° latitude equivalent, and inverse-distance weighting (in units of °latitude equivalent) otherwise. Finally, annual trends were calculated from differences between CRUTEM3 and 20CR anomalies where CRUTEM3 had sufficient data (Figure 1c). Because the influences of land surface forcings may differ between winter and summer (for example, the effects of snow cover in winter and of irrigation in the growing season), the trends were also calculated for November through April (Figure 2) and May through October (Figure 3), stipulating a minimum of 5 months’ data in each yearly 6-month season.

image

Figure 1. Annual surface air temperature trends during 1979–2008. (a) CRUTEM3 at locations with at least 10 months’ data in each year. (b) 20CR. (c) CRUTEM3 minus 20CR with the same sampling criterion as for CRUTEM3. The color scale applies to all three plots.

Download figure to PowerPoint

image

Figure 2. As in Figure 1 but for November through April and with at least 5 months' data in each year.

Download figure to PowerPoint

image

Figure 3. As in Figure 1 but for May through October and with at least 5 months' data in each year.

Download figure to PowerPoint

3. Global Trends

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Analysis Method
  5. 3. Global Trends
  6. 4. Spatial Patterns
  7. 5. Uncertainties in CRUTEM and in 20CR
  8. 6. Conclusion
  9. Appendix A
  10. Acknowledgments
  11. References
  12. Supporting Information

[8] Figure 1a shows trends of annual average land surface air temperature during 1979–2008 in CRUTEM3. The global average of these trends was 0.26°C per decade. Trends in the corresponding 20CR analysis are shown in Figure 1b, and Figure 1c shows trends of CRUTEM3 relative to 20CR. Over land areas with sufficient data, CRUTEM3 warmed on average by 0.05°C per decade relative to 20CR. This implies that 0.21°C per decade, or a little more than 80% of the total CRUTEM3 warming, took place in 20CR where CRUTEM3 had observations. This is very close to the 0.20°C per decade trend over the fully sampled global landmass in 1979–2008 in V3.3 of the Remote Sensing Systems lower-tropospheric temperature series (http://www.remss.com/data/msu/monthly_time_series/), though less close to the 0.16°C per decade trend in the corresponding V5.4 University of Alabama–Huntsville analysis (http://vortex.nsstc.uah.edu/data/msu/t2lt/uahncdc.lt). See also Figure 10 of Mears and Wentz [2009] for comparative maps, but note that surface and lower-tropospheric trends are not fully coupled, as stressed above.

[9] The warming of CRUTEM3 by an average of 0.05°C per decade relative to 20CR also implies that not quite 20% of the total net warming in CRUTEM3 arose from a combination of land use changes, aerosol increases and decreases, and changes in minor greenhouse gases not included in 20CR, as well as from possible residual urbanization influences and errors in CRUTEM3. Trends of 20CR and Brohan et al. [2006] temperatures over the oceans are not compared because the Rayner et al. [2003] sea surface temperatures used in 20CR force close agreement.

[10] Figures 2 and 3 correspond to Figure 1 but are for November through April and May through October, respectively. The warming of CRUTEM3 in November through April was 0.28°C per decade on a global average, and exceeded that at the same locations in 20CR by 0.09°C per decade. However, in May through October there was zero difference between CRUTEM3 at 0.25°C per decade and collocated 20CR. Urban heat islands tend to be stronger in the summer half-year, albeit with exceptions [Arnfield, 2003]. So Table 1, which shows the relative trends for each calendar month, and Figures 2 and 3 do not appear to support any substantial local urban influence on the relative warming of the Northern Hemisphere-dominated CRUTEM3.

Table 1. Global Average Trends, CRUTEM3 Minus Collocated 20CR, 1979–2008
MonthRelative Trend (°C/Decade)
January0.10
February0.09
March0.11
April0.10
May0.05
June−0.00
July0.02
August−0.01
September−0.01
October−0.00
November0.07
December0.09

[11] The global radiative forcing from CO2 increased by 0.73 W m−2 between 1979 and 2008, whereas that from other greenhouse gases increased by 0.31 W m−2 [Blunden et al., 2011, Figure 2.52]. The omission of the latter from 20CR will have caused an underestimation of global warming, though not by as much as 30% (0.31/(0.73+0.31)) because all forcings (and the associated lag of the climate system) are implicit in the observed sea surface temperatures used in 20CR. The underestimation is unlikely to have been balanced by aerosol changes also absent from 20CR, because “dimming” by increasing tropospheric aerosols up to the 1980s was followed by “brightening” in the 1990s and, to a lesser extent, in the 2000s [Wild et al., 2009] as tropospheric aerosols were, on a global average, reduced. The ongoing though slight global brightening implies that, for 1979–2008, the trend of combined aerosol and non-CO2 greenhouse gas forcing is very likely to be positive. Note that brightening affects continental surface air temperatures without much thermal lag. Therefore, part of the overall 0.05°C per decade warming in CRUTEM3 relative to 20CR in 1979–2008 probably arose from increasing non-CO2 greenhouse gases and a limited amount of brightening.

4. Spatial Patterns

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Analysis Method
  5. 3. Global Trends
  6. 4. Spatial Patterns
  7. 5. Uncertainties in CRUTEM and in 20CR
  8. 6. Conclusion
  9. Appendix A
  10. Acknowledgments
  11. References
  12. Supporting Information

[12] The patterns of warming at 2 m over land in 20CR (Figures 1b, 2b, and 3b) show similarities to those in CRUTEM (Figures 1a, 2a, and 3a): for example, strong warming over Scandinavia and/or Europe, and over northern China and Mongolia, and slight winter cooling in parts of the Middle East. These reflect the observed atmospheric circulation changes that were imposed by the assimilated surface pressure observations. However, several spatial features merit additional comment. In northern winter, 20CR did not warm as much as CRUTEM3 over Siberia (Figure 2). This could reflect an inability of 20CR to fully model the influence of snow cover on Siberian winter near-surface temperatures, although Figure A1 (bottom) in Appendix A suggests generally good 20CR skills in central Siberia in winter. The annual cycle of relative trends in Table 1 is aligned with that of snow cover. Recent reductions in tropospheric aerosols may also have raised temperatures in spring. The patterns of differences in Figures 1c, 2c, and 3c do not suggest warming owing to increases in black carbon aerosols [Jones et al., 2011] over China and India, possibly because this has been exceeded by cooling from other forms of aerosol. In northern summer, a cooling effect of irrigation, or of anthropogenic aerosols [Kim et al., 2011], is suggested over the northwestern USA in Figure 3c. Omission of these could have led to local overheating in 20CR (Figure 3b).

[13] Aerosol forcings vary regionally and locally according to aerosol type and boundary conditions such as soil moisture [Nair et al., 2011]. The trend patterns analyzed by McKitrick and Michaels [2007] and McKitrick and Nierenberg [2010] may reflect these aspects, along with atmospheric aerosol reductions in some socioeconomically developed areas. The residue of the overall 0.05°C per decade relative warming, which relates to the entire observed landmass, need not contradict the results of Efthymiadis and Jones [2010], whose upper limit of 0.02°C per decade urban warming influence for 1951–2009 was based on coastal areas: note that most instances of strong relative warming in Figure 1 are not near coasts. Nor does it contradict the findings of very small overall urban effects by Hansen et al. [2010] who did not attempt to adjust data to compensate for rural land use changes, regarding them as a larger-scale forcing.

5. Uncertainties in CRUTEM and in 20CR

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Analysis Method
  5. 3. Global Trends
  6. 4. Spatial Patterns
  7. 5. Uncertainties in CRUTEM and in 20CR
  8. 6. Conclusion
  9. Appendix A
  10. Acknowledgments
  11. References
  12. Supporting Information

[14] The relative trends in Figures 1c, 2c, and 3c are also likely to been affected locally by errors in the CRUTEM3 data. Indeed, 20CR has potential as a quality-improvement tool for observed surface air temperatures, though ideally accompanied by station history data to confirm suspected errors and biases, because 20CR also has some uncertainties. These are reflected, for example, in a few isolated strong trends in Figures 1b, 2b, and 3b, and in an underestimation of the extreme global average warmth of the lower troposphere during the 1998 El Niño event [Blunden et al., 2011, Figure 2.5]. In winter and spring, the extent of snow cover in 20CR may differ from that in the real world, reducing the fidelity of the 20CR 2 m air temperatures. Nonetheless, given additional historical surface pressure data that are being rescued, future versions of 20CR could be used to assess longer-term atmospheric circulation and other influences on surface temperature back into the nineteenth century.

6. Conclusion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Analysis Method
  5. 3. Global Trends
  6. 4. Spatial Patterns
  7. 5. Uncertainties in CRUTEM and in 20CR
  8. 6. Conclusion
  9. Appendix A
  10. Acknowledgments
  11. References
  12. Supporting Information

[15] The 20CR has been used to demonstrate that, on a global annual average, a little more than 80% of the land surface air temperature trends observed during 1979–2008 can be ascribed to global forcings (CO2, volcanic, and solar), and variations of atmospheric circulation. However, there are substantial regional and seasonal variations that may be assigned tentatively to land use changes, including urbanization, aerosol increases and decreases, and changes in minor greenhouse gases not accounted for in 20CR. Residual uncertainties in the CRUTEM data quality and in 20CR are amenable to improvement given the incorporation of newly rescued observational data.

Appendix A

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Analysis Method
  5. 3. Global Trends
  6. 4. Spatial Patterns
  7. 5. Uncertainties in CRUTEM and in 20CR
  8. 6. Conclusion
  9. Appendix A
  10. Acknowledgments
  11. References
  12. Supporting Information

[16] The strong, reliably reproduced influence of atmospheric circulation on continental surface air temperature in ensemble means of 20CR is shown by Figure A1. 20CR estimates day-to-day variations in temperature with anomaly correlations of the order of 0.8 with observations at two Northern Hemisphere sites remote from the ocean. Correlations are a little lower for an Australian site (Table A1), because day-to-day temperature variations are less strongly controlled by atmospheric circulation at lower latitudes; correlations are about 0.9 for central England which is an average of 3 or 4 sites [Parker et al., 1992] and therefore more comparable to a gridded value.

image

Figure A1. Daily mean air temperatures in 1993. (top) North Platte, USA (41°N, 101°W); (bottom) Kolpashevo, Russia (58°N, 83°E). Blue, observed; red dashed, 20CR; heavy black, observed 1961–1990 daily climatological average; yellow shading, range within the light black observed 1961–1990 5th and 95th percentiles. 20CR values are anomalies relative to 20CR 1961–1990 daily climatological average at its nearest grid point, added to the observed daily climatological average. Anomaly correlations of observations versus 20CR: North Platte, 0.75; Kolpashevo, 0.86.

Download figure to PowerPoint

Table A1. Correlations Between Daily Anomalies (Relative to 1961–1990 Climatology) of Observed Mean Temperature and of Mean Temperature Simulated by 20CR at Its Nearest Grid Point
Location198519931998
Alice Springs, Australia (24°S, 134°E)0.710.740.75
Kolpashevo, Russia (58°N, 83°E)0.810.860.87
North Platte, USA (41°N, 101°W)0.830.750.85
Central England (52°N, 2°W)0.910.900.89

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Analysis Method
  5. 3. Global Trends
  6. 4. Spatial Patterns
  7. 5. Uncertainties in CRUTEM and in 20CR
  8. 6. Conclusion
  9. Appendix A
  10. Acknowledgments
  11. References
  12. Supporting Information

[17] D. E. Parker was supported by the Joint DECC/Defra Met Office Hadley Centre Climate Programme (GA01101). D. Parker thanks three anonymous reviewers for their helpful comments. Support for the Twentieth Century Reanalysis Project data set is provided by the U.S. Department of Energy, Office of Science Innovative and Novel Computational Impact on Theory and Experiment (DOE INCITE) program, and Office of Biological and Environmental Research (BER), and by the National Oceanic and Atmospheric Administration Climate Program Office.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Analysis Method
  5. 3. Global Trends
  6. 4. Spatial Patterns
  7. 5. Uncertainties in CRUTEM and in 20CR
  8. 6. Conclusion
  9. Appendix A
  10. Acknowledgments
  11. References
  12. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Analysis Method
  5. 3. Global Trends
  6. 4. Spatial Patterns
  7. 5. Uncertainties in CRUTEM and in 20CR
  8. 6. Conclusion
  9. Appendix A
  10. Acknowledgments
  11. References
  12. Supporting Information
FilenameFormatSizeDescription
jgrd17482-sup-0001-t01.txtplain text document0KTab-delimited Table 1.
jgrd17482-sup-0002-taA01.txtplain text document0KTab-delimited Table A1.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.