Hansen et al. 2006
Supplementary material for the article "Global temperature change" includes additional text and three figures.
Global Temperature Analysis Method. Analyses of global temperature change by different groups, particularly, NASA Goddard Institute for Space Studies (GISS), the NOAA National Climate Data Center (NCDC), and the combination of the British Meteorological Office and the University of East Anglia (BMO/UEA), are generally in close agreement. The ranking of individual years, however, often depends upon differences of only several hundredths of a degree, which is finer than the accuracy that any method can claim given observational limitations.
One major source of differences is the fact that the GISS method extrapolates temperature anomalies to all areas that have at least one station located within 1,200 km, using weights for these stations that decrease linearly with distance from the station. At any given location, the temperature anomaly estimated in this way can be substantially in error, but the resulting increase of spatial coverage usually allows an improved estimate of the global temperature anomaly, as judged from tests made using spatially and temporally complete data sets generated by a general circulation model (1). However, in some cases, this extrapolation increases the error by giving undue weight to a single isolated station with anomalous temperature.
Another source of difference is the method of averaging over the world, given the fact that data are not available everywhere. In the GISS method, the Earth is divided into four latitude belts. Within each belt, all regions having at least one station within 1,200 km are included. Temperature anomalies are calculated for rectangular latitude-longitude boxes that have approximately equal latitude and longitude dimensions (1). The anomaly for an entire latitude belt is taken as the area-weighted anomaly for boxes within that belt that have defined temperature anomalies. The global anomaly is then the area-weighted mean of the four belts. This method gives equal weight to the hemispheres, but if one of the belts has little data and if those data are not actually representative of the entire belt, substantial error can occur.
The land (meteorological stations) data sets in the GISS, NCDC, and BMO/UEA analyses have considerable commonality, but they are not identical. Our approach, which has been described in detail (2), uses GHCN (Global Historical Climatology Network) meteorological station data obtained from NCDC supplemented by Antarctic station data from the British Antarctic Survey web site (www.antarctica.ac.uk/met/READER). Quality control of GHCN data relies primarily on data checking at NCDC (3), which occurs before our receipt of the data. Minimal additional quality checking is carried out at GISS (2), e.g., flagging stations with temperatures that differ by five standard deviations from their long-term mean, so that these can be checked against neighboring stations. Urban station data are adjusted so that the long-term trend of the urban station matches that of neighboring rural stations, with the distinction between urban and rural based on either population or nightlights observed by satellite (4).
Our ocean data set is version 2 of the "OI" analysis (5) for the period of satellite data, i.e., after 1982. Earlier ocean data are from ships and buoys (6). These two ocean data sets are combined by working with anomalies for both data sets and defining anomalies relative to a common period, specifically 1982-1992.
Two minor changes in our analysis were tested this year with the objective of making the calculations and results simpler and more transparent. The first of these changes was adopted in March 2006 and affects our analyses updated each month and made available on the World Wide Web, for data covering the entire period of record. The second change has not been accepted in our analyses for the reason given below.
In the first change to our analysis method, the ocean area where we used ship and buoy data (6) was reduced, so as to be identical to the ocean area where satellite data are reported (5). This eliminated the time variation of ocean area that occurs in the ship and buoy data (6) as sea ice area changes. Temperature anomalies at the times and small areas where ship and buoy data are no longer used are based on weighted means of anomalies in all gridboxes within 1,200 km, thus giving slightly more weight to surface air measurements as opposed to SST. The effect of this change on global temperature can be as much as a few hundredths of a degree. This change has been adopted in the temperature records in this paper and on our web site.
In the second change to our analysis, we replaced the calculation of global temperature anomaly, which proceeds by successively combining anomalies of subboxes, boxes, and latitude zones (1), with a simpler integration over the globe. Temperature anomalies at all longitudes were averaged to obtain zonal mean anomalies at resolution 2° of latitude, and the temperature anomalies for these latitude belts were combined to obtain the global anomaly. However, this method was found to be slightly less accurate than the subbox, box, latitude zone approach in tests that used a globally complete data set generated by climate model simulations for the period 1880-2003. Thus, this second simplification has not been adopted in our continuing data analyses.
Additional Sea Surface Temperature Data. Fig. 3 shows an absence of warming of 2001-2005 relative to 1870-1900 in the equatorial region of upwelling off the coast of South America. Fig. 7 shows that the same conclusion holds for linear trends of SSTs. Alternative choices for the beginning date, e.g., 1880 or 1900, do not alter this conclusion qualitatively. Maps of the temperature change for arbitrary choices of the beginning and ending dates are available at www.giss.nasa.gov/data/gistemp.
Fig. 8 shows annual mean SSTs in the WEP and EEP based on ship and buoy data (6) for 1870-1981 and satellite data (5) for subsequent years. The satellite data are adjusted by a small constant, as shown in Fig. 8A, such that the mean 1982-1992 temperature matches the in situ data (6) for that 11-year period. The 1983 and 1998 El Niños stand out in these annual mean plots as well as in the 12-month running means shown in Fig. 3B.
Fig. 5, using the same data source as Fig. 4A, extends the paleoclimate data for the WEP back to 1.35 million years before present. At least several interglacials in this longer period were warmer than the Holocene. As discussed in the text, alignment of the paleo temperatures and the modern instrumental data are uncertain by up to 1°C. If the paleo temperatures are shifted upward (by about one-half degree Celsius) such that the temperature in the late 1800s is near the lowest value in the Holocene, the current temperature is still within ≈1°C of the warmest interglacials.
Correlation of Local and Global Temperature Change. Fig. 9 shows the correlation of local (2°by 2°) annual surface temperature with the global annual surface temperature for the period 1880-2005. The strongest correlations with global temperature change occur for the Indian Ocean, Indonesian region, the near-equatorial Atlantic Ocean region, and the South Indian Ocean to the southwest of Australia.
References for Supplementary Text
- Hansen J, Lebedeff S (1987) J Geophys Res 92:13345-13372.
- Hansen J, Ruedy R, Glascoe J, Sato M (1999) J Geophys Res 104:30997-31022.
- Peterson TC, Vose R, Schmoyer R, Razuvev V (1998) Int J Climatol 18:1169-1179.
- Hansen J, Ruedy R, Sato M, Imhoff M, Lawrence W, Easterling D, Peterson T, Karl T (2001) J Geophys Res 106:23947-23963.
- Reynolds RW, Smith TM (1994) J Clim 7: 929-948.
- Rayner N, Parker D, Horton E, Folland C, Alexander L, Rowell D, Kent E, Kaplan A (2003) J Geophys Res 108:10.1029/2002JD002670.
- Medina-Elizade M, Lea DW (2005) Science 310:1009-1012.
PDF documents require a special viewer such as the free Adobe Reader.