The Temperature Record

“Don’t trust thermometers”

It may seem trivial to track global temperatures, but it is actually more difficult than it would seem.

Two analyses

We will examine the two most widely used analyses of global surface temperature. The first is a joint effort between the UK Hadley Centre and Climatic Research Unit of the University of East Anglia (HadCRUT)1. The other is from the NASA Goddard Institute for Space Studies, or NASA GISS (GISTEMP)2.

Meaningful coverage of the earth is not possible before a certain point. The Hadley/CRU analysis begins at 1850. NASA GISS starts at 1880.

Since the late 19th Century, global temperature has risen about 0.75 °. This might not sound like much, but we’ve already observed significant changes as a result (see section 10) and the continuing increase does not bode well for the future (see sections 11 and 12).

The rise and fall of global temperature is actually that of temperature anomalies. These are variations in temperature from a pre-determined base period. For example, if the average temperature of a location in June from 1951 to 1980 was 20 °C, and the current June temperature is 21 °C, the temperature anomaly for the current June is 1 °, or 1 ° warmer than the norm. By calculating anomalies from points all over the world we can approximate the global surface temperature anomaly.

The most visible difference between the two analyses is due to their choice of base periods. NASA GISS calculates anomalies with respect to 1951 to 1980, and Hadley/CRU calculates theirs with respect to 1961 to 1990. The different base periods change the position of zero, but an increase in 1 ° for either is exactly the same. The rate of warming is virtually identical between the two analyses.

“It’s only the Urban Heat Island (UHI) effect!”

The surface temperature over land is recorded at meteorological stations throughout the world (also called weather stations or surface stations). Skeptics do not trust these records because they are believed to be hopelessly contaminated by the urban heat island effect. Because asphalt and concrete absorb and radiate heat, and because industry, buildings and vehicles generate heat, increasing urbanization around thermometers would introduce a warming bias.

If we ignore the records from urban areas, we still know that the world has warmed for several reasons:

  • Rural areas are warming.

  • The oceans are warming.

  • Glaciers and permafrost are melting, as are the ice sheets and sea ice (see section 10).

  • Sea levels are rising (see section 10).

  • Plants and animals are moving pole ward.

All of these independent observations point to widespread warming.

Correcting problems

NASA GISS attempts to reduce the influence of urban warming3. The meteorological data that NASA GISS uses is contained within the Global Historical Climatology Network (GHCN) and the US Historical Climatology Network (USHCN) datasets. Before attempting to correct for urban warming, other problems must be addressed. Measurements with obvious problems are excluded — such as those inconsistent with surrounding sites. Measurements are not always taken at the same time of day, so time of observation bias (TOBS) must be addressed. If a meteorological station is replaced or moved, adjustments are often required to make the records consistent. Over the years, different kinds of equipment have measured temperature, and small adjustments are required to correct for these changes.

A location may have several stations and those are combined into a single continuous series. To reduce the influence of urban warming, NASA GISS adjusts the trends from urban locations to match those of at least three rural locations within 1000 km. Otherwise, the urban location is dropped. The closer it is to the urban location receiving the adjustment, the larger the weight given to the rural location’s trend. The overall shape of the global temperature series, and thus the rate of warming, is determined entirely by stations classified as rural.

For the US, NASA GISS uses night-time satellite photographs to determine whether a location is “urban” or “rural”4.

For the rest of the world, they use the population near the location.

The graph shows the 5 meteorological stations within the USHCN measuring temperature for Washington DC between 1881 and 20085.

After the trend is adjusted to account for urbanization, the oldest data in this example receive an upward correction of over 1 °. The corrected temperature series is represented by the thick black line.

A (small) victory for the skeptics

A problem with the NASA GISS analysis of the US lower 48 states was discovered in 20076. The National Oceanic and Atmospheric Administration (NOAA) applies its own corrections to the USHCN dataset. NASA GISS had assumed that NOAA would apply these corrections in real time, but they were not present after the year 2000. The combination of corrected and uncorrected data created a 0.15 ° warm bias from January 2000 to June of 2007 for the contiguous 48 states. Skeptics have dubbed this the “Y2K error.”

The red line is the 5 year moving average of the corrected data, and the green line is the same for the old data. The difference between the green and red represents the magnitude of the correction.

Because the US represents less than 2% of the earth’s surface, the total effect on global temperatures was approximately 0.003 °.

“It was warmer in the ‘30s!”

This correction led many to believe that it was warmer in the ‘30s. Yet, temperatures in the ‘30s are clearly not comparable to those of the past few decades.

As on the previous graph, the green line shows the 5 year average before the correction, and the red line shows the 5 year average after the correction. If you don’t see a green line, the enlargement may help. You might barely make out parts of the green line.

So where did people get the idea that the ‘30s were warmer than the present?

The US is not the world!

During the dust bowl, temperatures in the US were very similar to those of the present. Correcting the Y2K problem changed the ranking of the warmest years in the US, and many confused this with global temperatures. Most famously, it changed the ranking of the two warmest years for the lower 48 states: 1934 and 1998. Although the problem did not involve 1998 and 1934 directly, changes in recent data cause urban trend adjustments to ripple backwards. Until recently, 1934 had the higher anomaly, but 1998 had taken the lead for a short while. The new correction put 1934 back on top (1934 and 1998 are circled)8.

None of this makes any difference however, because both years are and were statistically tied for the warmest recorded US temperature anomaly. That is why NASA GISS has never declared 1998 to be the record US temperature.

NASA GISS’ most recent paper documenting their temperature analysis methods makes all of this clear in 20019.

The US annual mean temperature is slightly warmer in 1934 than in 1998 in the GISS analysis. [...] the difference between 1934 and 1998 mean temperatures is a few hundredths of a °. [...] In comparing temperatures of years separated by 60 or 70 years, the uncertainties in various adjustments lead to an uncertainty of at least 0.1 °C. Thus it is not possible to declare a record U.S. temperature with confidence until a result is obtained that exceeds the temperature of 1934 by more than 0.1 °C.

Up with satellites. Down with thermometers.

Skeptics often point to satellite measurements as the superior alternative to thermometers. There are two leading analyses of satellite temperature data. The first is from the University of Alabama in Huntsville, or UAH. The second is the Remote Sensing Systems analysis, or RSS. Both start with the same satellite data and from that use different methods to form conclusions.

Satellites use instruments known as Microwave Sounding Units (MSU) to infer atmospheric temperature. There are several satellite “channels” designed to measure the emission of microwave radiation at different altitudes. Each channel corresponds to thick layers of atmosphere that overlap significantly with channels above and below. Because the atmosphere simultaneously warms and cools depending on altitude (see section 7), inferring temperature change at a specific altitude is not possible. In addition, the earth’s surface reflects some of this microwave radiation, and the amount of interference this creates depends on the type of surface. This interference must be accounted for and, in some cases, problem areas are excluded from the analysis.

Perhaps the most significant issue is that no single satellite has stayed in operation for the past 30 years, so the data must be matched and stitched together.

“The satellites are the most accurate thermometers.”

Of particular importance is the temperature trend of the lower troposphere, the layer that most closely corresponds to the temperature of the surface. Both RSS and UAH have developed synthetic channels intended to represent lower altitudes10. They are considered “synthetic” because there is no actual satellite channel that measures those altitudes so narrowly. Both analyses attempt to remove the influence of the atmosphere above the lower troposphere.

The UAH analysis is the older of the two. For a long time it showed little or no warming, leading many skeptics to believe that the warming recorded in meteorological records was an artifact of the urban heat island effect.

Over the years, a lot of problems with the UAH analysis have been identified, but the most significant was discovered in 2005. UAH didn’t properly adjust for orbital decay, which changes the time of day calculation. After this flaw was corrected, warming was found to be 40% higher than previously thought11.

The primary difference between UAH and RSS is due to the merging of data from two satellites in the early ‘90s. This and other differences are on top of the UAH flaw discovered in 2005.

To illustrate the magnitude of these differences, two earlier versions of UAH for the time period ending in September 2001 are compared to the current versions of UAH and RSS12.

To emphasize the accumulated difference between the analyses, the anomalies are shifted up or down so each straight line fit (not shown) intersects with a common starting point, in this case January 1979. The benefit of this method is that any differences between the analyses become increasingly apparent as time goes on.

RSS has not been free of problems either. An error worth about 0.15 ° persisted throughout all of 2007.

The bar graph compares the significance of these problems in relation to the NASA GISS “Y2K Error” which received widespread media coverage13. In contrast, the 2005 UAH diurnal correction is about 50 times larger.

Additionally, there are two other analyses of satellite data and both show significantly more warming than RSS and UAH14. If satellite data are as reliable as many skeptics proclaim, none of these errors and adjustments would exist.

Both the satellite and surface analyses have seen and will continue to see their share of revisions as problems with historical data and processing methods are discovered and corrected.

Temperature series compared

Below are the four most prominent temperature analyses covering the satellite era from 1979 through 200815.

The latest version of UAH shows warming of about 0.13 ° per decade between 1979 and 2008.

This is still low compared to the other analyses, however. Over the same time period, the RSS and two meteorological analyses all show warming of about 0.16 ° per decade.

Sea surface temperatures

What differences exist between the two surface analyses are partly related to the sources of data for each. For ocean areas, Hadley/CRU (blue) and NASA GISS (red) use different analyses of ship and buoy-based measurements. Since late 1981, NASA GISS (green) uses satellite sea surface temperature readings16.

The popular emphasis on the urban heat island effect often overshadows the ocean data, but these measurements have perhaps the largest discontinuities. This is significant because the oceans represent about 70% of the Earth’s surface. The Hadley/CRU ship and buoy data have warmed faster in recent decades compared to the satellite data used by NASA GISS.

Geographic coverage

Another significant difference involves the area covered by each analysis. The NASA GISS analysis is shown on top with Hadley/CRU below.

NASA GISS has done a statistical analysis of temperature data that shows that although absolute temperature at various locations are certainly different, their anomalies are highly correlated within a radius of 1200 km17. By extending anomalies out 1200 km from a location, NASA GISS increases coverage of land surface and sea ice. This has become increasingly important as the availability of Arctic data in Canada and the former Soviet Union, and tropical data in South America and Africa has declined over the last two decades. Those areas, in addition to the poorly sampled interior of Antarctica, are more completely represented in the NASA GISS analysis.

Because both land surface and sea ice warm more quickly, a greater ratio of these areas to ocean surface increases the calculated rate of warming for the globe. The graph below shows the rate of warming based on a smaller 250 km radius, compared to NASA GISS’ standard 1200 km radius. The bar graph shows where this additional calculated warming is located18.

Over half of the difference is in polar areas, while 40% is in the tropics – all poorly sampled areas.

The satellites are limited in their scope as well. They cannot view the small areas surrounding the north and south poles. More significantly, the large Southern Hemisphere ice cover interferes with the satellite readings, as do the high altitude regions of the Himalayan and Andes mountains. As a result, RSS excludes those regions from its analysis. UAH includes them, potentially creating measurement problems.

For RSS, the result is shown19.

The unshaded areas are excluded from its analysis.

The big picture

Much of this discussion is largely an exercise in minutia, which is obvious when we put all of the data together20.

What’s apparent is that the satellite measurements of the lower troposphere are more variable than the surface measurements, although the trends are very similar. You can see this similarity on the left, which is the 5 year moving average for each temperature series over the satellite era beginning in 1979.

On this graph it’s evident that something unusual was going on in 1998, with temperatures showing a huge spike, especially for UAH and RSS. That is important because . . .

“Global Warming stopped in 1998!”

For three of the four analyses, 1998 remains the warmest year on record through 2008. Many skeptics assume that because the record has not been unambiguously broken, global warming has stopped.

But one glance at the graph shows that this is a disingenuous claim. 1998 was unusually warm due to the super strong El Niño effect that year. The only thing that it proves is that we haven’t had as strong an El Niño since. Regardless of the analysis, 2001, 2002, 2003, 2004, 2005, 2006, and 2007 have all been warmer than any year prior to 1998. In the NASA analysis, 2005 is the warmest year, and 2007 and 1998 are tied for second place.

Two major causes of natural fluctuations

In the short term, anthropogenic global warming (AGW) is not the strongest influence on global temperature, nor are changes in solar activity (see section 6). The two major causes of short term fluctuations are explosive volcanic eruptions and the El Niño/Southern Oscillation (ENSO).

Large volcanic eruptions eject debris into the stratosphere which encircles the earth. The large 1982 El Chichon and 1991 Pinatubo eruptions reduced solar radiation reaching the surface21.

Both eruptions lowered global temperatures about 0.3 ° when averaged over the course of 12 months.

But it is ENSO that is relevant here. ENSO is a pattern of ocean temperatures in the Pacific. Ordinarily, a pool of warm water lingers in the West Pacific. Occasionally, this warm water spreads out over much of the Pacific, warming the air and increasing the moisture of the atmosphere. The result is a temporary increase in global temperatures and changes in precipitation patterns.

The graph shows the positive and negative phases of ENSO from 1979 to 200822.

When ENSO is in its warm phase, it is called El Niño — Spanish for little ”little boy” because it often associated with Christmas. In its cool phase, it is called La Niña or “little girl”. There are other ocean oscillations, but ENSO has the strongest effect on global temperatures.

The results of the more significant volcanic and ENSO events between 1979 and 2008 are labeled below.

With respect to the last decade, the relative warmth of 1998 (El Niño) and coldness of 2008 (La Niña) are much stronger as measured by the satellites, which gives them the highest starting point and lowest end point (example below)23.

Averaged over 12 months, the temporary increase for the 1998 El Niño was between 0.2 and 0.4 °. The 2008 La Niña lowered global temperatures about 0.2 to 0.3 °. In contrast, anthropogenic global warming is currently responsible for less than 0.20 ° of warming per decade. Under short time scales, the oscillation of the oceans will bury AGW.

“If it’s getting warmer, why is it cold outside?”

Global warming does not eliminate cold weather. This is obvious if we look at the warmest month in the history of the NASA GISS analysis: January 200724.

You can see the extreme warmth of the high northern latitudes. You can also see that El Niño conditions exist due to the warm band of Pacific water along the equator. Despite this, there were still some cold areas. Globally, the temperature anomaly was very warm.

Alternatively, we can look at the coldest month in recent years, January 2008 — exactly one year later.

Most areas are relatively cold, but areas of extreme warmth exist, particularly in the Arctic. In contrast to the previous January, the cool water in the Pacific indicates a strong La Niña.

It is always cold some place in the world, and it is always warm some place in the world. However, whether you use thermometers on land, ships and buoys in the ocean, satellites that measure sea surface temperatures or satellites that measure atmospheric temperatures, the trend is up. Such data are consistent with our expectations strong natural fluctuations superimposed on anthropogenic global warming.


  1. (Brohan et. al. 2006) Online here. Data here

  2. (Hansen, Ruedy, & Sato, 2001) Online here. Data here

  3. (Hansen, Ruedy, Glascoe, & Sato, 1999) Online here, (Hansen, Ruedy, & Sato, 2001) Online here. An ongoing effort to disprove NASA GISS’ urban adjustments has resulted in some ironic findings. The Surface Stations project was set up to document the numerous quality control violations of various meteorological stations throughout the US. A preliminary analysis using the highest quality stations has produced a temperature series that is virtually identical to the one produced by GISTEMP.

  4. (Hansen, Ruedy, & Sato, 2001) Online here

  5. (NASA GISS) Station data here

  6. (Hansen, 2007) Online here. (Hansen, The Real Deal: Usafruct & the Gorilla, 2007) Online here

  7. ibid

  8. ibid

  9. (Hansen, Ruedy, & Sato, 2001) Online here

  10. (Mears & Wentz, Accepted) Preprint here, (Christy et. al, 2003) Online here.

  11. (Gentry, 2005) Online here. Spencer and Christy are generally considered to be skeptics, although they acknowledge that human beings are playing some role in the climate.

  12. (Mears & Wentz, Accepted) Preprint here. Data here. (Christy et. al, 2003) Online here. Data here. Archived UAH data here. For many of the graphs in this section, each temperature series is offset so that the linear trends have a common intercept. This makes it easier to see the accumulated difference between each series as time goes on. The trends are calculated and the offsets are applied before the series are smoothed.

  13. The magnitude of each error is approximate. The “Y2K Error” caused a 0.15 ° warm bias for the lower 48 states, which is less than 2% of the world’s surface. Worldwide, this implies closer to 0.002 ° error, but is rounded up to 0.003 °. The problem applied to January 2000 through June 2007, or 90 months, for a total of 0.27 “degree-months”. The “RSS 2007 Error” was about 0.15 ° and applied to all of 2007, for a total of 1.8 degree-months. The RSS-UAH “Step Change” begins in January 1992, and is about 0.05 ° based on the average difference for five years before the step change vs the average difference for five years after the step change. Through 2008, the time period totals 204 months. The 2005 UAH “Diurnal Correction” accumulated to 0.09 degrees based on trends from December 1978 to June 2005. The magnitude of the correction is the area of a triangle 0.09 ° high, and 307 months wide, or 13.8 degree-months.

  14. (Vinnikov et. al., 2006) Preprint here. (Fu & Johanson, 2005) Online here.

  15. (Mears & Wentz, Accepted) Preprint here. Data here. (Christy et. al, 2003) Online here. Data here. (Brohan et. al. 2006) Online here. Data here. (Hansen, Ruedy, & Sato, 2001) Online here. Data here.

  16. (Rayner et. al. 2006) Online here. Web page here. Data here. (Rayner et. al., 2003) Online here. Web page here. Data available by contacting Hadley Centre. (Reynolds et. al., 2002) Online here. Web page here. Instructions for time series here.

  17. (Hansen & Lebedeff, 1987). Abstract here

  18. (NASA GISS) Map creation tool here. The tool allows you to create plots with a 250 km smoothing radius. The anomaly is printed on the image. To calculate the location of this additional warming by latitude, create 1979 – 2008 trend maps for both 1200 and 250 km smoothing. Download the “zonal means plot” as a text file. Take the difference of both plots and weight each by area.

  19. (RSS) Online here

  20. (Mears & Wentz, Accepted) Preprint here. Data here. (Christy et. al, 2003) Online here. Data here. (Brohan et. al. 2006) Online here. Data here. (Hansen, Ruedy, & Sato, 2001) Online here. Data here.

  21. (Mauna Loa Observatory) Online here. Direct link to graphic here.

  22. (Wolter & Timlin, 1993) Online here. Website here. Data here.

  23. (Mears & Wentz, Accepted) Preprint here. Data here. (Hansen, Ruedy, & Sato, 2001) Online here. Data here.

  24. (NASA GISS) Map creation tool here

Sources used in The Temperature Record

Brohan, P., Kennedy, J., Harris, I., Tett, S., & Jones, P. (2006). Uncertainty estimates in regional and global observed temperature changes: a new dataset from 1850. Journal of Geophysical Research , 111.

Christy, J., Spencer, R. W., Norris, W. B., & Braswell, W. D. (2003). Error Estimates of Version 5.0 of MSU-AMSU Bulk Atmospheric Temperatures. Journal of Atmospheric and Oceanic Technology , 20, 613-629.

Christy, J., Spencer, R., Norris, W., & Braswell, W. (n.d.). Retrieved January 2009, from The National Space Science & Technology Center:

Fu, Q., & Johanson, C. M. (2005). Satellite-derived vertical dependence of tropical tropospheric temperature trends. Geophysical Research Letters , 32.

Gentry, P. (2005, August 11). California group's answer to climate puzzler improves the accuracy of global climate data. Retrieved June 5, 2008, from UAHuntsville: The University of Alabama in Huntsville:

Hansen, J. (2007, August 10). A Light On Upstairs. Retrieved June 6, 2008, from Dr. James E. Hansen:

Hansen, J. (2007, August 16). The Real Deal: Usafruct & the Gorilla. Retrieved June 5, 2008, from Dr. James E. Hansen:

Hansen, J., & Lebedeff, S. (1987). Global trends of mesasured surface air temperature. Journal of Geophysical Research , 13345-13372.

Hansen, J., Ruedy, R., & Sato, M. (2001). A closer look at United States and global surface temperature change. Journal of Geophysical Research , 23947-23963.

Hansen, J., Ruedy, R., Glascoe, J., & Sato, M. (1999). GISS analysis of surface temperature change. Journal of Geophysical Research , 30997-31002.

Mauna Loa Observatory. (n.d.). ESLR Global Monitoring Division - Climate. Retrieved June 5, 2008, from NOAA Earth System Research Laboratory:

Mears, C. A., & Wentz, F. J. (Accepted). Construction of the RSS V3.2 lower tropospheric temperature dataset from the MSU and AMSU microwave sounders. Journal of Amtospheric and Oceanic Technology .

NASA GISS. (n.d.). GISS Surface Temperature Analysis (GISTEMP). Retrieved January 2008, from Goddard Institute for Space Studies:

NASA GISS. (n.d.). Some Arctic region removed. Retrieved Jun 4, 2008, from Goddard Institute for Space Studies:

Rayner, N., Brohan, P., Parker, D., Folland, C., Kennedy, J., Vanicek, M., et al. (2006). Improved analyses of changes and uncertainties in sea surface temperature measured in situ since the mid nineteenth century: the HadSST2 data set. Journal of Climate , 19 (3), 446-469.

Rayner, N., Parker, D., Horton, E., Folland, C., Alexander, L., Rowell, D., et al. (2003). Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. Journal of Geophysical Research , 108 (D14).

Reynolds, R., Rayner, N., Smith, T., Stokes, D., & Wang, W. (2002). An improved in situ and satellite SST analysis for climate. Journal of Climate , 1609-1625.

RSS. (n.d.). RSS / MSU and AMSU Data Description . Retrieved January 2008, from Remote Sensing Systems:

Trenberth, K., Jones, P., Ambenje, P., R., B., Easterling, D., Tank, A. K., et al. (2007). Observations: Surface and Atmospheric Climate Change. In S. Solomon, D. Qin, M. Manning, Z. Chen, M. Marquis, K. Averyt, et al. (Eds.), Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (pp. 237-336). Cambridge, United Kingdom and New York, NY, USA: Cambridge University Press.

UK Hadley Centre and Climatic Research Unit of the University of East Anglia. (n.d.). Temperature Data (HADCRUT3 and CRUTEMP3). Retrieved January 2008, from Climatic Research Unit:

Vinnokov, K. Y., Grody, N. C., Robock, A., Stouffer, R. J., Jones, P. D., & Goldberg, M. D. (2006). Temperature trends at the surface and in the troposphere. Journal of Geophysical Research , 111.

Wolter, K. (2008, June 4). Multivariate ENSO Index. Retrieved June 5, 2008, from Earth System Research Laboratory: Physical Sciences Division:

Wolter, K., & Timlin, M. (1993). Monitoring ENSO in COADS with seasonally adjusted principal component index. Proceedings of the 17th Climate Diagnostics Workshop (pp. 52-57). Norman, OK: NOAA/NMC/CAC, NSSL, Oclahoma Clim. Survey, CIMMS and teh School of Metor., Univ. of Oklahoma.