Copernicus Climate Change Service, ECMWF
Uncertainties in ERA-Interim
Comparing alternative products helps to identify where and when they provide reliable information.
In the case of differences it may suggest causes and the product or products in which more
confidence can be placed.
Time series and maps of surface air temperatures from ERA-Interim are compared here with those
from three other datasets. One is a newer reanalysis than ERA-Interim, the Japan Meteorological
Agency’s JRA-55, which like ERA-Interim includes an analysis of surface air temperature
observations. As with ERA-Interim, analysed values over land are blended with values from the
background forecast model over sea. The other two datasets are the latest versions of products that
have been conventionally used to characterize the long-term warming of the atmosphere that has
occurred since the 19th century: HadCRUT4, produced by the Met Office in collaboration with the
Climatic Research Unit of the University of East Anglia, and NOAAGlobalTemp, produced by the US
National Centers for Environmental Information. These datasets combine analyses of climatological
reports of monthly-mean surface air temperature from stations over land with analyses of seasurface temperature (SST). SST is used rather than marine surface air temperature as the latter is
difficult to analyse reliably directly from observations.
The four datasets each use a different analysis of SST. The analysis used by ERA-Interim differs from
the others in that it includes data from satellites. HadCRUT4 is an ensemble of 100 realisations. The
ensemble samples uncertainty in the multi-decadal variability represented by the dataset, but does
not shed light on uncertainty in the representation of short-term variations. The HadCRUT4 results
presented here are from the medians of the monthly ensemble values.
Figure 1 shows time series of twelve-month running averages of the four estimates of global
temperature. Each of the datasets provides a similar overall picture: warming since the 1970s is not
in doubt, nor is the occurrence of warmer and colder spells linked with El Niño events, volcanic
eruptions and other sources of variability. All show a quite similar level of current warmth, with
temperatures in the range from 0.34OC to 0.38OC warmer than the 1981-2010 average.
The datasets nevertheless differ in their estimates of the strengths of individual warm spells, and in
their rankings of calendar years as to their warmth. ERA-Interim shows the largest peaks, exceeding
the 2014 calendar-year average for 12-month periods within 2005–2006 and 2009–2010, including
the calendar years of 2005, 2006 and 2010. JRA-55 has 2014 as the warmest year by a narrow
margin, but also has a slightly warmer 12-month spell in 2009–2010. NOAAGlobalTemp shows the
lowest maxima in the period from 1999 to 2013; 2014 clearly is the warmest year for this particular
dataset. The temperature anomaly for the year 2005 varies from 0.23OC for NOAAGlobalTemp to
0.35OC for ERA-Interim.
A significant factor behind these differences is the limited number of direct measurements of
temperature made in the high Arctic and in the Southern Ocean and Antarctica, which results in data
voids in the HadCRUT4 and NOAAGlobalTemp datasets. A second factor is uncertainty in the
analyses of sea-surface temperature, whose production entails making adjustments to the data from
various observing platforms that sample sea water at different depths and with different
measurement biases. ERA-Interim has been shown to be in reasonable agreement with both JRA-55
and such direct temperature measurements as are available in the high Arctic, where anomalously
September 2015
Copernicus Climate Change Service, ECMWF
warm temperatures contribute more to the 2005 and 2012 peaks in global temperature in the case
of the reanalyses. JRA-55 nevertheless does not produce as high a peak as ERA-Interim in 2005,
partly because of relatively low SSTs and partly because of differences between the two reanalyses
over Antarctica. NOAAGlobalTemp also has relatively low SSTs at the time, compared with both
HadCRUT4 and ERA-Interim.
Figure 2 shows time series of average temperatures over European land areas. Variability is much
higher for this considerably smaller domain, but the region is well observed, and all four datasets are
in good agreement. The reanalyses exhibit slightly larger maxima and minima, as does HadCRUT4
compared with NOAAGlobalTemp.
Maps of annual-mean temperature differences from climatological averages for 1981-2010 are
presented in Figure 3 for the years 2012 and 2014. HadCRUT4 and NOAAGlobalTemp data are
supplied on a relatively coarse 5Ox5O grid, HadCRUT4 data as deviations from the 1961-1990 average
and NOAAGlobalTemp as deviations from 1971-2000. Values adjusted to be relative to 1981-2010
are shown only for grid squares where there are data for at least 90% of the months from 1981 to
2010, and only for grid squares at which a value is provided for every month of 2012 and 2014
respectively. HadCRUT4, unlike NOAAGlobalTemp, does not use extrapolation or interpolation to
provide values for 5Ox5O grid squares for which it cannot produce a value due to lack of direct
observations of temperature. Absence of values in a region does not necessarily mean that
measurements of air or sea temperatures are currently lacking for the region, as a value may not be
plotted because a climatological value cannot be established from past measurements.
The Arctic void in data from HadCRUT4 and NOAAGlobalTemp covers only a small part of the globe,
but it is the region where the largest temperature deviations from climatology occur, associated
with anomalous winter sea-ice conditions. As noted above, 2012 was a year with unusually warm
mean Arctic temperatures. This is shown clearly by the maps for the reanalyses in Figure 3. The
warm temperatures are supported by HadCRUT4 values where available. NOAAGlobalTemp provides
fewer data values in the region in question, and where data are provided the values are less
anomalous.
Elsewhere the patterns and amplitudes of the temperature anomalies shown in Figure 3 are in
generally good agreement where observational coverage is good. ERA-Interim and JRA-55 differ
most over western and southern Africa, over South America (where it is ERA-Interim that is the more
consistent with HadCRUT4 and NOAAGlobalTemp) and over Antarctica. HadCRUT4 and
NOAAGlobalTemp do not provide values over sea-ice off the coast of Antarctica, where ERA-Interim
and JRA-55 indicate below-average temperatures associated with a higher than usual concentration
of sea ice. This is especially so in 2014; the quite close agreement among all four datasets as regards
global average temperatures for this year appears to result from cancellation of above-average
temperatures in the Arctic and below-average temperatures in the Antarctic.
Differences from monthly climatological averages for 1981-2010 are compared in Figure 4 for
January and July 2015. Similarly to what was done for Figure 3, values for HadCRUT4 and
NOAAGlobalTemp are plotted for a particular month only for grid squares for which there are at
least 28 values available to define the month’s climatological average for 1981-2010. The four
datasets are in overall agreement for both months, although local differences can be seen. For
example, HadCRUT4 has two anomalously cold grid squares over the Mozambique Channel in July
September 2015
Copernicus Climate Change Service, ECMWF
2015 that are not seen in any of the other datasets; this suggests a quality-control issue that may be
remedied in a future version of the product. JRA-55 has a pronounced cold anomaly over western
Africa in July 2015, and generally has a less-warm anomaly than ERA-Interim over Africa and South
America for this month. In contrast, ERA-Interim has a larger cold anomaly over the Antarctic
Plateau. NOAAGlobalTemp produces a more complete coverage than HadCRUT4, but smooths out
detail on which there is agreement between the other datasets.
In summary, these comparisons indicate that ERA-Interim is a reasonable choice of product for
providing a regularly updated monitoring of global temperature, notwithstanding reservations for
regions with a sparse long-term record of direct measurements, notably over the Antarctic
continent.
Figure 1 Running twelve-month averages of global-mean surface temperature anomalies (OC) relative to 19812010, from ERA-Interim (top), JRA-55 (upper middle), HadCRUT4 (version 4.0.0; lower middle) and
NOAAGlobalTemp (version 4.0.0.201507; bottom), from January 1979 to July 2015. The darker coloured bars
are the averages for each of the calendar years from 1979 to 2014. The dotted line show the average of the
four datasets for 2014.
September 2015
Copernicus Climate Change Service, ECMWF
Figure 2 As Figure 1, but for averages over European land areas.
September 2015
Copernicus Climate Change Service, ECMWF
Figure 3 Surface temperature anomalies (OC) relative to 1981-2010 for 2012 (left) and 2014 (right), from ERAInterim (top), JRA-55 (upper middle), HadCRUT4 (version 4.0.0; lower middle) and NOAAGlobalTemp (version
4.0.0.201507; bottom).
September 2015
Copernicus Climate Change Service, ECMWF
Figure 4 Surface temperature anomalies (OC) relative to 1981-2010 for January 2015 (left) and July 2015 (right),
from ERA-Interim (top), JRA-55 (upper middle), HadCRUT4 (version 4.0.0; lower middle) and NOAAGlobalTemp
(version 4.0.0.201507; bottom).
September 2015