Frequently Asked Questions

Q: What anomaly period have you used?

A: Anomalies have been calculated relative to the 1961-1990 average. The method used to calculate the land station normals is described in Jones et al. 2012. The method used to derive the SST climatology is described in Rayner et al. 2006.

Q: Where do the observations come from?

A: The land observations are from an archive of land temperature station data that is maintained by the Climatic Research Unit at the University of East Anglia. Individual stations files are available from the CRUTEM4 data page. The CRU archive comprises data from several sources and a more complete description is given in the CRUTEM4 paper. Updates to data sources included in the CRU archive since the release of the CRUTEM4 paper are described on the CRUTEM4 version updates page.

The SST observations come from version 2.5 of the International Comprehensive Ocean Atmosphere Data Set (ICOADS). Updates from 2007 to present are taken from the Global Telecommunication System (GTS).

Q: What are the uncertainties and what implications do they have for the records?

A: The accuracy with which we can calculate a global average varies from year to year depending on many factors. The new version of HadCRUT provides much more detailed information enabling anyone to see what the uncertainties are and whether they affect their analysis.

For example, we don't know the exact date when canvas buckets were phased out in favour of modern rubber buckets for sea surface temperature observations on ships, but we do know approximately. How quickly they were phased out could affect our estimates of global temperature trends, but by trying a variety of different timelines for the changeover we can assess how large that effect would be. Rather than choosing a precise start and end date for the switch we produce a range of data sets with different start and end dates applied. For the first time, we are providing these individual versions of HadCRUT to users. We use the different versions ourselves to calculate a likely range of global average temperatures.

For more details see the HadCRUT4 paper.

Q: Was 2010 (or 1998 or 2005) the warmest year on record?

A: The short answer is, maybe. It is not possible to calculate the global average temperature anomaly with perfect accuracy because the underlying data contain measurement errors and because the measurements do not cover the whole globe. However, it is possible to quantify the accuracy with which we can measure the global temperature and that forms an important part of the creation of the HadCRUT4 data set. The accuracy with which we can measure the global average temperature of 2010 is around one tenth of a degree Celsius. The difference between the median estimates for 1998 and 2010 is around one hundredth of a degree, which is much less than the accuracy with which either value can be calculated. This means that we can't know for certain - based on this information alone - which was warmer. However, the difference between 2010 and 1989 is around four tenths of a degree, so we can say with a good deal of confidence that 2010 was warmer than 1989, or indeed any year prior to 1996.

HadCRUT4 has more comprehensive data. It includes an increased number of land stations, as well as more up-to-date records available from land stations that were already in HadCRUT3. These updates to the land station database are described in the CRUTEM4 paper. HadCRUT4 also includes a larger number of sea-surface temperature records and new bias adjustments for sea surface temperature measurements. These bias adjustments for sea-surface temperature records are discussed in detail in part 2 of the HadSST3 paper.

There has also been a change in the way that HadCRUT4 is presented to users. In order to preserve information about the temporal and spatial structure of uncertainties in the data, HadCRUT4 is presented as an ensemble data set in which the 100 ensemble members sample the distribution of observational uncertainty described by the underlying bias model. In addition fields of measurement and sampling errors are supplied for which the correlation structure is sufficiently simple to be encoded in a covariance matrix. For more details please read the HadCRUT4 paper.

A: Errors in climate data sets come from various sources. Some, like transcription errors, affect a single reading. Others, like the calibration error on a single ship's thermometer affect measurements over a much wider area, but are unlikely to affect estimates of long-term trends spanning decades. These types of errors are handled in the traditional way and are presented as a set of error-covariance matrices that encode this information.

Systematic errors such as those associated with station moves, or widespread changes in instrumentation have complex spatial and temporal structures that cannot be concisely summarised. In order to preserve information about the temporal and spatial structure of these uncertainties, HadCRUT4 is presented as an ensemble data set in which the 100 ensemble members sample the distribution of observational uncertainty described by the underlying bias model. For more details please read the HadCRUT4 and HadSST3 papers.

A: When we calculate the trend in the global average temperature, we calculate the global annual average and the trend for each of the 100 realisations. The distribution of the 100 values gives an idea of the uncertainty in that derived quantity that arises from bias-like errors. Additional uncertainties associated with more localised measurement and sampling errors are handled differently. For more detail see Section 5 of the HadCRUT4 paper

Likewise, it should be possible to run your analysis on each of the realisations and the spread in results gives an idea of the sensitivity of the analysis to bias-like errors in observations. A separate file contains the measurement and sampling uncertainties that are not strongly correlated over long time periods.

If you are interested in exploring the observational uncertainty, we recommend that you use a variety of data sets in addition to HadCRUT4. In the early record and in data sparse regions, the differences in the treatment of missing data, random measurement errors and quality control become more significant components of the overall uncertainty.

A: The data for the land and sea components are available from the CRUTEM4 and HadSST3 web pages.

A: To compensate for different observational coverage of the northern and southern hemispheres, global time series are calculated as the average of northern and southern hemispheric averages, computed as (NH+SH)/2. This prevents the better observed northern hemisphere from dominating the global average. Annual averages are computed by first forming monthly time series and then averaging the monthly series to form the annual values. Best estimate series are computed as the median of the 100 HadCRUT4 ensemble member time series.

Further details on the computation of HadCRUT4 time series and uncertainties in these series can be found in the HadCRUT4 paper

Decadally smoothed annual time series are formed by applying a 21-point binomial filter to annual time series. The filter is a weighted moving average of the data. Its weights are centred on the year of interest.

Toward the ends of the series there are not enough points to calculate the smoothed value. For example, to calculate the smoothed value for 2005 we would need to know what the annual averages were for the 21-year period 1995-2015, but we only currently have annual data for the period 1995-2010. Ideally the smoothing should stop before the filter 'runs off' the end of the series, but a series that has been shortened in this way appears not to be up-to-date.

In order to extend the simple smoothing to the very ends of the time series it is necessary to either extend the data series, or shorten the filter. Howsoever it is done, the data near the endpoints will be treated differently to data in the middle of the series. Extending the data series can be done in a number of ways, but the method used for HadCRUT4 is simply to continue the series by repeating the final value.

There is an increase in uncertainty associated with estimation of decadally smoothed anomaly values where fewer than 21 years of data are available toward the ends of series. The uncertainty in smoothed anomaly values arising from application of the method for extending the smoothed series is estimated as part of the coverage uncertainty calculation, resulting in increased uncertainty ranges in decadally smoothed HadCRUT4 anomaly values toward the ends of the smoothed series.

A: Data set development is an ongoing project. There remains a lack of data for some regions of the world recently - most notably the Arctic and Antarctic regions - but also for much larger regions in the early record. If and when new sources of data become available, these may be included in HadCRUT but it is not possible to say how this may affect the dataset in future. We will continue to update the dataset using the latest data over the Global Telecommunication System (GTS) and in Monthly Climatic Data for the World publications and will continue to incorporate the latest research findings into the dataset.

Sea-surface temperature observations will be updated using data taken from the GTS. Retrospective updates incorporating newly digitised data are also possible.

A. Diagrams are Crown Copyright. Source should be acknowledged as Met Office Hadley Centre.

Commercial and media enquiries

You can access the Met Office Customer Centre, any time of the day or night by phone, fax or e-mail. Trained staff will help you find the information or products that are right for you.
Contact the Met Office Customer Centre