Editing IPCC:AR6/WGI/Chapter-2 (section)

==== 2.3.1.1 Surface Temperatures ====

<div id="h3-12-siblings" class="h3-siblings"></div>

<div id="2.3.1.1.1" class="h4-container"></div>

<span id="temperatures-of-the-deep-past-65-ma-to-8-ka"></span>
===== 2.3.1.1.1 Temperatures of the deep past (65 Ma to 8 ka) =====

<div id="h4-7-siblings" class="h4-siblings"></div>

This assessment of the paleo reference periods (Cross-Chapter Box 2.1) draws from studies based mostly or entirely on indirect observational evidence from geological archives (i.e., proxy records) rather than reconstructions that rely more heavily on modelled parameters and those based on deep-ocean temperatures (e.g., [[#Köhler--2015|Köhler et al., 2015]] ; [[#Friedrich--2016|Friedrich et al., 2016]] ). In contrast to AR5, temperature estimates from climate models are not included in the assessed values for paleo reference periods in this chapter. The AR5 concluded that the reconstructed GMST during the PETM was 4°C–7°C warmer than pre-PETM mean climate ( ''low confidence'' ), and that the EECO and the MPWP were 9°C–14°C and 1.9°C–3.6°C warmer than pre-industrial, respectively ( ''medium confidence'' ). The GMST during the LIG was assessed at 1°C–2°C warmer than pre-industrial ''(medium confidence'' ), whereas SROCC narrowed the range to 0.5°C–1.0°C warmer, but did not state a confidence level. The AR5 further concluded that it was ''very'' ''likely'' that the LGM was 3°C–8°C colder than pre-industrial, and ''likely'' that the maximum rate of global warming during the subsequent deglacial period was 1°C–1.5°C kyr <sup>–1</sup> .

For the PETM, new reconstructions agree with those assessed by AR5. A major new compilation of proxy temperature data ( [[#Hollis--2019|Hollis et al., 2019]] ) analysed using multiple statistical approaches ( [[#Inglis--2020|Inglis et al., 2020]] ) indicates that GMST was 10°C–25°C (90% range) warmer than 1850–1900, or about 5°C warmer relative to the pre-PETM state. A related synthesis study also estimates that PETM warmed by 5°C (no uncertainty assigned; [[#Zhu--2019|Zhu et al., 2019]] ). A recent benthic isotope compilation ( [[#Westerhold--2020|Westerhold et al., 2020]] ) transformed to GMST based on the formulation by J. Hansen et al. (2013; Cross-Chapter Box 2.1, Figure 1), and adjusted to 1850–1900 by adding 0.36°C, shows an increase of GMST by about 10°C during the PETM. This reflects the expected higher variability at single sites that were used to splice together the composite time series, compared to the globally averaged composite time series of [[#Zachos--2008|Zachos et al. (2008)]] . The latter was originally used by J. [[#Hansen--2013|]] [[#Hansen--2013|Hansen et al. (2013)]] to reconstruct GMST, and is the preferred representation of the global average bottom water conditions, despite its less well-refined chronology.

For the EECO, new GMST reconstructions fall at the high end of the range assessed by AR5. These include estimates of 7°C–18°C (90% range; [[#Inglis--2020|Inglis et al., 2020]] ) and 12°C–18°C (95% range; [[#Zhu--2019|Zhu et al., 2019]] ) warmer than 1850–1900, and 10°C–16°C warmer than 1995–2014 ‘recent past’ conditions (2 standard error range; [[#Caballero--2013|Caballero and Huber, 2013]] ). Together, they indicate that GMST was 10°C–18°C warmer during the EECO compared with 1850–1900 ( ''medium confidence'' ).

The AR5 did not assess the GMST for the MCO. Reconstructions based on data from multiple study sites include estimates of about 4°C (uncertainty range not specified; [[#You--2009|You et al., 2009]] ) and 5°C–10°C (2 standard error range; [[#Goldner--2014|Goldner et al., 2014]] ) warmer than 1850–1900. Together, these studies indicate that GMST was 4°C–10°C warmer during the MCO ( ''medium confidence'' ).

For the MPWP, new proxy-based estimates of global sea surface temperatures (SST) are about 2.0°C–3.5°C warmer than 1850–1900, depending on which proxy types are included in the analysis ( [[#Foley--2019|Foley and Dowsett, 2019]] ; [[#McClymont--2020|McClymont et al., 2020]] ). On the basis of model-derived relationships between land versus sea surface temperatures under different climate states (Figure 3.2b), the increase in GMST is estimated to have been roughly 15% greater than the increase in global SST. Therefore, GMST during the MPWP is estimated to have been 2.5°C–4.0°C warmer than 1850–1900 ( ''medium confidenc'' e).

For the LIG (Cross-Chapter Box 2.1, Figure 1, and Figure 2.11), a major new compilation of marine proxy data ( [[#Turney--2020|Turney et al., 2020]] ) from 203 sites indicates that the average SST from 129–125 ka was 1.0°C ± 0.2°C (2 SD) warmer than 1850–1900 (reported relative to 1981–2010 and adjusted here by 0.8°C). These temperatures represent the time of peak warmth, which may not have been synchronous among these sites. This compares with two other SST estimates for 125 ka of 0.5°C ± 0.3°C (± 2 SD) warmer at 125 ka relative to 1870–1889 ( [[#Hoffman--2017|Hoffman et al., 2017]] ), and about 1.4°C (no uncertainty stated) warmer at 125 ka relative to 1850–1900 ( [[#Friedrich--2020|Friedrich and Timmermann, 2020]] ; reported relative to 10–5 ka and adjusted here by 0.4°C; [[#Kaufman--2020a|Kaufman et al., 2020a]] ). The average of these post-AR5 global SST anomalies is 1°C. Commensurately (Figure 3.2b), GMST is estimated to have been roughly 1.1°C above 1850–1900 values, although this value could be too high if peak warmth was not globally synchronous ( [[#Capron--2017|Capron et al., 2017]] ). A further estimate of peak GMST anomalies of 1.0°C–3.5°C (90% range; adjusted here to 1850–1900 by adding 0.2°C) based on 59 marine sediment cores ( [[#Snyder--2016|Snyder, 2016]] ) is considerably warmer than remaining estimates and is therefore given less weight in the final assessment. The warmest millennium of the LIG GMST reconstruction in J. [[#Hansen--2013|]] [[#Hansen--2013|Hansen et al. (2013)]] is 1.5°C above 1850–1900. In summary, GMST during the warmest millennia of the LIG (within the interval of around 129–125 ka) is estimated to have reached 0.5°C–1.5°C higher values than the 1850–1990 reference period ( ''medium confidence'' ).

<div id="_idContainer030" class="Basic-Text-Frame"></div>

[[File:db18d2c58474b72f5157a77d07c68327 IPCC_AR6_WGI_Figure_2_11.png]]

'''Figure''' '''2.11 |''' '''Earth’s surface temperature history with key findings annotated within each panel. (a)''' GMST over the Holocene divided into three time scales: (i) 12 kyr–1 kyr in 100-year time steps; (ii) 1000–1900 CE, 10-year smooth; and (iii) 1900–2020 CE (from panel (c)). Median of the multi-method reconstruction (bold lines), with 5th and 95th percentiles of the ensemble members (thin lines). Vertical bars are the assessed ''medium confidence'' ranges of GMST for the Last Interglacial and mid-Holocene ( [[#2.3.1.1|Section 2.3.1.1]] ). The last decade value and ''very likely'' range arises from [[#2.3.1.1.3|Section 2.3.1.1.3]] . '''(b)''' Spatially resolved trends (°C per decade) for HadCRUTv5 over (upper map) 1900–1980, and (lower map) 1981–2020. Significance is assessed following AR(1) adjustment after [[#Santer--2008|Santer et al. (2008)]] , ‘×’ marks denote non-significant trends. '''(c)''' Temperature from instrumental data for 1850–2020, including (upper panel) multi-product mean annual time series assessed in [[#2.3.1.1.3|Section 2.3.1.1.3]] for temperature over the oceans (blue line) and temperature over the land (red line) and indicating the warming to the most recent 10 years; and annually (middle panel) and decadally (bottom panel) resolved averages for the GMST datasets assessed in [[#2.3.1.1.3|Section 2.3.1.1.3]] . The grey shading in each panel shows the uncertainty associated with the HadCRUT5 estimate ( [[#Morice--2021|Morice et al., 2021]] ). All temperatures relative to the 1850–1900 reference period. Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

New GMST reconstructions for the LGM fall near the middle of AR5’s ''very likely'' range, which was based on a combination of proxy reconstructions and model simulations. Two of these new reconstructions use marine proxies to reconstruct global SST that were scaled to GMST based on different assumptions. One indicates that GMST was 6.2 [4.5 to 8.1] °C cooler than the late Holocene average ( [[#Snyder--2016|Snyder, 2016]] ), and the other, 5.7°C ± 0.8°C (2 SD) cooler than the average of the first part of the Holocene (10–5 ka) ( [[#Friedrich--2020|Friedrich and Timmermann, 2020]] ). A third new estimate ( [[#Tierney--2020|Tierney et al., 2020]] ) uses a much larger compilation of marine proxies along with a data-assimilation procedure, rather than scaling, to reconstruct a GMST of 6.1°C ± 0.4°C (2 SD) cooler than the late Holocene. Assuming that the 1850–1900 reference period was 0.2°C and 0.4°C cooler than the late and first part of the Holocene, respectively ( [[#Kaufman--2020a|Kaufman et al., 2020a]] ), the midpoints of these three new GMST reconstructions average –5.8°C relative to 1850–1900. The coldest multi-century period of the LGM in the J. [[#Hansen--2013|]] [[#Hansen--2013|Hansen et al. (2013)]] reconstruction is 4.3°C colder than 1850–1900. This compares to land- and SST-only estimates of about –6.1°C ± 2°C and –2.2°C ± 1°C, respectively (2 SD), which are based on AR5-generation studies that imply a warmer GMST than more recent reconstructions (Figure 1c in [[#Harrison--2015|Harrison et al., 2015]] ; Figure 7 in [[#Harrison--2016|Harrison et al., 2016]] ). A major new pollen-based data-assimilation reconstruction averages 6.9°C cooler over northern extratropical land ( [[#Cleator--2020|Cleator et al., 2020]] ). LGM temperature variability on centennial scales was about four times higher globally than during the Holocene, and even greater at high latitudes ( [[#Rehfeld--2018|Rehfeld et al., 2018]] ). In summary, GMST is estimated to have been 5°C–7°C lower during the LGM (around 23–19 ka) compared with 1850–1900 ( ''medium confidence'' ).

For the LDT (Cross-Chapter Box 2.1, Figure 1), no new large-scale studies have been published since AR5 ( [[#Shakun--2012|Shakun et al., 2012]] ) to further assess the rate of GMST change during this period of rapid global warming (estimated at 1°C–1.5°C per kyr). The reconstruction of [[#Shakun--2012|Shakun et al. (2012)]] was based primarily on SST records and therefore underrepresents the change in GMST during the LDT. Temperature over Greenland increased by about ten times that rate during the centuries of most rapid warming ( [[#Jansen--2020|Jansen et al., 2020]] ).

<div id="2.3.1.1.2" class="h4-container"></div>

<span id="temperatures-of-the-post-glacial-period-past-7000-years"></span>
===== 2.3.1.1.2 Temperatures of the post-glacial period (past 7000 years) =====

<div id="h4-8-siblings" class="h4-siblings"></div>

The AR5 did not include an assessment of large-scale temperature estimates for the MH, although it assigned ''high confidence'' to the long-term cooling trend over mid- to high-latitudes of the Northern Hemisphere (NH) during the 5 kyr that preceded recent warming. For average annual NH temperatures, the period 1983–2012 was assessed as ''very likely'' the warmest 30-year period of the past 800 years ( ''high confidence'' ) and ''likely'' the warmest 30-year period of the past 1.4 kyr ( ''medium confidence'' ); the warm multi-decadal periods prior to the 20th century were unsynchronized across regions, in contrast to the warming since the mid-20th century ( ''high confidence'' ), although only sparse information was available from the SH.

This section concerns the Holocene period prior to industrialization when GMST was overall highest. Whereas SR1.5 focussed upon the ‘Holocene thermal maximum’ when regional temperatures were up to 1°C higher than 1850–1900, though peak warming occurred regionally at different times between around 10 and 5 ka greatly complicating interpretation. A multi-method reconstruction ( [[#Kaufman--2020a|Kaufman et al., 2020a]] ) based on a quality-controlled, multi-proxy synthesis of paleo-temperature records from 470 terrestrial and 209 marine sites globally ( [[#Kaufman--2020b|Kaufman et al., 2020b]] ) indicates that the median GMST of the warmest two-century-long interval was 0.7 [0.3 to 1.8] °C warmer than 1800–1900 (which averaged 0.03°C colder than 1850–1900; [[#PAGES%202k%20Consortium--2019|PAGES 2k Consortium, 2019]] ), and was centred around 6.5 ka. This is similar to [[#Marcott--2013|Marcott et al. (2013)]] , which is based on a smaller dataset (73 sites) and different procedures to estimate a maximum warmth of 0.8°C ± 0.3°C (2 SD) at around 7.0 ka, adjusted here by adding 0.3°C to account for differences in reference periods. These may be underestimates because averaging inherently smoothed proxy records with uncertain chronologies reduces the variability in the temperature reconstruction (e.g., Dolman and Laepple, (2018) for sedimentary archives). However, the general coincidence between peak warmth and astronomically driven boreal summer insolation might reflect a bias toward summer conditions ( [[#Liu--2014|Liu et al., 2014]] ; [[#Hou--2019|Hou et al., 2019]] ; [[#Bova--2021|Bova et al., 2021]] ), suggesting that the estimate is too high. This possibility is supported by AR5-generation proxy data focusing on 6 ka ( [[#Harrison--2014|Harrison et al., 2014]] ), the long-standing MH modelling target (Cross-Chapter Box 2.1), that indicate surface temperatures for land and ocean were indistinguishable from ‘pre-industrial’ climate (Figure 1c in [[#Harrison--2015|Harrison et al., 2015]] ; Figure 7 in [[#Harrison--2016|Harrison et al., 2016]] ). In contrast, the GMST estimate from the multi-method global reconstruction ( [[#Kaufman--2020a|Kaufman et al., 2020a]] ) for the millennium centred on 6 ka is only about 0.1°C colder than the warmest millennium.

Taking all lines of evidence into account, the GMST averaged over the warmest centuries of the current interglacial period (sometime between around 6 and 7 ka) is estimated to have been 0.2°C–1.0°C higher than 1850–1900 ( ''medium confidence'' ). It is therefore ''more likely than not'' that no multi-centennial interval during the post-glacial period was warmer globally than the most recent decade (which was 1.1°C warmer than 1850–1900; [[#2.3.1.1.3|Section 2.3.1.1.3]] ); the LIG (129–116 ka) is the next most recent candidate for a period of higher global temperature. Zonally averaged mean annual temperature reconstructions ( [[#Routson--2019|Routson et al., 2019]] ) indicate that MH warmth was most pronounced north of 30°N latitude, and that GMST subsequently decreased in general, albeit with multi-century variability, with greater cooling in the NH than in the SH ( [[#Kaufman--2020a|Kaufman et al., 2020a]] ).

The temperature history of the last millennium and the methods used to reconstruct it have been studied extensively, both prior to and following AR5, as summarized recently by [[#Smerdon--2016|Smerdon and Pollack (2016)]] and [[#Christiansen--2017|Christiansen and Ljungqvist (2017)]] . New regional (e.g., [[#Shi--2015|Shi et al., 2015]] ; [[#Stenni--2017|Stenni et al., 2017]] ; [[#Werner--2018|Werner et al., 2018]] ), global ocean ( [[#McGregor--2015|McGregor et al., 2015]] ), quasi-hemispheric ( [[#Neukom--2014|Neukom et al., 2014]] ; [[#Schneider--2015|Schneider et al., 2015]] ; [[#Anchukaitis--2017|Anchukaitis et al., 2017]] ), and global ( [[#Tardif--2019|Tardif et al., 2019]] ) temperature reconstructions, and new regional proxy data syntheses ( [[#Lüning--2019a|Lüning et al., 2019a]] , b) have been published, extending back 1–2 kyr. In addition, a major new global compilation of multiproxy, annually resolved paleo-temperature records for the CE ( [[#PAGES%202k%20Consortium--2017|PAGES 2k Consortium, 2017]] ) has been analysed using a variety of statistical methods for reconstructing temperature ( [[#PAGES%202k%20Consortium--2019|PAGES 2k Consortium, 2019]] ). The median of the multi-method GMST reconstruction from this synthesis (Figure 2.11a) generally agrees with the AR5 assessment, while affording more robust estimates of the following major features of GMST during the CE: (i) an overall millennial-scale cooling trend of –0.18 [–0.28 to 0.00] °C kyr <sup>–1</sup> prior to 1850; (ii) a multi-centennial period of relatively low temperature beginning around the 15th century, with GMST averaging –0.03 [–0.30 to 0.06] °C between 1450 and 1850 relative to 1850–1900; (iii) the warmest multi-decadal period occurring most recently; and (iv) the rate of warming during the second half of the 20th century (from instrumental data) exceeding the 99th percentile of all 51-year trends over the past 2 kyr. Moreover, the new proxy data compilation shows that the warming of the 20th century was more spatially uniform than any other century-scale temperature change of the CE ( ''medium confidence'' ) ( [[#Neukom--2019|Neukom et al., 2019]] ). A new independent temperature reconstruction extending back to 1580 is based on an expanded database of subsurface borehole temperature profiles, along with refined methods for inverse modelling ( [[#Cuesta-Valero--2021|Cuesta-Valero et al., 2021]] ). The borehole data, converted to GMST based on the modelled relation between changes in land versus sea surface temperature outlined previously, indicate that average GMST for 1600–1650 was 0.12°C colder than 1850–1900, which is similar to the PAGES 2k reconstruction (0.09°C colder), although both estimates are associated with relatively large uncertainties (0.8°C (95% range) and 0.5°C (90% range), respectively).

To conclude, following approximately 6 ka, GMST generally decreased, culminating in the coldest multi-century interval of the post-glacial period (since 8 ka), which occurred between around 1450 and 1850 ( ''high confidence'' ). This multi-millennial cooling trend was reversed in the mid-19th century. Since around 1950, GMST has increased at an observed rate unprecedented for any 50-year period in at least the last 2000 years ( ''high confidence'' ).

<div id="2.3.1.1.3" class="h4-container"></div>

<span id="temperatures-during-the-instrumental-period-surface"></span>
===== 2.3.1.1.3 Temperatures during the instrumental period – surface =====

<div id="h4-9-siblings" class="h4-siblings"></div>

The AR5 concluded that it was certain that GMST had increased since the late 19th century. Total warming in GMST was assessed as 0.85 [0.65 to 1.06] °C over 1880–2012, while the change from 1850–1900 to 2003–2012 was assessed at 0.78 [0.72 to 0.85] °C, and from 1850–1900 to 1986–2005 at 0.61 [0.55 to 0.67] °C. The SR1.5 reported warming of GMST from 1850–1900 to 2006–2015 of 0.87°C, with an 1880–2012 trend of 0.86°C and an 1880–2015 trend of 0.92°C. The SRCCL concluded that since the pre-industrial period, surface air temperature over land areas has risen nearly twice as much as the global mean surface temperature ( ''high confidence'' ).

Since AR5, there have been substantial improvements in the availability of instrumental archive data both over the ocean and on land. A new version of the International Comprehensive Ocean-Atmosphere Dataset (ICOADS Release 3.0, [[#Freeman--2017|Freeman et al., 2017]] ) comprises over 450 million in situ marine reports and incorporates newly digitized data, increasing coverage in data sparse regions and times (e.g., polar oceans and World War I). The International Surface Temperature Initiative released a much improved collection of fundamental land surface air temperature records ( [[#Rennie--2014|Rennie et al., 2014]] ) comprising more than 35,000 station records. These advances, both of which have substantially improved spatial coverage, have reduced uncertainties in assessments of both land and marine data.

<span id="marine-domain"></span>
====== Marine domain ======

For SST analyses, three products – HadSST4 (1850–present, [[#Kennedy--2019|Kennedy et al., 2019]] ), ERSSTv5 (1850–present, [[#Huang--2017|Huang et al., 2017]] ) and COBE SST2 (1880–present, ( [[#Hirahara--2014|Hirahara et al., 2014]] ) – now have bias adjustments applied throughout the record. The new SST datasets account for two major issues previously identified in AR5: that globally averaged buoy SSTs are about 0.12°C cooler than ship-based SSTs ( [[#Kennedy--2011|Kennedy et al., 2011]] ; [[#Huang--2015|Huang et al., 2015]] ), and that SSTs from ship engine room intakes may have biases for individual ships depending upon the sensor set-up ( [[#Kent--2006|Kent and Kaplan, 2006]] ) but have an overall warm bias when globally aggregated ( [[#Kennedy--2019|Kennedy et al., 2019]] ). The first issue primarily affects data since 1990, when buoys began to increasingly contribute to the observation network ( [[#Woodruff--2011|Woodruff et al., 2011]] ), and the second issue has its largest effect from the 1940s to the 1970s. From the standpoint of uncertainty, ERSSTv4 (W. [[#Liu--2015|]] [[#Liu--2015|Liu et al., 2015]] ; [[#Huang--2016|Huang et al., 2016]] ) and subsequent versions ( [[#Huang--2017|Huang et al., 2017]] ), and HadSST4 have estimates presented as ensembles that sample parametric uncertainty. Comparisons between these independently-derived analyses and the assessed uncertainties ( [[#Kennedy--2014|Kennedy, 2014]] ; [[#Kent--2017|Kent et al., 2017]] ) show unambiguously that global mean SST increased since the start of the 20th century, a conclusion that is insensitive to the method used to treat gaps in data coverage ( [[#Kennedy--2014|Kennedy, 2014]] ).

A number of recent studies also corroborate important components of the SST record ( [[#Hausfather--2017|Hausfather et al., 2017]] ; [[#Kent--2017|Kent et al., 2017]] ; [[#Cowtan--2018|Cowtan et al., 2018]] ; [[#Kennedy--2019|Kennedy et al., 2019]] ). In particular, ATSR SST satellite retrievals ( [[#Merchant--2012|Merchant et al., 2012]] ; [[#Berry--2018|Berry et al., 2018]] ), the near-surface records from hydrographical profiles ( [[#Gouretski--2012|Gouretski et al., 2012]] ; [[#Huang--2018|Huang et al., 2018]] ), and coastal observations ( [[#Cowtan--2018|Cowtan et al., 2018]] ) have all been shown to be broadly consistent with the homogenized SST analyses. [[#Hausfather--2017|Hausfather et al. (2017)]] also confirmed the new estimate of the rate of warming seen in ERSSTv4 since the late 1990s through comparison with independent SST data sources such as Argo floats and satellite retrievals. Nevertheless, dataset differences remain in the mid-20th century when there were major, poorly-documented, changes in instrumentation and observational practices ( [[#Kent--2017|Kent et al., 2017]] ), particularly during World War II, when ship observations were limited and disproportionately originated from US naval sources ( [[#Thompson--2008|Thompson et al., 2008]] ). [[#Kennedy--2019|Kennedy et al. (2019)]] also identify differences between the new HadSST4 dataset and other SST datasets in the 1980s and 1990s, indicating that some level of structural uncertainty remains during this period, whilst [[#Chan--2019|Chan et al. (2019)]] and [[#Davis--2019|Davis et al. (2019)]] document residual uncertainties in the early and later 20th century records respectively.

Historically, SST has been used as a basis for global temperature assessment on the premise that the less variable SST data provides a better estimate of marine temperature changes than marine air temperature (MAT) ( [[#Kent--2021|Kent and Kennedy, 2021]] ). However, MAT products are used to adjust SST biases in the NOAA SST product because they are assessed to be more homogeneous ( [[#Huang--2017|Huang et al., 2017]] ). Observational datasets exist for night-marine air temperature (NMAT) (e.g., [[#Cornes--2020|Cornes et al., 2020]] ; [[#Junod--2020|Junod and Christy, 2020]] ; [[#Rayner--2020|Rayner et al., 2020]] ) and there are methods to adjust daytime MATs ( [[#Berry--2004|Berry et al., 2004]] ), but there is to date no regularly updated dataset which combines MAT with temperatures over land. MAT datasets are more sparse in recent decades than SST datasets as marine datasets have become increasingly dependent on drifting buoys ( [[#Centurioni--2019|Centurioni et al., 2019]] ) which generally measure SST but not MAT, and there are almost no recent winter MAT data south of 40°S ( [[#Swart--2019|Swart et al., 2019]] ). However, the situation reverses in the 19th century with a greater prevalence of MAT than SST measurements available in the ICOADS data repository ( [[#Freeman--2017|Freeman et al., 2017]] , 2019; [[#Kent--2021|Kent and Kennedy, 2021]] ).

<span id="land-domain"></span>
====== Land domain ======

The GHCNMv4 dataset ( [[#Menne--2018|Menne et al., 2018]] ) includes many more land stations than GHCNMv3, arising from the databank efforts of [[#Rennie--2014|Rennie et al. (2014)]] , and calculates a 100-member parametric uncertainty ensemble drawing upon the benchmarking analysis of [[#Williams--2012|Williams et al. (2012)]] , as well as accounting for sampling effects. A new version of the CRUTEM dataset (CRUTEMv5, [[#Osborn--2021|Osborn et al., 2021]] ) has increased data completeness and additional quality control measures. A new global land dataset, the China Land Surface Air Temperature (CLSAT) dataset ( [[#Xu--2018|Xu et al., 2018]] ) has higher network density in some regions (particularly Asia) than previously existing datasets. Global trends derived from CLSAT are generally consistent with those derived from other land datasets through 2014 ( [[#Xu--2018|Xu et al., 2018]] ).

The AR5 identified diurnal temperature range (DTR) as a substantial knowledge gap. The most recent analysis of Thorne et al. (2016a, b) compared a broad range of gridded estimates of change in DTR, including a new estimate derived from the ISTI databank release using the pairwise homogenization algorithm used to create GHCNMv4, and estimates derived from [[#Vose--2005|Vose et al. (2005)]] , HadEX2 ( [[#Donat--2013a|Donat et al., 2013a]] ), HadGHCND ( [[#Donat--2013b|Donat et al., 2013b]] ), GHCNDEX ( [[#Donat--2013b|Donat et al., 2013b]] ), Berkeley Earth ( [[#Rohde--2013|Rohde et al., 2013]] ), and CRU TS ( [[#Harris--2014|Harris et al., 2014]] ). The analysis highlighted substantial ambiguity in pre-1950 estimates arising from sparse data availability. After 1950 estimates agreed that DTR had decreased globally with most of that decrease occurring over the period 1960–1980. A subsequent DTR analysis using CLSAT further confirmed this behaviour (X. [[#Sun--2018|]] [[#Sun--2018|Sun et al., 2018]] ).

No recent literature has emerged to alter the AR5 finding that it is ''unlikely'' that any uncorrected effects from urbanization (Box 10.3), or from changes in land use or land cover ( [[#2.2.7|Section 2.2.7]] ), have raised global Land Surface Air Temperature (LSAT) trends by more than 10%, although larger signals have been identified in some specific regions, especially rapidly urbanizing areas such as eastern China (Y. [[#Li--2013|]] [[#Li--2013|Li et al., 2013]] ; [[#Liao--2017|Liao et al., 2017]] ; Z. [[#Shi--2019|]] [[#Shi--2019|Shi et al., 2019]] ). There is also no clear indication that site-specific data homogeneity issues have had any significant impact on global trends since the early 20th century; there is more uncertainty in the 19th century, mainly arising from a lack of standardization of instrument shelters, which has been largely accounted for in data from central Europe ( [[#Jones--2012|Jones et al., 2012]] ), but less so elsewhere.

<span id="combined-data-products"></span>
====== Combined data products ======

At the time of AR5 a limitation of conventional datasets was the lack of coverage, especially in high latitudes, which although recognized as an issue ( [[#Simmons--2010|Simmons et al., 2010]] ) had not been addressed in most products. Interpolation involves the statistical imputation of values across regions with limited data and can add both systematic and random uncertainties ( [[#Lenssen--2019|Lenssen et al., 2019]] ). [[#Cowtan--2014|Cowtan and Way (2014)]] applied a kriging-based method to extend existing datasets to polar regions, while [[#Kadow--2020|Kadow et al. (2020)]] used an artificial intelligence-based method, and [[#Vaccaro--2021|Vaccaro et al. (2021)]] used gaussian random Markov fields, for the same purpose, although only [[#Kadow--2020|Kadow et al. (2020)]] uses the most recent generation of datasets as its base. The Berkeley Earth merged product ( [[#Rohde--2020|Rohde and Hausfather, 2020]] ), HadCRUT5 ( [[#Morice--2021|Morice et al., 2021]] ) and NOAA GlobalTemp-Interim ( [[#Vose--2021|Vose et al., 2021]] ) all include interpolation over reasonable distances across data sparse regions which results in quasi-global estimates from the late 1950s when continuous Antarctic observations commenced. Interpolated datasets with substantial coverage of high latitudes show generally stronger warming of GMST than those with limited data in polar regions ( [[#Vose--2021|Vose et al., 2021]] ), and their strong warming at high northern latitudes is consistent with independent estimates from reanalyses ( [[#Simmons--2017|Simmons et al., 2017]] ; [[#Lenssen--2019|Lenssen et al., 2019]] ) and satellites ( [[#Cowtan--2014|Cowtan and Way, 2014]] ). Given the spatial scales of surface temperature variations and the verification of the methods, it is ''extremely likely'' that interpolation results in a less-biased estimate of the actual global temperature change than ignoring regions with limited or no data.

In total there are five conventional datasets which meet spatial coverage requirements and draw from the most recent generation of SST analyses, four of which have sufficient data in the 1850–1900 period to allow an assessment of changes from that baseline (Table 2.3). A fifth dataset is added to the assessment for changes over land areas. Datasets share SST and LSAT data products and in several cases differ solely in the post-processing interpolation applied meaning that there are far fewer methodological degrees of freedom than implied by a straight count of the number of available estimates.

<div id="_idContainer033" class="Basic-Text-Frame"></div>

Table 2.3 | '''Principal characteristics of GMST in situ data products considered in AR6 WGI, highlighting interdependencies in underlying land and SST products and whether inclusion criteria are met.'''

{| class="wikitable"
|-
| '''Dataset'''

| '''Period of Record'''

| '''Land Component'''

| '''SST Component'''

| '''Ensemble Uncertainties?'''

| '''Meets all Inclusion Criteria?'''

| '''Principal Reference'''

|-
| '''HadCRUT5'''

| 1850–2020

| CRUTEM5

| HadSST4

| Yes

| Yes

| [[#Morice--2021|Morice et al. (2021)]]

|-
| '''NOAA GlobalTemp – Interim'''

| 1850–2020

| GHCNv4

| ERSSTv5

| Yes, on earlier version

| Yes

| [[#Vose--2021|Vose et al. (2021)]]

|-
| '''Berkeley Earth'''

| 1850–2020

| Berkeley

| HadSST4

| No

| Yes

| [[#Rohde--2020|Rohde and Hausfather (2020)]]

|-
| '''Kadow et al.'''

| 1850–2020

| CRUTEM5

| HadSST4

| No

| Yes

| [[#Kadow--2020|Kadow et al. (2020)]]

|-
| '''China – MST'''

| 1856–2020

| CLSAT

| ERSSTv5

| No

| Land only

| [[#Sun--2021|Sun et al. (2021)]]

|-
| '''GISTEMP'''

| 1880–2020

| GHCNv4

| ERSSTv5

| Yes

| Post-1880 only

| [[#Lenssen--2019|Lenssen et al. (2019)]]

|-
| '''Cowtan and Way'''

| 1850–2020

| CRUTEM4

| HadSST3

| Yes

| No

| [[#Cowtan--2014|Cowtan and Way (2014)]]

|-
| '''Vaccaro et al.'''

| 1850–2020

| CRUTEM4

| HadSST3

| No

| No

| [[#Vaccaro--2021|Vaccaro et al. (2021)]]

|}

Estimates of GMST have also benefitted from improved estimation of parametric uncertainties. New versions of three long-standing products from NASA GISTEMP v4 ( [[#Lenssen--2019|Lenssen et al., 2019]] ), NOAA GlobalTempv5 ( [[#Huang--2019b|]] [[#Huang--2019|B. Huang et al., 2019]] b ) and HadCRUT5 ( [[#Morice--2021|Morice et al., 2021]] ) are all now available as ensemble estimates. These ensembles each account for a variety of systematic and random uncertainty effects in slightly different ways, giving broadly similar results, which are incorporated into the present assessment, with the total uncertainty generally declining up until the mid-20th century as data coverage improves.

Another significant development has been the incorporation of reanalysis products ( [[IPCC:Wg1:Chapter:Chapter-1#1.5.2|Section 1.5.2]] ) into operational monitoring of GSAT. It was reported in AR5 that various reanalyses were broadly consistent with conventional surface datasets in the representation of trends since the mid-20th century. Since that time, [[#Simmons--2017|Simmons et al. (2017)]] found that the ERA-Interim ( [[#Dee--2011|Dee et al., 2011]] ) and JRA-55 ( [[#Kobayashi--2015|Kobayashi et al., 2015]] ) reanalyses continued to be consistent, over the last 20 years, with those surface datasets which fully represented the polar regions. GSAT trends from ERA5 reanalysis ( [[#Hersbach--2020|Hersbach et al., 2020]] ) are also broadly consistent with GMST trends from conventional surface datasets. However, the MERRA-2 reanalysis ( [[#Gelaro--2017|Gelaro et al., 2017]] ) GSAT spuriously cooled sharply relative to ERA-Interim and JRA-55 in about 2007 ( [[#Funk--2019|Funk et al., 2019]] ). Since the early 2000s, analyses of surface temperature, from which near-surface temperature may be derived, have also been available from various satellites ( [[#Famiglietti--2018|Famiglietti et al., 2018]] ; [[#Prakash--2018|Prakash et al., 2018]] ; [[#Susskind--2019|Susskind et al., 2019]] ), which have the potential to improve assessments of temperature changes over data-sparse regions.

Most land areas in the extratropical Northern Hemisphere (NH) have warmed faster than the GMST average over both the 1900–2020 and 1980–2020 periods (Figure 2.11b), although at more regional scales, particularly in data sparse regions, considerable uncertainty is introduced by sometimes large differences in trends between different LSAT datasets ( [[#Rao--2018|Rao et al., 2018]] ). Temperatures averaged over land areas globally have warmed by 1.59 [1.34 to 1.83] °C from 1850–1900 to 2011–2020, substantially higher than the SST warming of 0.88 [0.68 to 1.01] °C. The four conventional surface temperature products which meet all criteria to be included in the final assessment (Table 2.4) agree that each of the last four decades has consecutively been the warmest globally since the beginning of their respective records (Figure 2.11c and Table 2.4). Each of the six years 2015 to 2020 has ''very likely'' been at least 0.9°C warmer than the 1850–1900 average.

<div id="_idContainer034" class="Basic-Text-Frame"></div>

'''Table 2.4''' '''|''' '''Observed increase (°C) in GMST and underlying LSAT and SST estimates in various datasets.''' Numbers in square brackets indicate 5–95% confidence ranges. Trend values are calculated with ordinary least squares following [[#Santer--2008|Santer et al. (2008)]] and expressed as a total change over the stated period. Datasets considered in this table are those with data for at least 90% of global grid points in each year from 1960 onwards. GMST and SST are shown only for data sets which use air temperature (as opposed to climatological SST values) over sea ice. Changes from an 1850–1900 baseline are calculated only for those datasets which have data in at least 80% of years over 1850–1900. GMST values for each year are calculated as the mean of hemispheric means for the NH and SH, while LSAT and SST values are calculated from hemispheric means weighted according to the proportion of land (ocean) in the two hemispheres. This may vary from the methods used by individual data set providers in their own reporting. Products which meet all criteria to be included in the final assessment and contribute to the average are shown in italics. Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

{| class="wikitable"
|-
| Diagnostic/

Dataset

|
| '''1850–1900 to 1995–2014''' (°C)

| '''1850–1900 to 2001–2020''' (°C)

| '''1850–1900 to 2011–2020''' (°C)

| '''Trend'''

'''1880–2020''' (°C)

| '''Trend'''

'''1960–2020''' (°C)

| '''Trend'''

'''1980–2020''' (°C)

|-
| rowspan="3"| '''HadCRUT5'''

| GMST

| ''0.87''

''[0.81 to 0.94]''

| ''1.01''

''[0.94 to 1.09]''

| ''1.12''

''[1.06 to 1.18]''

| ''1.10''

''[0.89 to 1.32]''

| ''1.04''

''[0.93 to 1.14]''

| ''0.76''

''[0.65 to 0.87]''

|-
| LSAT

| ''1.23''

''[1.06 to 1.38]''

| ''1.44''

''[1.26 to 1.59]''

| ''1.55''

''[1.39 to 1.70]''

| ''1.43''

''[1.16 to 1.70]''

| ''1.50''

''[1.33 to 1.67]''

| ''1.20''

''[1.04 to 1.36]''

|-
| SST

| ''0.73''

''[0.69 to 0.78]''

| ''0.85''

''[0.81 to 0.90]''

| ''0.94''

''[0.90 to 0.99]''

| ''1.03''

''[0.80 to 1.25]''

| ''0.90''

''[0.80 to 0.99]''

| ''0.62''

''[0.51 to 0.72]''

|-
| rowspan="3"| '''NOAA GlobalTemp – Interim'''

| GMST

| ''0.76''

| ''0.91''

| ''1.02''

| ''1.06''

''[0.80 to 1.32]''

| ''1.01''

''[0.90 to 1.11]''

| ''0.75''

''[0.63 to 0.87]''

|-
| LSAT

| ''1.34''

| ''1.55''

| ''1.69''

| ''1.58''

''[1.32 to 1.84]''

| ''1.54''

''[1.40 to 1.68]''

| ''1.19''

''[1.04 to 1.35]''

|-
| SST

| ''0.53''

| ''0.65''

| ''0.75''

| ''0.85''

''[0.59 to 1.12]''

| ''0.79''

''[0.69 to 0.89]''

| ''0.57''

''[0.44 to 0.70]''

|-
| rowspan="3"| '''GISTEMP v4'''

| GMST

|
| ''1.07''

''[0.80 to 1.34]''

| ''1.05''

''[0.94 to 1.16]''

| ''0.79''

''[0.67 to 0.90]''

|-
| LSAT

|
| ''1.48''

''[1.19 to 1.78]''

| ''1.56''

''[1.40 to 1.72]''

| ''1.23''

''[1.07 to 1.39]''

|-
| SST

|
| ''0.91''

''[0.65 to 1.17]''

| ''0.84''

''[0.74 to 0.95]''

| ''0.61''

''[0.49 to 0.72]''

|-
| rowspan="3"| '''Berkeley Earth'''

| GMST

| ''0.89''

| ''1.03''

| ''1.14''

| ''1.17''

''[0.94 to 1.40]''

| ''1.09''

''[1.00 to 1.19]''

| ''0.79''

''[0.68 to 0.90]''

|-
| LSAT

| ''1.28''

| ''1.49''

| ''1.60''

| ''1.50''

''[1.25 to 1.76]''

| ''1.51''

''[1.36 to 1.66]''

| ''1.16''

''[1.00 to 1.32]''

|-
| SST

| ''0.73''

| ''0.85''

| ''0.96''

| ''1.04''

''[0.81 to 1.26]''

| ''0.93''

''[0.84 to 1.01]''

| ''0.64''

''[0.54 to 0.74]''

|-
| '''China-MST'''

| LSAT

| ''1.18''

| ''1.38''

| ''1.49''

| ''1.48''

''[1.21 to 1.75]''

| ''1.48''

''[1.31 to 1.65]''

| ''1.16''

''[1.00 to 1.32]''

|-
| rowspan="3"| '''Kadow et al.'''

| GMST

| ''0.86''

| ''1.00''

| ''1.09''

| ''1.15''

''[0.95 to 1.35]''

| ''1.01''

''[0.92 to 1.10]''

| ''0.73''

''[0.63 to 0.82]''

|-
| LSAT

| ''1.29''

| ''1.49''

| ''1.61''

| ''1.60''

''[1.37 to 1.82]''

| ''1.46''

''[1.30 to 1.61]''

| ''1.14''

''[0.99 to 1.30]''

|-
| SST

| ''0.69''

| ''0.80''

| ''0.88''

| ''0.97''

''[0.78 to 1.16]''

| ''0.83''

''[0.76 to 0.90]''

| ''0.56''

''[0.48 to 0.65]''

|-
| rowspan="3"| '''Cowtan-Way'''

| GMST

| 0.82

[0.75 to 0.89]

| 0.96

[0.89 to 1.03]

| 1.04

[0.97 to 1.11]

| 1.03

[0.84 to 1.22]

| 0.94

[0.82 to 1.07]

| 0.77

[0.67 to 0.87]

|-
| LSAT

| 1.23

| 1.43

| 1.54

| 1.42

[1.15 to 1.68]

| 1.48

[1.31 to 1.65]

| 1.20

[1.04 to 1.36]

|-
| SST

| 0.66

| 0.76

| 0.84

| 0.88

[0.71 to 1.05]

| 0.73

[0.61 to 0.84]

| 0.61

[0.52 to 0.69]

|-
| rowspan="3"| '''Vaccaro et al.'''

| GMST

| 0.76

| 0.89

| 0.97

| 0.99

[0.81 to 1.17]

| 0.89

[0.77 to 1.00]

| 0.72

[0.63 to 0.81]

|-
| LSAT

| 1.15

| 1.35

| 1.47

| 1.40

[1.13 to 1.67]

| 1.47

[1.29 to 1.64]

| 1.21

[1.06 to 1.36]

|-
| SST

| 0.60

| 0.70

| 0.77

| 0.82

[0.67 to 0.97]

| 0.66

[0.55 to 0.76]

| 0.53

[0.44 to 0.61]

|-
| rowspan="2"| '''ERA5'''

| GSAT

|
| 0.78

[0.64 to 0.92]

|-
| LSAT

|
| 1.21

[1.02 to 1.40]

|-
| Average – GMST

|
| 0.85

| 0.99

| 1.09

| 1.11

| 1.04

| 0.76

|-
| Average – LSAT

|
| 1.27

| 1.47

| 1.59

| 1.50

| 1.51

| 1.18

|-
| Average – SST

|
| 0.67

| 0.79

| 0.88

| 0.96

| 0.86

| 0.60

|}

To conclude, from 1850–1900 to 1995–2014, GMST increased by 0.85 [0.69 to 0.95] °C, to the first two decades of the 21st century (2001–2020) by 0.99 [0.84 to 1.10] °C, and to the most recent decade (2011–2020) by 1.09 [0.95 to 1.20] °C. Each of the last four decades has in turn been warmer than any decade that preceded it since 1850. Temperatures have increased faster over land than over the oceans since 1850–1900, with warming to 2011–2020 of 1.59 [1.34 to 1.83] °C versus 0.88 [0.68 to 1.01] °C, respectively.

<div id="2.3.1.2" class="h3-container"></div>

<span id="temperatures-during-the-instrumental-period-free-atmosphere"></span>