Records of CO₂ from the AIS formed during the last glacial period and the LDT show century-scale fluctuations of up to 9.6 ppm (Ahn et al., 2012; Ahn and Brook, 2014; Marcott et al., 2014; Bauska et al., 2015; Rubino et al., 2019). Although these rates are an order of magnitude lower than those directly observed over 1919–2019 CE (Section 2.2.3.3.1), they provide information on non-linear responses of climate-biogeochemical feedbacks (Section 5.1.2). Multiple records for 0–1850 CE show CO₂ mixing ratios of 274–285 ppm. Offsets among ice core records are about 1%, but the long-term trends agree well and show coherent multi-centennial variations of about 10 ppm (Ahn et al., 2012; Bauska et al., 2015; Rubino et al., 2019). Multiple records show CO₂ concentrations of 278.3 ± 2.9 ppm in 1750 and 285.5 ± 2.1 ppm in 1850 (Siegenthaler et al., 2005; MacFarling Meure et al., 2006; Ahn et al., 2012; Bauska et al., 2015). CO₂ concentration increased by 5.0 ± 0.8 ppm during 970–1130 CE, followed by a decrease of 4.6 ± 1.7 ppm during 1580–1700 CE. The greatest rate of change over the CE prior to 1750 is observed at about 1600 CE, and ranges from –6.9 to +4.7 ppm per century in multiple high-resolution ice core records (Siegenthaler et al., 2005; MacFarling Meure et al., 2006; Ahn et al., 2012; Bauska et al., 2015; Rubino et al., 2019). Although ice core records present low-pass filtered time series due to gas diffusion and gradual bubble close-off in the snow layer over the ice sheet (Fourteau et al., 2020), the rate of increase since 1850 CE (about 125 ppm increase over about 170 years) is far greater than implied for any 170-year period by ice core records that cover the last 800 ka (very high confidence).

CH₄ concentrations over the past 110 kyr are higher in the Northern Hemisphere (NH) than in the Southern Hemisphere (SH), but closely correlated on centennial and millennial timescales (Buizert et al., 2015). On glacial to interglacial cycles, approximately 450 ppb oscillations in CH₄ concentrations have occurred (Loulergue et al., 2008). On millennial timescales, most rapid climate changes observed in Greenland and other regions are coincident with rapid CH₄ changes (Buizert et al., 2015; Rhodes et al., 2015, 2017). The variability of CH₄ on centennial timescales during the early Holocene does not significantly differ from that of the late Holocene prior to about 1850 (Rhodes et al., 2013; Yang et al., 2017). The LGM concentration was 390.5 ± 6.0 ppb (Kageyama et al., 2017). The global mean concentrations during 0–1850 CE varied between 625 and 807 ppb. High-resolution ice core records from Antarctica and Greenland exhibit the same trends with an inter-polar difference of 36–47 ppb (Sapart et al., 2012; L. Mitchell et al., 2013). There is a long-term positive trend of about 0.5 ppb per decade during the CE until 1750 CE. The most rapid CH₄ changes prior to industrialization were as large as 30–50 ppb on multi-decadal timescales. Global mean CH₄ concentrations estimated from Antarctic and Greenland ice cores are 729.2 ± 9.4 ppb in 1750 and 807.6 ± 13.8 ppb in 1850 (L. Mitchell et al., 2013).

New records show that N₂O concentration changes are associated with glacial-interglacial transitions (Schilt et al., 2014). The most rapid change during the last glacial termination is a 30 ppb increase in a 200-year period, which is an order of magnitude smaller than the modern rate (Section 2.2.3.3). During the LGM, N₂O was 208.5 ± 7.7 ppb (Kageyama et al., 2017). Over the Holocene the lowest value was 257 ± 6.6 ppb during 6–8 ka, but millennial variation is not clearly detectable due to analytical uncertainty and insufficient ice core quality (Flückiger et al., 2002; Schilt et al., 2010). Recently acquired high-resolution records from Greenland and Antarctica for the last 2 kyr consistently show multi-centennial variations of about 5–10 ppb (Figure 2.4), although the magnitudes vary over time (Ryu et al., 2020). Three high temporal resolution records exhibit a short-term minimum at about 600 CE of 261 ± 4 ppb (MacFarling Meure et al., 2006; Ryu et al., 2020). It is very likely that industrial N₂O increase started before 1900 CE (Machida et al., 1995; Sowers, 2001; MacFarling Meure et al., 2006; Ryu et al., 2020). Multiple ice cores show N₂O concentrations of 270.1 ± 6.0 ppb in 1750 and 272.1 ± 5.7 ppb in 1850 (Machida et al., 1995; Flückiger et al., 1999; Sowers, 2001; Rubino et al., 2019; Ryu et al., 2020).

There has been a positive trend in globally averaged surface CO₂ mixing ratios since 1958 (Figure 2.5a), that reflects the imbalance of sources and sinks (Section 5.2). The growth rate has increased overall since the 1960s (Figure 2.5a inset), while annual growth rates have varied substantially, for example, reaching a peak during the strong El Niño events of 1997–1998 and 2015–2016 (Bastos et al., 2013; Betts et al., 2016). The average annual CO₂ increase from 2000 through 2011 was 2.0 ppm yr^–1 (standard deviation 0.3 ppm yr^–1), similar to what was reported in AR5. From 2011 through 2019 it was 2.4 ppm yr^–1 (standard deviation 0.5 ppm yr^–1), which is higher than that of any comparable time period since global measurements began. Global networks consistently show that the globally averaged annual mean CO₂ has increased by 5.0% since 2011, reaching 409.9 ± 0.4 ppm in 2019 (NOAA measurements). Further assessment of changing seasonality is undertaken in Section 2.3.4.1.

Figure 2.5 | Globally averaged dry-air mole fractions of greenhouse gases. (a) CO₂ from SIO, CSIRO, and NOAA/GML (b) CH₄ from NOAA, AGAGE, CSIRO, and UCI; and (c) N₂O from NOAA, AGAGE, and CSIRO (Table 2.2). Growth rates, calculated as the time derivative of the global means after removing seasonal cycle are shown as inset figures. Note that the CO₂ series is 1958–2019 whereas CH₄, and N₂O are 1979–2019. Units are parts per million (ppm) or parts per billion (ppb). Further details on data are in Annex III, and on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

The globally averaged surface mixing ratio of CH₄ in 2019 was 1866.3 ± 3.3 ppb, which is 3.5% higher than 2011, while observed increases from various networks range from 3.3–3.9% (Table 2.2 and Figure 2.5b). There are marked growth rate changes over the period of direct observations, with a decreasing rate from the late-1970s through the late-1990s, very little change in concentrations from 1999–2006, and resumed increases since 2006. Atmospheric CH₄ fluctuations result from complex variations of sources and sinks. A detailed discussion of recent methane trends and our understanding of their causes is presented in Cross-Chapter Box 5.2.

The AR5 reported 324.2 ± 0.1 ppb for global surface annual mean N₂O in 2011; since then, it has increased by 2.4% to 332.1 ± 0.4 ppb in 2019. Independent measurement networks agree well for both the global mean mixing ratio and relative change since 2011 (Table 2.2). Over 1995–2011, N₂O increased at an average rate of 0.79 ± 0.05 ppb yr^–1. The growth rate has been higher in recent years, amounting to 0.96 ± 0.05 ppb yr^–1 from 2012 to 2019 (Figure 2.5c and Section 5.2.3.5).

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 et al., 2015; 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^–1.

For the PETM, new reconstructions agree with those assessed by AR5. A major new compilation of proxy temperature data (Hollis et al., 2019) analysed using multiple statistical approaches (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 et al., 2019). A recent benthic isotope compilation (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 et al. (2008). The latter was originally used by J. 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 et al., 2020) and 12°C–18°C (95% range; 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 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 et al., 2009) and 5°C–10°C (2 standard error range; 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 and Dowsett, 2019; 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 confidence).

For the LIG (Cross-Chapter Box 2.1, Figure 1, and Figure 2.11), a major new compilation of marine proxy data (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 et al., 2017), and about 1.4°C (no uncertainty stated) warmer at 125 ka relative to 1850–1900 (Friedrich and Timmermann, 2020; reported relative to 10–5 ka and adjusted here by 0.4°C; 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 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) 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 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).

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 (Section 2.3.1.1). The last decade value and very likely range arises from 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 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 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 Section 2.3.1.1.3. The grey shading in each panel shows the uncertainty associated with the HadCRUT5 estimate (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).

Open figure

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), 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 and Timmermann, 2020). A third new estimate (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 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 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 et al., 2015; Figure 7 in Harrison et al., 2016). A major new pollen-based data-assimilation reconstruction averages 6.9°C cooler over northern extratropical land (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 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 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 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 et al., 2020).

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 et al., 2020a) based on a quality-controlled, multi-proxy synthesis of paleo-temperature records from 470 terrestrial and 209 marine sites globally (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 2k Consortium, 2019), and was centred around 6.5 ka. This is similar to 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 et al., 2014; Hou et al., 2019; 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 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 et al., 2015; Figure 7 in Harrison et al., 2016). In contrast, the GMST estimate from the multi-method global reconstruction (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; 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 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 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 and Pollack (2016) and Christiansen and Ljungqvist (2017). New regional (e.g., Shi et al., 2015; Stenni et al., 2017; Werner et al., 2018), global ocean (McGregor et al., 2015), quasi-hemispheric (Neukom et al., 2014; Schneider et al., 2015; Anchukaitis et al., 2017), and global (Tardif et al., 2019) temperature reconstructions, and new regional proxy data syntheses (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 2k Consortium, 2017) has been analysed using a variety of statistical methods for reconstructing temperature (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^–1 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 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 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).

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 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 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.

Marine domain

For SST analyses, three products – HadSST4 (1850–present, Kennedy et al., 2019), ERSSTv5 (1850–present, Huang et al., 2017) and COBE SST2 (1880–present, (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 et al., 2011; 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 and Kaplan, 2006) but have an overall warm bias when globally aggregated (Kennedy et al., 2019). The first issue primarily affects data since 1990, when buoys began to increasingly contribute to the observation network (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 et al., 2015; Huang et al., 2016) and subsequent versions (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; 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).

A number of recent studies also corroborate important components of the SST record (Hausfather et al., 2017; Kent et al., 2017; Cowtan et al., 2018; Kennedy et al., 2019). In particular, ATSR SST satellite retrievals (Merchant et al., 2012; Berry et al., 2018), the near-surface records from hydrographical profiles (Gouretski et al., 2012; Huang et al., 2018), and coastal observations (Cowtan et al., 2018) have all been shown to be broadly consistent with the homogenized SST analyses. 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 et al., 2017), particularly during World War II, when ship observations were limited and disproportionately originated from US naval sources (Thompson et al., 2008). 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 et al. (2019) and 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 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 et al., 2017). Observational datasets exist for night-marine air temperature (NMAT) (e.g., Cornes et al., 2020; Junod and Christy, 2020; Rayner et al., 2020) and there are methods to adjust daytime MATs (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 et al., 2019) which generally measure SST but not MAT, and there are almost no recent winter MAT data south of 40°S (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 et al., 2017, 2019; Kent and Kennedy, 2021).

Land domain

The GHCNMv4 dataset (Menne et al., 2018) includes many more land stations than GHCNMv3, arising from the databank efforts of Rennie et al. (2014), and calculates a 100-member parametric uncertainty ensemble drawing upon the benchmarking analysis of Williams et al. (2012), as well as accounting for sampling effects. A new version of the CRUTEM dataset (CRUTEMv5, 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 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 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 et al. (2005), HadEX2 (Donat et al., 2013a), HadGHCND (Donat et al., 2013b), GHCNDEX (Donat et al., 2013b), Berkeley Earth (Rohde et al., 2013), and CRU TS (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 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 (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 et al., 2013; Liao et al., 2017; Z. 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 et al., 2012), but less so elsewhere.

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 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 et al., 2019). Cowtan and Way (2014) applied a kriging-based method to extend existing datasets to polar regions, while Kadow et al. (2020) used an artificial intelligence-based method, and Vaccaro et al. (2021) used gaussian random Markov fields, for the same purpose, although only Kadow et al. (2020) uses the most recent generation of datasets as its base. The Berkeley Earth merged product (Rohde and Hausfather, 2020), HadCRUT5 (Morice et al., 2021) and NOAA GlobalTemp-Interim (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 et al., 2021), and their strong warming at high northern latitudes is consistent with independent estimates from reanalyses (Simmons et al., 2017; Lenssen et al., 2019) and satellites (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.

Estimates of GMST have also benefitted from improved estimation of parametric uncertainties. New versions of three long-standing products from NASA GISTEMP v4 (Lenssen et al., 2019), NOAA GlobalTempv5 (B. Huang et al., 2019b) and HadCRUT5 (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 (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 et al. (2017) found that the ERA-Interim (Dee et al., 2011) and JRA-55 (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 et al., 2020) are also broadly consistent with GMST trends from conventional surface datasets. However, the MERRA-2 reanalysis (Gelaro et al., 2017) GSAT spuriously cooled sharply relative to ERA-Interim and JRA-55 in about 2007 (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 et al., 2018; Prakash et al., 2018; 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 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.

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.

There have been updated radiosonde estimates from the University of Vienna (RAOBCORE and RICH; Haimberger et al., 2012) and a new dataset from the State University of New York (UAHRD, Zhou et al., 2020). There are new versions of AMSU products from the University of Alabama in Huntsville (UAHv6.0; Spencer et al., 2017) and Remote Sensing Systems (RSSv4.0; Mears and Wentz, 2017). These updates have led to convergence in the lower stratosphere layer (Maycock et al., 2018); in particular, the move to UAHv6.0 has addressed homogeneity issues identified by Seidel et al. (2016), although residual differences remain (Christy et al., 2018). Reanalyses products had identified limitations near the 300 hPa level where the contribution of aircraft observations has increased rapidly in recent years (Dee et al., 2011; Gelaro et al., 2017), leading to identified biases (Dee and Uppala, 2009), that have been addressed in ERA5 (Hersbach et al., 2020). Modern reanalyses are generally well aligned with radiosonde and satellite observations in the middle and lower troposphere and lower stratosphere. A new operational mid- and upper-stratospheric dataset (STAR) has been developed by Zou and Qian (2016), merging the previous 1979–2006 SSU dataset (Zou et al., 2014) with a dataset from 1998 onwards drawn from relevant AMSU channels (Wang and Zou, 2014). Further stratospheric satellite-based datasets from various combinations of satellites have been developed by McLandress et al. (2015) and Randel et al. (2016).

New assessments of free-atmosphere temperature are available through radio occultation (RO) and Atmospheric Infrared Sounder (AIRS) products which begin in the early 2000s (Section 1.5.1.1). Global Navigation Satellite System (GNSS)-RO datasets have been compared against AMSU data records, finding almost identical trends (Khaykin et al., 2017). Comparison of RO with collocated radiosondes, Vaisala RS90/92 and GCOS Reference Upper Air Network data (RS92-GDP; Dirksen et al., 2014), show very good correspondence with global annual mean differences of less than 0.2°C in the upper troposphere and lower stratosphere. Radiosonde daytime radiation biases were identified at higher altitudes (Ladstädter et al., 2015; Ho et al., 2017). The stability of RO makes this data a useful comparator for AMSU (Chen and Zou, 2014) and radiosondes (Ho et al., 2017; Tradowsky et al., 2017), as well as anchoring post-2006 reanalyses datasets and improving their consistency in the lower and middle stratosphere (Long et al., 2017; Ho et al., 2020). The effective vertical resolution of RO measurements in the upper troposphere and lower stratosphere was found to be up to 100 m at the tropical tropopause (Zeng et al., 2019a), which is favourable for resolving atmospheric variability (Scherllin-Pirscher et al., 2012; Wilhelmsen et al., 2018; Stocker et al., 2019). Temperature trends in RO products are most consistent with each other and with other observations between 8 km and 25 km (Ho et al., 2012; Steiner et al., 2013, 2020a). The uncertainty increases above 25 km for the early RO period, for which data are based on the single-satellite CHAMP mission, but data at higher altitudes become more reliable for later missions based on advanced receivers (Steiner et al., 2020a), along with the application of corrections for ionospheric effects (Danzer et al., 2020). The uncertainty due to the changing number of observations is reduced by correcting for the sampling uncertainty in RO climatological fields (e.g., Scherllin-Pirscher et al., 2011). For AIRS, thus far, stability of the instrument has been constrained to less than 0.03°C per decade for selected window channels in a comparison to SSTs measured by ocean buoys (Aumann et al., 2019). Trends were inter-compared with trends in RO data and reanalysis data to assess systematic uncertainties (Leroy et al., 2018).

Warming has continued in the lower troposphere according to all radiosonde, reanalyses and satellite datasets, with a rate over 1980–2019 similar to surface warming rates (Table 2.5; c.f. Table 2.4). Radiosonde-based products generally show greater warming rates for 1980–2019 than satellite-based products and reanalyses. They also extend further back to the 1950s and trends since quasi-global coverage around 1960 also show warming (Table 2.5). Trends in RO and AIRS data, supported by radiosonde datasets, exhibit a warming trend in most of the mid- to upper- troposphere at all non-polar latitudes over 2002–2019. These also exhibit faster warming rates in the tropics in the upper troposphere than those observed at or near the surface (Figure 2.12); with the lowermost stratosphere also warming while above it is cooling. There is some spread between different data types in the tropics near the 15km level, although these differences are reduced to near zero if a subset of radiosonde data, using only high-quality instruments, is used (Steiner et al., 2020b). AMSU tropical middle troposphere data also show that warming rates are near or above those in the lower troposphere, but they are measuring much broader layers which greatly complicates interpretation (Steiner et al., 2020b).

Temperatures averaged through the full lower stratosphere (roughly 10–25 km) have decreased over 1980–2019 in all data products, with the bulk of the decrease prior to 2000. The decrease holds even if the influence of the El Chichon (1982) and Pinatubo (1991) volcanic eruptions on the trend, found by Steiner et al. (2020a) to have increased the 1979–2018 cooling trend by 0.06°C per decade, is removed. Most datasets show no significant or only marginally significant trends over 2000–2019, and the results of Philipona et al. (2018) show weak increases over 2000–2015 in the very lowermost stratosphere sampled by radiosondes.

The STAR dataset shows cooling in the middle and upper stratosphere with a trend of –0.56°C ± 0.16°C per decade for the mid-stratosphere and –0.62°C ± 0.29°C per decade for the upper stratosphere over 1980–2019, although both cooling rates have slowed substantially since the mid-1990s. The overall post-1980 trend is reduced in magnitude by about 0.10°C per decade at both levels if the influences of the El Chichon and Pinatubo eruptions, and the solar cycle, are removed (Zou and Qian, 2016). The results obtained by McLandress et al. (2015) for 1980–2012, Randel et al. (2016) for 1979–2015, and Maycock et al. (2018) for 1979–2016 are broadly consistent with this.

A rise in the tropopause height of 40 to 120 m per decade between 1981 and 2015 was determined from both radiosonde and reanalysis datasets (Xian and Homeyer, 2019). Local studies (e.g., Tang et al., 2017; X. Chen et al., 2019) found stronger trends in some regions near the subtropical jet linked to tropical expansion (Section 2.3.1.4.1). Whilst Seidel and Randel (2006) found that the tropopause height was more closely coupled with temperatures in the stratosphere than those in the troposphere, it is not yet clear whether the rate of increase in tropopause height has experienced a similar recent slowdown to that of the cooling of the lower stratosphere, as short-period trends are typically inconclusive due to significant natural variability (Scherllin-Pirscher et al., 2021). RO data (Gao et al., 2015) indicate little change in tropopause height over the short period from 2006 to 2014, but a warming below the tropopause is observed over 2002 to 2019 (Figure 2.12).

Figure 2.12 | Temperature trends in the upper air. (a) Zonal cross-section of temperature anomaly trends (2007–2016 baseline) for 2002–2019 in the upper troposphere and lower stratosphere region. The climatological tropopause altitude is marked by a grey line. Significance is not indicated due to the short period over which trends are shown, and because the assessment findings associated to this figure relate to difference between trends at different heights, not the absolute trends. (b, c) Trends in temperature at various atmospheric heights for 1980–2019 and 2002–2019 for the near-global (70°N–70°S) domain. (d, e) as for (b, c) but for the tropical (20°N–20°S) region. Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

In summary, the troposphere has warmed since the mid-20th century. There is medium confidence that temperatures in the tropical upper troposphere have warmed faster than those at the surface since 2001, but low confidence in changes prior to 2001. It is virtually certain that the lower stratosphere has cooled since the mid-20th century. However, most datasets show that lower stratospheric temperatures have stabilized since the mid-1990s with no significant change over the last 20 years. It is likely that middle and upper stratospheric temperatures have decreased since 1980, but there is low confidence in the magnitude. It is virtually certain that the tropopause height has risen over 1980–2019 but there is low confidence in the magnitude of this rise, or whether the rate of change has reduced commensurate with stabilized lower stratospheric temperatures.

The AR5 assessed large-scale indicators of terrestrial paleo hydroclimate, including as part of its assessment of paleo floods and droughts, but did not assess proxy evidence for paleo hydroclimate indicators over continental and larger scales. The paleoclimate evidence assessed in AR5 was broadly consistent with global hydroclimate scaling with temperature: warmer periods were wetter (e.g., the Pliocene; increased precipitation) with colder periods being drier (e.g., the LGM; decreased precipitation).

Substantial limitations exist in reconstructing the global hydrological cycle prior to the Quaternary, particularly during the Eocene, due to the lack of high-resolution proxy records and their sparsity. Spatial heterogeneity complicates identification of wetting and drying signals during the PETM and the EECO, with paleo data and model simulations suggesting an intensified global hydrological cycle (Carmichael et al., 2016, 2017; Hyland et al., 2017; West et al., 2020), in particular an increased specific humidity (Winnick et al., 2015; van Dijk et al., 2020). Conditions wetter than present were inferred for the MPWP (Cross Chapter Box 2.4), with intensified Asian monsoons (An et al., 2015) but with nevertheless drier conditions over tropical and subtropical SH locations (Pontes et al., 2020). A new global reconstruction of hydroclimate proxies for the LIG points to stronger boreal precipitation compared to 1850–1900 over high latitudes and especially over monsoon areas, with a more heterogeneous signal for the SH (Scussolini et al., 2019). This heterogeneity is also present in the tropics, characterized by large zonal differences in precipitation change due to the variations in the intensity of Walker circulation (Section 2.3.1.4.1). Available records indicate reduced global vegetation cover and abundant atmospheric dust deposition during the LGM (increased aridity), particularly over the tropics and high latitudes (Lamy et al., 2014; Újvári et al., 2017). This agrees with models and moisture-sensitive proxies, suggesting an overall decrease in global precipitation during the LGM relative to recent decades, albeit with regional-scale heterogeneity (Cao et al., 2019). Despite lower global precipitation amounts, research since AR5 has identified a wetting of mid-latitudes during the LGM (Putnam and Broecker, 2017; Lowry and Morrill, 2018; Morrill et al., 2018), thereby complicating the characterization of the LGM as a relatively ‘dry’ period. Low evaporation rates and increased top-soil moisture during the LGM may have contributed to elevated levels of large closed-basin lakes located in the 30°–45° latitudinal belts (Putnam and Broecker, 2017; Scheff et al., 2017), such as the south-west United States (e.g., Ibarra et al., 2018), southern Australia (Petherick et al., 2013; Fitzsimmons et al., 2015; Sniderman et al., 2019) and Patagonia (e.g., Quade and Kaplan, 2017).

New analyses suggest that during the Holocene, the NH mid-latitudes became increasingly wet, in phase with the strength of the latitudinal temperature and insolation gradients (Shuman and Marsicek, 2016; Routson et al., 2019). Nevertheless, there was also considerable spatial heterogeneity and variability on centennial to millennial timescales (Newby et al., 2014; Shuman and Marsicek, 2016; H. Zhang et al., 2018; Liefert and Shuman, 2020). The NH tropics and many regions of the SH deep tropics experienced wetting up until the early to mid-Holocene but drying thereafter (Shanahan et al., 2015; Nash et al., 2016; Muñoz et al., 2017; Quade et al., 2018). Evidence for the SH is limited, with a wetting trend during the Holocene in low latitudes of South America (Kanner et al., 2013; Mollier-Vogel et al., 2013) and parts of the African tropics (Schefuß et al., 2011; Chevalier and Chase, 2015) but a drying tendency over southern Australia and New Zealand (van den Bos et al., 2018; Barr et al., 2019) and South America (Quade and Kaplan, 2017; Moreno et al., 2018).

For the CE, new proxy records have led to the creation of continental drought atlases (Cook et al., 2015; Palmer et al., 2015; Stahle et al., 2016; Morales et al., 2020) and millennial reanalyses (Steiger et al., 2018; Tardif et al., 2019). These reconstructions highlighted the occurrence of multi-decadal regional mega-droughts in the NH before 1600 CE, particularly during 800–1200 CE, with a predominance of wet periods after 1700 CE (Cook et al., 2015; Rodysill et al., 2018; Shuman et al., 2018). In the SH, much of South America and the African tropics experienced a reduction of precipitation during 900–1200 CE and a wetting peak during 1500–1800 CE (Tierney et al., 2015; Nash et al., 2016; Fletcher et al., 2018; Lüning et al., 2018; Campos et al., 2019), with an opposite pattern in southern subtropical Africa (Woodborne et al., 2015; Lüning et al., 2018). Large multi-decadal variability was documented over Australia and New Zealand during the 800–1300 CE period, followed by a well-defined wet period during 1500–1800 CE (Barr et al., 2014; Evans et al., 2019).

To summarize, since AR5 there has been considerable progress in detecting the variations of the global hydrological cycle prior to the instrumental period. There are indications from multiple sources of a wetting trend during the Holocene, particularly for the NH and parts of the SH tropics (medium confidence). Hydroclimate during the CE is dominated by regional variability, generally precluding definitive statements on changes at continental and larger scales, with a general reduction of mega-drought occurrences over the last about 500 years (medium confidence). Availability of proxy data for assessing Holocene hydroclimate variability is biased towards the NH, with medium evidence butlow agreement for the assessment of SH changes.

The AR5 reported very likely widespread increases in near-surface air specific humidity since the 1970s, abating from around 2000 to 2012 (medium confidence). This abatement resulted in a recent decline in relative humidity over the land.

Near surface humidity has been monitored using in-situ data (e.g., NOCSv2.0; Berry and Kent, 2011), satellite-derived estimations (e.g., HOAPS3, Liman et al., 2018; J-OFURO3, Tomita et al., 2019), global gridded products such as HadISDH (Willett et al., 2014, 2020), and reanalyses (e.g., ERA5, JRA-55 and 20CRv3). In-situ based humidity products suffer from uncertainties over poorly sampled regions particularly in the SH (Berry and Kent, 2011; Kent et al., 2014; Willett et al., 2014). There is general consensus in the inter-annual variability and sign of trends implying high confidence in increasing specific humidity since the 1970s and decreasing relative humidity since 2000, particularly over land (Simmons et al., 2010; Willett et al., 2014, 2020). Since 2012, specific humidity over land and ocean has remained well above the 1973–2019 average and reached record or near-record values (Figure 2.13b), with the strong 2015–2016 El Niño event boosting surface moisture levels (Byrne and O’Gorman, 2018). The abatement from around 2000 to 2012 reported in AR5 has not persisted. This is consistent with increases in total column water vapour (Section 2.3.1.3.3) and a resumption of rapid warming in surface temperatures (Section 2.3.1.1.3). The global averaged relative humidity however has remained depressed since 2000 (Figure 2.13d; Simmons et al., 2010; Willett et al., 2014, 2020; Dunn et al., 2017; Vicente-Serrano et al., 2018).

Since 1973, increases in specific humidity have been widespread and significant across the majority of the land and ocean regions where observations are available (Figure 2.13a). In contrast, trends in relative humidity show distinct spatial patterns with generally increasing trends over the higher latitudes and the tropics and generally decreasing trends over the sub-tropics and mid-latitudes, particularly over land areas (Figure 2.13c). Near-surface specific humidity over the oceans has increased since the 1970s according to several in-situ, satellite and reanalysis data records (Kent et al., 2014; Robertson et al., 2020; Willett et al., 2020). According to the HadISDH product, increases in specific humidity and decreases in relative humidity are significant particularly over the NH mid-latitudes (Figure 2.13a,c). Poor data coverage over the SH south of 20°S does not allow for the robust assessment of trends. Sources of uncertainty include the initial measurement accuracy, homogenization over land, observational height at ships and instrument bias adjustment over ocean, and sparse spatio-temporal sampling (Prytherch et al., 2015; Roberts et al., 2019; Willett et al., 2020).

Figure 2.13 | Changes in surface humidity. (a) Trends in surface specific humidity over 1973–2019. Trends are calculated using OLS regression with significance assessed following AR(1) adjustment after Santer et al. (2008); ‘×’ marks denote non-significant trends). (b) Global average surface specific humidity annual anomalies (1981–2010 base period). (c) as (a) but for the relative humidity. (d) as (b) but for the global average surface relative humidity annual anomalies. Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

In summary, observations since the 1970s show avery likely increase in near surface specific humidity over both land and oceans. Avery likely decrease in relative humidity has occurred over much of the global land area since 2000, particularly over mid-latitude regions of the NH, with increases at northern high latitudes.

The AR5 concluded that total column water vapour (TCWV)very likely increased since the 1970s, at a rate that was overall consistent with the Clausius-Clapeyron relationship (about 7% per °C) given the observed increase in atmospheric temperature.

Records prior to the instigation of quasi-global coverage by radiosondes require the use of statistical relationships to infer TCWV from historical SST observations or the evaluation of centennial-scale reanalysis products (Smith and Arkin, 2015). These approaches reveal two periods of positive trends, one from 1910 to 1940 and the other from 1975 onwards (Zhang et al., 2013; Mieruch et al., 2014; Shi et al., 2018), concurrent with periods of positive SST trends (Figure 2.11). Potential sources of errors in the SST-based estimation of TCWV include both uncertainties in historical SST and uncertainties in the parameters that define the relationship between the variables (Smith and Arkin, 2015). Trends based on 20CRv2c, ERA-20C and ERA-20CM indicate an increase in TCWV over much of the global ocean since the beginning of the 20th century, particularly over the tropics (Bordi et al., 2015; Smith and Arkin, 2015; Poli et al., 2016). TCWV trends estimated since the middle of the 20th century from radiosonde observations show significant increases over North America and large portions of Eurasia, while decreases are restricted to Australia, eastern Asia and the Mediterranean region (Y. Zhang et al., 2018). Overall, there is a significant increase in TCWV over global land areas since 1979 (Chen and Liu, 2016).

Since the late 1970s a range of satellite missions permit a quasi-global assessment of TCWV. Several satellite products provide water vapour retrievals based upon distinct spectral domains, in addition to products from radiosondes, reanalyses and GNSS radio occultation. The GEWEX Water Vapour Assessment (G-VAP) provided an intercomparison of several TCWV data records, with global coverage but limited timespan (Schröder et al., 2018). The various global products generally exhibit a positive trend since 1979 (Figure 2.14; Allan et al., 2014; Mieruch et al., 2014; Schröder et al., 2016; J. Wang et al., 2016), most evident over the tropics (Gu and Adler, 2013; Chen and Liu, 2016; Mears et al., 2018; Wang and Liu, 2020; Salamalikis et al., 2021). The existence of apparent breakpoints in several products, which are generally coincident with changes in the observing system, lead to trend estimates that are not in line with theoretical expectations imposed by the Clausius-Clapeyron relationship (Schröder et al., 2019), although other factors such as regional moisture divergence/convergence could account for the observed TCWV-temperature scaling. Substantial potential inhomogeneities affect trend estimates based on satellite, reanalysis and merged products in particular over Central Africa, the Sahara and central South America (Schröder et al., 2016, 2019; J. Wang et al., 2016). Moreover, data gaps in observations from ground-based GNSS receivers and radiosondes lead to low confidence in TCWV estimation in these regions.

Figure 2.14 | Time series of global mean total column water vapour annual anomalies (mm) relative to a 1988–2008 base period. Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

In summary, positive trends in global total column water vapour are very likely since 1979 when globally representative direct observations began, although uncertainties associated with changes in the observing system imply medium confidence in estimation of the trend magnitudes. Low confidence in longer-term trends arises from uncertainties in the SST-TCWV relationship and current centennial scale reanalyses, particularly during the first half of the 20th century.

The AR5 concluded that there was low confidence in precipitation change averaged over global land areas prior to 1950, and medium confidence thereafter with no significant global trends. There was a likely overall increase in precipitation in the well-sampled NH mid-latitudes, with high confidence after 1951.

In situ precipitation records over land extend back for centuries in a few locations, and to the early to mid-20th century quasi-globally. Datasets differ in their input data, completeness of records, period covered, and the gridding procedures applied, which, given spatial clustering and the small spatial scales of precipitation, results in differences in global and regional estimates of precipitation changes (Q. Sun et al., 2018; Nogueira, 2020). The spatial variability of observed long-term trends (1901–2019) based on GPCC V2020 and CRU TS 4.04 (Figure 2.15a,b) indicates significant increases in precipitation mainly over eastern North America, northern Eurasia, southern South America and north-western Australia. Decreases are strongest across tropical western and equatorial Africa and southern Asia. The temporal evolution of global annual land precipitation anomalies exhibits little consistency between GPCC V2020, CRU TS 4.04 and GHCNv4 datasets, especially prior to 1950, that is associated with limitations in data coverage (Figure 2.15c; Wu et al., 2013; Shen et al., 2014; Gu and Adler, 2015). These disagreements between datasets prior to the 1950s result in differences in trend estimates over global land (Table 2.6). A qualitative consistency in decadal and interdecadal variations between the products is only observed since the 1950s, with primarily positive land precipitation anomalies during the 1950s, 1970s and during 2000 to 2019 (Figure 2.15c).

Figure 2.15 | Changes in observed precipitation. (a, b) Spatial variability of observed precipitation trends over land for 1901–2019 for two global in-situ products. Trends are calculated using OLS regression with significance assessed following AR(1) adjustment after Santer et al. (2008) (‘×’ marks denote non-significant trends). (c) Annual time series and decadal means from 1891 to date relative to a 1981–2010 climatology (note that different products commence at distinct times). (d, e) as(a, b), but for the periods starting in 1980. (f) is for the same period for the globally complete merged GPCP v2.3 product. Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

Several satellite-based precipitation datasets improve the representation of the spatio-temporal changes since the late 20th century. Some of these are based exclusively on satellite data (e.g., CMORPH, Joyce et al., 2004; GSMaP, Okamoto et al., 2005), with others being combinations of in situ observations, reanalyses and satellite retrievals (e.g., CMAP, Xie and Arkin, 1997; TRMM 3B43 V7, Huffman et al., 2007 ; PERSIANN-CDR, Ashouri et al., 2015; CHIRPS, Funk et al., 2015; GPCP V2.3, Adler et al., 2018). These can be affected by systematic and random uncertainties due to inhomogeneities in the satellite-derived precipitation and station data and the uncertainties of blending algorithms (Hegerl et al., 2015; Q. Sun et al., 2018). The spatial coverage of these products is near-global, with available estimations formally covering 60°S–60°N with decreasing quality from low to high latitudes, depending on the sensors and algorithms used (Hu et al., 2019). A detailed description of the most relevant satellite products is provided in section 10.2.1.1.

Recent trends (1980–2019) for GPCC V2020, CRU TS 4.04 and GPCP V2.3 show significant increases in land precipitation over tropical Africa, the eastern portions of Europe and North America, central Asia and the Maritime Continent (Figure 2.14d–f). Significant decreases are observed over central South America, western North America, northern Africa and the Middle East. A detailed assessment of the recent regional precipitation trends using the same datasets can be found in the Atlas. Global trends for 1980–2019 show a general increase in annual precipitation over land, which is particularly marked for CRU TS 4.04 and GHCNv4 (Table 2.6). These changes have been accompanied by a strengthening of precipitation seasonality over tropical land areas, although with broad spread between different satellite-based (GPCP, MSWEP_V1.2, PERSIANN-CDR) and in situ gridded datasets (GPCC, CRU TS; Chou et al., 2013; Li et al., 2016; Tan et al., 2020). Increasing trends since 1980, in contrast to longer-term declining trends since 1901, are particularly evident over much of Africa, while more widespread negative trends were observed over much of southern South America in the more recent period (Atlas 7.2; Knutson and Zeng, 2018). A faster recent increase in precipitation over global land is inferred comparing the precipitation trends over 1960–2019 with 1980–2019 (Table 2.6). Over the global ocean, the comparison between precipitation datasets is compromised by the different measurement periods, as well as the spatial coverage of the available products (Adler et al., 2017; Nguyen et al., 2018; Jaber and Abu-Allaban, 2020; Nogueira, 2020), limiting the ability to assess the sign and magnitude of precipitation trends. The GPCPv2.3 database (Adler et al., 2017, 2018) exhibits an increase of 2.94 mm yr^–1 per decade over 1980–2019, principally due to the trends over the Indian ocean and in the tropical western Pacific (Figure 2.15f). The regional patterns of recent trends are consistent with the documented increase in precipitation over tropical wet regions and the decrease over dry areas, estimated through GPCP v2.2 data (Liu and Allan, 2013; Trammell et al., 2015; Kao et al., 2017; Polson and Hegerl, 2017).

In summary, globally averaged land precipitation has likely increased since the middle of the 20th century (medium confidence), with low confidence in trends prior to 1950. A faster increase in global land precipitation was observed since the 1980s (medium confidence), with large interannual variability and regional heterogeneity. Over the global ocean there is low confidence in the estimates of precipitation trends, linked to uncertainties in satellite retrievals, merging procedures and limited in situ observations.

The AR5 concluded that the pattern of precipitation minus evaporation (P–E) over the ocean had been enhanced since the 1950s (medium confidence). Saline surface waters had become saltier, while the relatively fresh surface waters had become fresher. The inferred changes in P–E were consistent with the observed increased TCWV, although uncertainties in the available products prevented identifying robust trends.

Estimating global-scale trends in P–E using direct observations alone is challenging due to limited evaporation measurements and inhomogeneities in satellite-derived precipitation and evaporation datasets (Hegerl et al., 2015; López et al., 2017). Hence, the assessment of global P–E trends is generally performed using reanalyses, although changes in the observing system imply considerable uncertainty (Skliris et al., 2014). Since the second half of the 20th century, several reanalyses and observational datasets have shown increases in P–E over global land, although 75% of land areas exhibit no significant changes and both internal variability and observational uncertainty are substantial (Greve et al., 2014; Robertson et al., 2016). The recently released ERA5 (Hersbach et al., 2020) showed improvements in the representation of tropical precipitation, although it overestimates global precipitation trends in comparison to ERA-Interim and GPCP (Nogueira, 2020), and suffers from temporal changes in the annual balance between precipitation and evaporation (Hersbach et al., 2020). The spatial pattern of P–E trends over 1980–2019 (Figure 2.16a) are largely consistent with the trends in the GPCP v2.3 precipitation dataset (Figure 2.15f and Section 2.3.1.3.4) and agrees in sign with the trends from other reanalyses such as JRA-55 and MERRA-2 (L. Yu et al., 2020).

Figure 2. 16 | Changes in precipitation minus evaporation. (a) Trends in precipitation minus evaporation (P–E) between 1980 and 2019. Trends are calculated using OLS regression with significance assessed following AR(1) adjustment after Santer et al. (2008) (‘×’ marks denote non-significant trends). Time series of (b) global, (c) land-only and (d) ocean-only average annual P–E (mm day^–1). Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

A variety of reanalysis products exhibit diverse temporal evolutions of P–E (Figure 2.16b–d). Globally MERRA-2, ERA20C and ERA20CM exhibit little change whereas JRA-55, ERA5 and 20CRv3 all imply long-term changes (Figure 2.16d). A potential limitation in estimating P–E from some reanalysis products is readily apparent when considering the temporal evolution of global P–E from CFSR and MERRA (Figure 2.16d) which both exhibit strong discontinuities over the global ocean in the late 1990s. Over global land as a whole, precipitation exceeds evaporation (P–E >0) for all the reanalysis products (Figure 2.16c), with decreasing trends in P–E for ERA5 and JRA-55 and increasing trends for MERRA-2 and CFSR. The P–E over the global ocean is negative (evaporation exceeding precipitation) for most reanalyses (Figure 2.16d), with declining trends in ERA5 and MERRA-2 dominated by trends in evaporation (Bosilovich et al., 2017; Hersbach et al., 2020) (Figure 2.16d). The recent increase in ocean evaporation was also documented for several reanalyses (Craig et al., 2017) and in satellite data (Andersson et al., 2011; Robertson et al., 2014), although with considerable differences between available estimates (Chandanpurkar et al., 2017; L. Yu et al., 2020). An alternative indirect approach to estimate P–E changes is based on near-surface ocean salinity (Section 2.3.3.2), which is partially driven by the freshwater flux at the ocean surface. The near-surface salinity trends are more spatially coherent compared to those revealed by P–E estimates from reanalyses, with an intensification of the water cycle over oceans, especially in subtropical regions (Durack et al., 2012; Skliris et al., 2014; L. Yu et al., 2020). However, the precise rate of water cycle intensification implied by salinity trends is sensitive to methodological choices (e.g., Skliris et al., 2016; Zika et al., 2018).

In conclusion, observational uncertainty yields low confidence in globally averaged trends in P–E over the 20th century, with a spatial pattern dominated by precipitation changes over land and by evaporation increases over the ocean. Different reanalyses disagree on the sign of long-term changes in the global mean P–E.

The AR5 concluded that there was low confidence in a positive trend in global river discharge during the 20th century. It noted that many of the largest rivers with long term streamflow records have been impacted by non-climatic human influences such as dam construction or land-use change.

River discharge is monitored widely, although gaps remain at a subcontinental scale over central Asia and Africa (Wei et al., 2020). Substantial recent efforts have been made to generate new global streamflow datasets, consolidating observations from many stream gauges to create streamflow indices (Do et al., 2018; Gudmundsson et al., 2018) and gridded products using neural networks (Barbarossa et al., 2018) or combinations between observations and reanalyses (Suzuki et al., 2018; Ghiggi et al., 2019).

Human intervention on river discharge linked to increases in evapotranspiration and some reduction of intra-annual streamflow variability (Jaramillo and Destouni, 2015; Chai et al., 2020) might affect the detection of trends in extreme daily streamflow events (Do et al., 2017; Gudmundsson et al., 2019). However, these activities have a minor impact on annual streamflow compared to climate variations (Dai et al., 2009; Alkama et al., 2013). Available global studies post-1950 generally concur that there have been more rivers experiencing decreases than increases in runoff (Do et al., 2017; Su et al., 2018; Gudmundsson et al., 2019; X. Shi et al., 2019). Most of the rivers have not experienced statistically significant changes in streamflow, and when globally aggregated there is no significant change (Dai and Zhao, 2017). Global streamflow variability is strongly modulated by ENSO and PDV, with below-normal global streamflow as a response to El Niño events and vice-versa during La Niña episodes (Dai, 2016; Liang et al., 2016; Kim, 2019). The response of streamflow to changes in precipitation associated with ENSO and PDV has heterogeneous regional patterns at subcontinental scales (Section 8.3.2.9.1). No significant trends are found for reanalysis-based discharge estimates over 1993 to 2015 (Chandanpurkar et al., 2017). Uncertainties in global streamflow trends arise predominantly from changes in instrumentation, gauge restoration, recalibration of rating curves, flow regulation or channel engineering (Alkama et al., 2011; Gudmundsson et al., 2018; Ghiggi et al., 2019).

In summary, the sign of global streamflow trends remains uncertain, with slightly more globally gauged rivers experiencing significantly decreasing flows than significantly increasing flows since the 1950s (low confidence).

The AR5 reported low confidence in trends in the strength of the Hadley circulation (HC) and the Walker circulation (WC) due to uncertainties in available reanalysis datasets and the large interannual-to-decadal variability of associated circulation patterns. However, AR5 indicated a likely widening of the tropical belt since the 1970s, albeit with large uncertainty in the magnitude of this change. There was high confidence that the post-1990s strengthening of the Pacific WC reversed its weakening observed from the mid-19th century to the 1990s.

Paleo reconstructions of rainfall and trade winds extending over the last 100 kyr show an intensification of the NH HC concurrently with a weakening of the SH HC and a southward shift of the inter tropical convergence zone (ITCZ) during Heinrich stadials (Deplazes et al., 2013; McGee et al., 2018; Stríkis et al., 2018; Wendt et al., 2019). An intensification of the HC associated with conditions similar to La Niña (northward migrations of both the ITCZ and the SH westerlies) was found in reconstructions for the MH (McGee et al., 2014; Mollier-Vogel et al., 2019). Changes in insolation from the mid to late Holocene favoured a southward migration in the position of the ITCZ and the descending branch of the HC in the NH, approaching its current width and position (Wirth et al., 2013; Thatcher et al., 2020). Tree ring chronologies from the NH mid-latitudes over the last 800 years show that the northern edge of the HC tended to migrate southward during positive phases of ENSO and PDV, with northward shifts during negative phases (Alfaro-Sánchez et al., 2018). Between 1400 and 1850 CE the HC over both hemispheres and the ITCZ were displaced southward, consistent with occurrence of drought conditions in several NH regions (Wirth et al., 2013; Burn and Palmer, 2014; Lechleitner et al., 2017; Alfaro-Sánchez et al., 2018; Flores-Aqueveque et al., 2020). Moreover, several proxy records showed not only inter-hemispheric shifts in the ITCZ but a contraction of the tropical belt during 1400–1850 CE, which followed an expansion during 950–1250 CE (Denniston et al., 2016; Griffiths et al., 2016).

From centennial-scale reanalyses, Liu et al. (2012) and D’Agostino and Lionello (2017) found divergent results on HC extent over the last 150 years, although with unanimity upon an intensification of the SH HC. A substantial discrepancy between HC characteristics in centennial-scale reanalyses and in ERA-Interim (D’Agostino and Lionello, 2017) since 1979 yields significant questions regarding their ability to capture changes in HC behaviour. Taken together with the existence of apparent non-climatic artefacts in the datasets (Nguyen et al., 2015), this implies low confidence in changes in the extent and intensity of HC derived from centennial-scale reanalyses. However, using multiple observational datasets and centennial-scale reanalyses, Bronnimann et al. (2015) identified a southward shift in the NH HC edge from 1945 to 1980 of about 0.25° latitude per decade, consistent with observed changes in global land monsoon precipitation (Section 2.3.1.4.2).

Since AR5 several studies based upon a range of metrics and different reanalyses products have suggested that the annual mean HC extent has shifted poleward at an approximate rate of 0.1°–0.5° latitude per decade over the last about 40 years (Allen and Kovilakam, 2017; Davis and Birner, 2017; Grise et al., 2018; Staten et al., 2018, 2020; Studholme and Gulev, 2018; Grise and Davis, 2020). The observed widening of the annual mean HC, revealed by a variety of metrics, is primarily due to poleward shift of the Northern Hemisphere HC. There have been stronger upward trends in the NH extent of HC after 1992 (Figure 2.17a). The estimated magnitude of the recent changes based on modern-era reanalyses is not as large as that in AR5, due to apparent biases in older-generation reanalyses (Grise et al., 2019). Moreover, large interannual variability leads to uncertainties in estimates of long-term changes (Nguyen et al., 2013; Garfinkel et al., 2015b; Seviour et al., 2018; Staten et al., 2018), particularly for the NH given its zonal asymmetries (Staten et al., 2020; Wang et al., 2020). These large-scale features of the HC based on reanalyses agree with estimates revealed from the Integrated Global Radiosonde Archive (IGRA) during 1979–2012 (Lucas and Nguyen, 2015; Mathew et al., 2016). Recent trends based on reanalyses indicate a larger seasonal widening in the HC for summer and autumn in each hemisphere, although the magnitude of changes in HC extent is strongly dependent on dataset and metrics used (Grise et al., 2018; Y. Hu et al., 2018; Staten et al., 2018). The shifts in the HC position were accompanied by a narrowing ITCZ over the Atlantic and Pacific basins, with no significant change in its location and increases in the precipitation intensity (Byrne et al., 2018).

Figure 2.17 | Time series of the annual mean Northern Hemisphere (NH, top curves) and Southern Hemisphere (SH, bottom curves) Hadley cell extent (a) and Hadley cell intensity (b) since 1979. Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

Trends in the HC intensity since 1979 differ between reanalyses, although there is a tendency toward HC intensification (Figure 2.17b; Nguyen et al., 2013; Chen et al., 2014; D’Agostino and Lionello, 2017; R. Huang et al., 2019), which is more marked in the NH than the SH (Studholme and Gulev, 2018). However, the ability of reanalyses to represent the HC strength has been questioned due to inaccurate representation of latent heating distribution, which is directly related to tropical convection and influences the HC dynamics (Chemke and Polvani, 2019; Mathew and Kumar, 2019).

Paleo evidence during the LGM indicates a weaker WC over the Indian Ocean (DiNezio et al., 2018; Windler et al., 2019) with a stronger Pacific WC (DiNezio and Tierney, 2013). During the Holocene, a transition from a strong WC located more westward during the Early-to-Mid Holocene towards a weak and eastward shifted WC during the late Holocene was inferred from proxy records from the Pacific Warm Pool and South East Asia (Barr et al., 2019; Dang et al., 2020; Griffiths et al., 2020), in concurrence with changes in ENSO activity (Section 2.4.2). Reconstructions for the CE showed weakened WC during 1000–1250 and since 1850, with an intensified circulation during 1500–1850 CE (Xu et al., 2016; Deng et al., 2017).

Considering instrumental records, there is considerable interdecadal variability in the strength of the WC, resulting in time-period dependent magnitude and even sign of trends (Carilli et al., 2015; Bordbar et al., 2017; Hou et al., 2018), with some studies reporting weakening over the 20th century (e.g., Power and Kociuba, 2011; Liu et al., 2019), while others reported strengthening (Z. Li et al., 2020), particularly over the last 30–40 years (e.g., Hu et al., 2013; L’Heureux et al., 2013; Yim et al., 2017). Based on estimation of changes in mid-tropospheric velocity from changes in observed cloud cover, Bellomo and Clement (2015) suggest a weakening and eastward shift of the WC over 1920–2010, however the robustness of this signal is questionable due to high uncertainty in the ship-reported cloud data used before 1954. Using centennial-scale 20CR reanalysis Tseng et al. (2019) showed that the vertical westerly wind shear over the western Pacific does not indicate any long-term change during 1900–1980, but shows a marked increase since the 1980s that is not present in ERA-Interim and JRA-55, again calling into question the ability of centennial-scale reanalyses to capture tropical circulation changes. Recent strengthening together with a westward shift of the WC (Bayr et al., 2014; Ma and Zhou, 2016) was identified across several reanalysis products and observational datasets, and using different metrics for quantifying WC. Nevertheless, satellite observations of precipitation and analyses of upper tropospheric humidity suggest substantially weaker strengthening of the WC than implied by reanalyses (Chung et al., 2019). This recent strengthening in the WC is associated with enhanced precipitation in the tropical western Pacific, anomalous westerlies in the upper troposphere, strengthened downwelling in the central and eastern tropical Pacific, and anomalous surface easterlies in the western and central tropical Pacific (Dong and Lu, 2013; McGregor et al., 2014; Choi et al., 2016). Positive trends in sea level pressure over the eastern Pacific and concurrent negative trends over the Indonesian region result in a pattern implying a shift towards a La Niña-like WC regime, with strengthening of the Pacific Trade Winds mainly over 1979–2012 (L’Heureux et al., 2013; England et al., 2014; Sohn et al., 2016; Zhao and Allen, 2019). Seasonal assessment of the WC showed significant changes in the vertical westerly wind shear over the Pacific during the austral summer and autumn implying a strengthening (Clem et al., 2017).

In summary, there has been a likely widening of the Hadley circulation since the 1980s, mostly due to its extension in the NH, although there is only medium confidence in the extent of the changes. This has been accompanied by a strengthening of the Hadley circulation, particularly in the NH (medium confidence). There is low confidence in the estimation of long-term trends in the strength of the Walker circulation, which are time period dependent and subject to dataset uncertainties. Trends since 1980 are better characterized and consistent with avery likely strengthening that resembles a La Niña-like Walker circulation and a westward shift of the Walker circulation, although with medium confidence in the magnitude of the changes, arising from the differences between satellite observations and reanalysis products.

The AR5 reported a weakening of the global monsoon (GM) circulation as well as a decrease of global land monsoon rainfall over the second half of the 20th century. Nevertheless, there was low confidence in the observed circulation trends due to uncertainties in reanalysis products and in the definition of the monsoon area. From a paleo perspective, AR5 only assessed regional monsoon changes.

New research based on high-resolution proxies reinforces previous findings on the influence of orbital cycles on GM variability on millennial time scales. The intensity of the monsoon systems is generally out of phase between hemispheres, being associated with the precession cycle (about 21–23 kyr) (An et al., 2015; P.X. Wang et al., 2017; Seth et al., 2019), with intensified NH monsoon systems during precession minima (Toucanne et al., 2015; Wagner et al., 2019). The eccentricity forcing (about 100 kyr cycle) shows stronger GM during interglacial periods (P.X. Wang et al., 2014, 2017; An et al., 2015; Mohtadi et al., 2016). Changes in obliquity (about 41 kyr cycle) modify the strength of monsoon systems, with increased summer monsoon rainfall when obliquity is maximal (Y. Liu et al., 2015b; Mohtadi et al., 2016). Millennial scale variability in GM during the LDT was also linked to the occurrences of Heinrich stadials, resulting in weakened NH monsoons and intensified SH monsoons (An et al., 2015; P.X. Wang et al., 2017; Margari et al., 2020).

An intensification of the NH monsoons in the early to mid-Holocene with increased precipitation and regional expansions of rainfall areas identified through a variety of proxy records is shown by Biasutti et al. (2018) and P.X. Wang et al. (2017). The response for the SH monsoons during this period indicates a weakening in both summer and winter precipitation (P.X. Wang et al., 2014, 2017; Sachs et al., 2018). A decline in GM precipitation and a retraction of the northern fringes of monsoon areas was inferred from the mid-Holocene onwards, with some regions experiencing wetter conditions during the mid to late Holocene compared with present and a strengthening of the SH monsoons (P.X. Wang et al., 2014, 2017; Sachs et al., 2018). For the CE, GM reconstructions exhibit inter-hemispheric contrast during the period 950–1250 CE, with intensified NH monsoons and weakened SH monsoons, and the opposite pattern during 1400–1850 CE (P.X. Wang et al., 2014; An et al., 2015).

Direct observations highlight that the GM land precipitation, particularly over the NH, experienced a slight increase from 1900 through the early 1950s, followed by an overall decrease from the 1950s to the 1980s, and then an increase to present (Kitoh et al., 2013; B. Wang et al., 2018, 2021; X. Huang et al., 2019b). This highlights the existence of multi-decadal variations in the NH monsoon circulation patterns and precipitation intensity (Wang et al., 2013; P.X. Wang et al., 2014, 2017; Monerie et al., 2019). An overall increase in monsoon precipitation during extended boreal summer (JJAS) over the NH since 1979 is revealed by GPCP (Deng et al., 2018; Han et al., 2019) and CMAP for 1980–2010 (Jiang et al., 2016). SH summer monsoon behaviour is dominated by strong interannual variability and large regional differences (Kitoh et al., 2013; Lin et al., 2014; Jiang et al., 2016; Kamae et al., 2017; Deng et al., 2018; Han et al., 2019), with no significant trends reported by GPCP and CMAP (Deng et al., 2018). Uncertainty predominantly arises from the observed increase in tropical precipitation seasonality (Feng et al., 2013) and the estimation of GM precipitation over the ocean areas, leading to a large apparent spread across datasets (Kitoh et al., 2013; Kamae et al., 2017).

In summary, observed trends during the last century indicate that the GM precipitation decline reported in AR5 has reversed since the 1980s, with a likely increase mainly due to a significant positive trend in the NH summer monsoon precipitation (medium confidence). However, GM precipitation has exhibited large multi-decadal variability over the last century, creating low confidence in the existence of centennial-length trends in the instrumental record. Proxy reconstructions show a likely NH monsoons weakening since the mid-Holocene, with opposite behaviour for the SH monsoons.

The AR5 reported a likely poleward shift of storm tracks and jet streams since the 1970s from different datasets, variables and approaches. These trends were consistent with the HC widening and the poleward shifting of the circulation features since the 1970s. There was low confidence in any large-scale change in blocking.

Proxy records consistent with modelling results imply a southward shift of the storm tracks over the North Atlantic during the LGM (Raible et al., 2021). A variety of proxies are available for the changes in the position of the extratropical jets/westerlies during the Holocene. Recent syntheses of moisture-sensitive proxy records indicate drier-than-present conditions over mid-latitudes of western North America (Hermann et al., 2018; Liefert and Shuman, 2020) during the MH, which together with a weakened Aleutian Low (Bailey et al., 2018) implies that the winter North Pacific jetstream was shifted northward. A synthesis of lines of evidence from the SH indicates that the westerly winds were stronger over 14–5 ka, followed by regional asymmetry after 5 ka (Fletcher and Moreno, 2012). There is no consensus on the shifts of the SH westerlies with some studies implying poleward migrations (Lamy et al., 2010; Voigt et al., 2015; Turney et al., 2017; Anderson et al., 2018) and others suggesting an equatorward shift (Kaplan et al., 2016) in the MH.

During 950–1400 CE, hydroclimate indicators suggest a northward shift of Pacific storm tracks over North America (McCabe-Glynn et al., 2013; Steinman et al., 2014) which was comparable in magnitude to that over 1979–2015 (J. Wang et al., 2017a). Storm tracks over the North Atlantic-European sector shifted northward as indicated by multi-proxy indicators over the North Atlantic (Wirth et al., 2013; Orme et al., 2017) and Mediterranean (Roberts et al., 2012). Reconstructed westerly winds in the SH suggest a poleward shift (Lamy et al., 2010; Schimpf et al., 2011; Goodwin et al., 2014; Koffman et al., 2014; Moreno et al., 2018), with latitudinal change comparable to that during recent decades (Swart and Fyfe, 2012; Manney and Hegglin, 2018).

Multiple reanalyses show that since 1979 the subtropical jet wind speeds have generally increased in winter and decreased in summer in both hemispheres, but the trends are regionally dependent (Pena-Ortiz et al., 2013; Manney and Hegglin, 2018; S.H. Lee et al., 2019). Over NH mid-latitudes, the summer zonal wind speeds have weakened in the mid-troposphere (Francis and Vavrus, 2012; Coumou et al., 2014, 2015; Haimberger and Mayer, 2017). Meanwhile there are indications of enhanced jetstream meandering in boreal autumn at the hemispheric scale (Francis and Vavrus, 2015; Di Capua and Coumou, 2016), whereas the regional arrangement of meandering depends on the background atmospheric state (Cohen et al., 2020). These meandering trends, however, are sensitive to the metrics used (Screen and Simmonds, 2013; Hassanzadeh et al., 2014; Cattiaux et al., 2016; Vavrus, 2018). Hypothesized links to Arctic warming are assessed in Cross-Chapter Box 10.1.

Multiple reanalyses and radiosonde observations show an increasing number of extratropical cyclones over the NH since the 1950s (Chang and Yau, 2016; X.L. Wang et al., 2016). The positive trends are generally consistent among reanalyses since 1979, though with considerable spread (Tilinina et al., 2013; X.L. Wang et al., 2016). In recent decades the number of deep extratropical cyclones has increased over the SH (Section 8.3.2.8.1 and Figure 8.12; Reboita et al., 2015; X.L. Wang et al., 2016), while the number of deep cyclones has decreased in the NH in both winter and summer (Neu et al., 2013; Coumou et al., 2015; Chang et al., 2016; J. Wang et al., 2017a; Gertler and O’Gorman, 2019). The regional changes for different intensity extratropical cyclones are assessed in Section 8.3.2.8.1. The assessment of trends is complicated by strong interannual to decadal variability, sensitivity to dataset choice and resolution (Tilinina et al., 2013; Lucas et al., 2014; X.L. Wang et al., 2016; Pepler et al., 2018; Rohrer et al., 2018) and cyclone identification/tracking methods (Neu et al., 2013; Grieger et al., 2018). Thus there is overall low confidence for recent changes in global extratropical storm tracks.

A consistent poleward shift of the tropospheric extratropical jets since 1979 is reported by multiple reanalyses (Figure 2.18; Davis and Rosenlof, 2012; Davis and Birner, 2013; Pena-Ortiz et al., 2013; Manney and Hegglin, 2018), and radiosonde winds (Allen et al., 2012). This is generally consistent with the previously reported shifts retrieved from satellite temperature observations (Fu and Lin, 2011; Davis and Rosenlof, 2012). After the 1960s the magnitude of meridional shifts in extratropical jets over both the North Atlantic and North Pacific in August is enhanced compared to multi-century variability (Trouet et al., 2018). Despite some regional differences (Woollings et al., 2014; Norris et al., 2016; J. Wang et al., 2017a; Xue and Zhang, 2017; Ma and Zhang, 2018; Melamed-Turkish et al., 2018), overall poleward deflection of storm tracks in boreal winter over both the North Atlantic and the North Pacific was identified during 1979–2010 (Tilinina et al., 2013). Over the SH extra-tropics there is a similarly robust poleward shift in the polar jet since 1979 (Pena-Ortiz et al., 2013; Manney and Hegglin, 2018; WMO, 2018), although after 2000 the December–January–February (DJF) tendency to poleward shift of the SH jet stream position ceased (Banerjee et al., 2020). The general poleward movement in midlatitude jet streams (Lucas et al., 2014) is consistent with the expansion of the tropical circulation (Section 2.3.1.4.1). The changes of extratropical jets and westerlies are also related to the annular modes of variability (Section 2.4 and Annex IV).

Figure 2.18 | Trends in ERA5 zonal-mean zonal wind speed. Shown are (a) DJF (December–January–February); (b) MAM (March–April–May); (c) JJA (June–July–August); and (d) SON (September–October–November). Climatological zonal winds during the data period are shown in solid contour lines for westerly winds and in dashed lines for easterly. Trends are calculated using OLS regression with significance assessed following AR(1) adjustment after Santer et al. (2008) (‘×’ marks denote non-significant trends). Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

Robust trends in blocking have only been found in certain regions and specific seasons during recent decades. Increases in blocking frequency have occurred over low-latitude regions in the North Atlantic in boreal winter (Davini et al., 2012), the South Atlantic in austral summer (Dennison et al., 2016) and the southern Indian Ocean in austral spring (Schemm, 2018). Over the subpolar North Atlantic sustained periods of positive Greenland blocking were identified during 1870–1900 and from the late 1990s to 2015 (Hanna et al., 2015). Further analysis of association of Greenland blocking with the NAM is provided in Section 2.4.1.1. Meanwhile, a reduced blocking frequency has been found over winter in Siberia (Davini et al., 2012) and the south-western Pacific in austral spring (Schemm, 2018). Over eastern European Russia and western Siberia (40°E–100°E) a tendency towards longer blocking events was reported by Luo et al. (2016) for 2000–2013 and by Tyrlis et al. (2020) for 1979–2017. Inter-annual variance in the number of blocking events over the SH (Oliveira et al., 2014) and North Atlantic (Kim and Ha, 2015) has enhanced. Blocking events and their trends are sensitive to choice of datasets, calculation periods and methods (Cheung et al., 2013; Barnes et al., 2014; Pepler et al., 2018; Rohrer et al., 2018; Woollings et al., 2018b; Kononova and Lupo, 2020). As a result, hemispheric and global trends in blocking frequency have overall low confidence.

In summary, the total number of extratropical cyclones has likely increased since the 1980s in the NH (low confidence), but with fewer deep cyclones particularly in summer. The number of strong extratropical cyclones has likely increased in the SH (medium confidence). The extratropical jets and cyclone tracks have likely been shifting poleward in both hemispheres since the 1980s with marked seasonality in trends (medium confidence). There is low confidence in shifting of extratropical jets in the NH during the mid-Holocene and over 950–1400 CE to latitudes that likely were similar to those since 1979. There is low confidence in observed global-scale changes in the occurrence of blocking events.

The AR5 concluded that surface winds over land had generally weakened. The confidence for both land and ocean surface wind trends was lowowing to uncertainties in datasets and measures used. Sea level pressure (SLP) was assessed to have likely decreased from 1979–2012 over the tropical Atlantic and increased over large regions of the Pacific and South Atlantic, but trends were sensitive to the period analysed.

Terrestrial in situ wind datasets have been updated and the quality-control procedures have been improved, with particular attention to homogeneity and to better retaining true extreme values (Dunn et al., 2012, 2014, 2016). Global mean land wind speed (excluding Australia) from HadISD for 1979–2018 shows a reduction (stilling) of 0.063 m s^–1 per decade (Azorin-Molina et al., 2019). Trends are broadly insensitive to the subsets of stations used. Although the meteorological stations are unevenly distributed worldwide and sparse in South America and Africa, the majority exhibit stilling particularly in the NH (Figure 2.19). Regionally, strong decreasing trends are reported in central Asia and North America (–0.106 and –0.084 m s^–1 per decade respectively) during 1979–2018 (McVicar et al., 2012; Vautard et al., 2012; J. Wu et al., 2018; Azorin-Molina et al., 2019). This stilling tendency has reversed after 2010 and the global mean surface winds have strengthened (Zeng et al., 2019b; Azorin-Molina et al., 2020), although the robustness of this reversal is unclear given the short period and interannual variability (Kousari et al., 2013; Kim and Paik, 2015; Azorin-Molina et al., 2019).

Figure 2.19 | Trends in surface wind speed. (a) Station observed winds from the integrated surface database (HadISD v2.0.2.2017f); (b) Cross-Calibrated Multi-Platform wind product; (c) ERA5; and (d) wind speed from the Objectively Analyzed Air-Sea Heat Fluxes dataset, release 3 (OAFLUX, release 3). White areas indicate incomplete or missing data. Trends are calculated using OLS regression with significance assessed following AR(1) adjustment after Santer et al. (2008); ‘×’ marks denote non-significant trends. Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

Over the ocean, datasets demonstrate considerable disagreement in surface wind speed trends and spatial features (Kent et al., 2013). Global ocean surface winds from NOCv2.0 demonstrate upward trends of about 0.11 m s^–1 per decade (1979–2015) with somewhat smaller trends from WASwind for 1979–2011 (Azorin-Molina et al., 2017, 2019). The trends are consistent until 1998, but diverge thereafter. Both ERA5 and JRA-55 reanalyses show consistently increasing global marine wind speeds over 1979–2015, though flattening since 2000, whereas MERRA-2 agrees until 1998, but then exhibits increased variability and an overall decrease in the last two decades (Azorin-Molina et al., 2019). This agrees with estimates by Sharmar et al. (2021) showing upward ocean wind trends from 1979 to 2000 which are consistent in ERA-Interim, ERA5 and MERRA-2, but disagree with CFSR trends for the same period. Over 2000–2019 all reanalyses show diverging tendencies. An updated multiplatform satellite database (comprising data from altimeters, radiometers, and scatterometers) from 1985–2018 shows small increases in mean wind speed over the global ocean, with the largest increase observed in the Southern Ocean (Young and Ribal, 2019), consistent with signals in ERA-Interim, ERA5 and MERRA-2 (Sharmar et al., 2021). Overall, most products suggest positive trends over the Southern Ocean, western North Atlantic and the tropical eastern Pacific since the early 1980s.

The modern era reanalyses exhibit SLP increases over the SH subtropics with stronger increases in austral winter over 1979–2018. Over the NH, SLP increased over the mid-latitude Pacific in boreal winter and decreased over the eastern subtropical and mid-latitude North Atlantic in boreal summer. Discrepancies in the low-frequency variations during the first half of the 20th century exist in the centennial-scale reanalysis products (Befort et al., 2016). Overall, modern reanalysis datasets support the AR5 conclusion that there is no clear signal for trends in the strength and position of the permanent and quasi-permanent pressure centres of action since the 1950s. Instead, they highlight multi-decadal variations. Large-scale SLP is strongly associated with the changes in modes of variability (Section 2.4 and Annex IV).

In summary, since the 1970s a worldwide weakening of surface wind has likely occurred over land, particularly marked in the NH, with low confidence in a recent partial recovery since around 2010. Differences between available wind speed estimates lead to low confidence in trends over the global ocean as a whole but with most estimates showing strengthening globally over 1980–2000 and over the last four decades in the Southern Ocean, western North Atlantic and the tropical eastern Pacific.

The AR5 assessed changes in the polar vortices and reported a likely decrease in the lower-stratospheric geopotential heights over Antarctica in spring and summer at least since 1979.

Multiple definitions for the polar vortex strength and sudden stratospheric warming (SSW) events have been proposed and compared (Butler et al., 2015; Palmeiro et al., 2015; Waugh et al., 2017; Butler and Gerber, 2018), and new techniques identifying daily vortex patterns and SSWs have been developed (D.M. Mitchell et al., 2013; Kretschmer et al., 2018). Errors in reanalysis stratospheric winds were assessed and discrepancies in stratospheric atmospheric circulation and temperatures between reanalyses, satellites and radiosondes have been reported (D.M. Mitchell et al., 2013; Duruisseau et al., 2017).

The northern stratospheric polar vortex has varied intra-seasonally and with altitude during recent decades. Multiple reanalysis and radiosonde datasets show that the midwinter lower stratospheric geopotential height (150 hPa) over the polar region north of 60°N has increased significantly since the early 1980s (Bohlinger et al., 2014; Garfinkel et al., 2017). This signal extends to the middle and upper stratosphere. In January-February zonal winds north of 60°N at 10 hPa have been weakening (Kim et al., 2014; Kretschmer et al., 2018). Daily atmospheric circulation patterns over the northern polar stratosphere exhibit a decreasing frequency of strong vortex events and commensurate increase in more-persistent weak events, which largely explains the observed significant weakening of the vortex during 1979–2015 (Kretschmer et al., 2018). The northern polar vortex has weakened in early winter but strengthened during late winter (Bohlinger et al., 2014; Garfinkel et al., 2015a, 2017; Ivy et al., 2016; Seviour, 2017; Kretschmer et al., 2018). In the middle and upper stratosphere, a strengthening trend of the northern polar vortex during DJF has occurred since 1998, contrasting the weakening trend beforehand (D. Hu et al., 2018). The position of the polar vortex also has long-term variations, exhibiting a persistent shift toward Northern Siberia and away from North America in February over the period 1979–2015 (Zhang et al., 2016; J. Zhang et al., 2018). Multiple measures show similar location changes (Seviour, 2017).

Sudden stratospheric warming (SSW), a phenomenon of rapid stratospheric air temperature increases (sometimes by more than 50°C in 1–2 days), is tightly associated with the reversal of upper stratospheric zonal winds, and a resulting collapse or substantial weakening of the stratospheric polar vortex (Butler et al., 2015; Butler and Gerber, 2018) and on average occurs approximately 6 times per decade in the NH winter (Charlton et al., 2007; Butler et al., 2015). The SSW record from all modern reanalyses is very consistent. There is a higher occurrence of major midwinter SSWs in the 1980s and 2000s with no SSW events during 1990–1997 (Reichler et al., 2012; Butler et al., 2015). An assessment of multi-decadal variability and change in SSW events is sensitive to both chosen metric and methods (Palmeiro et al., 2015). Due to the lack of assimilation of upper air data, the centennial-scale reanalyses do not capture SSW events, even for the most recent decades (Butler et al., 2015, 2017) and hence cannot inform on earlier behaviour. There has been considerably less study of trends in the SH stratosphere polar vortex strength despite the interest in the ozone hole and the potential impact of the SH stratosphere polar vortex strength on it. The occurrence of SSW events in the SH is not as frequent as in the NH, with only 3 documented events in the last 40 years (Shen et al., 2020).

In summary, it is likely that the northern lower stratospheric polar vortex has weakened since the 1980s in midwinter, and its location has shifted more frequently toward the Eurasian continent. The short record and substantial decadal variability yields low confidence in any trends in the occurrence of SSW events in the NH winter and such events in the SH are rare.

The AR5 reported that the annual mean Arctic sea-ice extent (SIE)very likely decreased by 3.5–4.1% per decade between 1979 and 2012 with the summer sea-ice minimum (perennial sea ice)very likely decreasing by 9.4–13.6% per decade. This was confirmed by SROCC reporting the strongest reductions in September (12.8 ± 2.3% per decade; 1979–2018) and stating that these changes were likely unprecedented in at least 1 kyr (medium confidence). The spatial extent had decreased in all seasons, with the largest decrease for September (high confidence). The AR5 reported also that the average winter sea ice thickness within the Arctic Basin had likely decreased by between 1.3 m and 2.3 m from 1980 to 2008 (high confidence), consistent with the decline in multi-year and perennial ice extent. The SROCC stated further that it was virtually certain that Arctic sea ice had thinned, concurrent with a shift to younger ice. Lower sea ice volume in 2010–2012 compared to 2003–2008 was documented in AR5 (medium confidence). There was high confidence that, where the sea ice thickness had decreased, the sea-ice drift speed had increased.

Proxy records are used in combination with modelling to assess Arctic paleo sea ice conditions to the extent possible. For the Pliocene, limitedproxyevidence of a reduced sea ice cover compared to ‘modern’ winter conditions (Knies et al., 2014; Clotten et al., 2018) and model simulations of a largely ice-free Arctic Ocean during summer (Howell et al., 2016; Feng et al., 2019; F. Li et al., 2020) imply medium confidence that the Arctic Ocean was seasonally ice covered. Over the LIG, sparse proxy reconstructions (Stein et al., 2017; Kremer et al., 2018) and proxy evidence from marine sediments (Kageyama et al., 2021b) provide medium confidence of perennial sea ice cover.

Over the past 13 kyr proxy records suggest extensive sea-ice coverage during the Younger Dryas (at the end of the LDT), followed by a decrease in sea ice coverage during the Early Holocene, and increasing sea-ice coverage from the MH to the mid-15th century (De Vernal et al., 2013; Belt et al., 2015; Cabedo-Sanz et al., 2016; Armand et al., 2017; Belt, 2018). There is limited evidence that the Canadian Arctic had less multiyear sea ice during the Early Holocene than today (Spolaor et al., 2016). For more regional details on paleo arctic sea ice see Section 9.3.1.1.

Pan-Arctic SIE conditions (annual means and late summer) during the last decade were unprecedented since at least 1850 (Figure 2.20a; Walsh et al., 2017, 2019; Brennan et al., 2020), while, as reported in SROCC, there remains medium confidence that the September (late summer) Arctic sea ice loss during the last decade was unprecedented during the past 1 kyr. Sea-ice charts since 1850 (Walsh et al., 2017, 2019) suggest that there was no significant trend before the 1990s, but the uncertainty of these estimates is large and could mask a trend, a possibility illustrated by Brennan et al. (2020), who found a loss of Arctic sea ice between 1910 and 1940 in an estimate based on a data assimilation approach.

Figure 2.20 | Changes in Arctic and Antarctic sea ice area. (a) Three time series of Arctic sea-ice area (SIA) for March and September from 1979 to 2020 (passive microwave satellite era). In addition, the range of SIA from 1850–1978 is indicated by the vertical bar to the left. (b) Three time series of Antarctic sea ice area for September and February (1979–2020). In both (a) and (b), decadal means for the three series for the first and most recent decades of observations are shown by horizontal lines in grey (1979–1988) and black (2010–2019). SIA values have been calculated from sea ice concentration fields. Available data for 2020 (OSISAF) is shown in both (a) and (b). Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

There has been a continuing decline in SIE and Arctic sea ice area (SIA) in recent years (Figure 2.20a). To reduce grid-geometry associated biases and uncertainties (Notz, 2014; Ivanova et al., 2016; Meier and Stewart, 2019) SIA is used in addition to, or instead of SIE herein (see also section 9.3.1). A record-low Arctic SIA since the start of the satellite era (1979) occurred in September 2012 (Figure 2.20a). Decadal SIA means based on the average of three different satellite products decreased from 6.23 to 3.76 million km² for September and 14.52 to 13.42 million km² for March SIA (Figure 2.20a). Initial SIA data for 2020 (OSISAF) are within the range of these recent decadal means or slightly below (Figure 2.20a). SIA has declined since 1979 across the seasonal cycle (Figure 9.13). Most of this decline in SIA has occurred after 2000, and is superimposed by substantial interannual variability. The sharp decline in Arctic summer SIA coincides with earlier surface melt onset (Mortin et al., 2016; Bliss et al., 2017), later freeze-up, and thus a longer ice retreat and open water period (Stammerjohn et al., 2012; Parkinson, 2014; Peng et al., 2018).

Over the past two decades, first-year sea ice has become more dominant and the oldest multiyear ice (older than 4 years) which in March 1985 made up 33% of the Arctic sea-ice cover, has nearly disappeared, making up 1.2% in March 2019 (Perovich et al., 2020). The loss of older ice is indicative of a thinning overall of ice cover (Tschudi et al., 2016), but also the remaining older ice has become thinner (E. Hansen et al., 2013). Since in situ ice thickness measurements are sparse, information about ice thickness is mainly based on airborne and satellite surveys. Records from a combination of different platforms show for the central and western Arctic Ocean (Arctic Ocean north of Canada and Alaska) negative trends since the mid-1970s (Lindsay and Schweiger, 2015; Kwok, 2018), with a particularly rapid decline during the 2000s, which coincided with a large loss of multiyear sea ice. Direct observations from 2004 and 2017 indicate a decrease of modal ice thickness in the Arctic Ocean north of Greenland by 0.75 m, but with little thinning between 2014 and 2017 (Haas et al., 2017). This agrees with data based on satellite altimetry and airborne observations, showing no discernible thickness trend since 2010 (Kwok and Cunningham, 2015; Kwok, 2018; Kwok and Kacimi, 2018; see Figure 2.21). However, sea-ice thickness derived from airborne and spaceborne data is still subject to uncertainties imposed by snow loading. For radar altimeters, insufficient penetration of radar signal into the snowpack results in overestimation of ice thickness (e.g., Ricker et al., 2015; King et al., 2018; Nandan et al., 2020). Negative trends in ice thickness since the 1990s are also reported from the Fram Strait in the Greenland Sea, and north of Svalbard (E. Hansen et al., 2013; Renner et al., 2014; King et al., 2018; Rösel et al., 2018; Spreen et al., 2020). Thickness data collected in the Fram Strait originate from ice exported from the interior of the Arctic Basin and are representative of a larger geographical area upstream in the transpolar drift. A reduction of survival rates of sea ice exported from the Siberian shelves by 15% per decade has interrupted the transpolar drift and affected the long-range transport of sea ice (Krumpen et al., 2019). The thinner and on average younger ice has less resistance to dynamic forcing, resulting in a more dynamic ice cover (Hakkinen et al., 2008; Spreen et al., 2011; Vihma et al., 2012; Kwok et al., 2013).

Figure 2.21 | Arctic sea ice thickness changes (means) for autumn (red/dotted red) and winter (blue/dotted blue). Shadings (blue and red) show 1 standard error (S.E.) ranges from the regression analysis of submarine ice thickness and expected uncertainties in satellite ice thickness estimates. Data release area of submarine data ice thickness data is shown in inset. Satellite ice thickness estimates are for the Arctic south of 88°N. Thickness estimates from more localized airborne/ground electromagnetic surveys near the North Pole (diamonds) and from Operation IceBridge (circles) are shown within the context of the larger scale changes in the submarine and satellite records. Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

The SROCC noted the lack of continuous records of snow on sea ice. Nevertheless in recent decades, more snow on sea ice has been observed in the Atlantic sector in the Arctic than in the western Arctic Ocean (Webster et al., 2018). Previously, Warren et al. (1999) showed that over 1954–1991 there were weak trends towards declining snow depth on sea ice in the Pacific sector. Recent observations indicate a substantial thinning of the spring snowpack in the western Arctic (Cavalieri et al., 2012; Brucker and Markus, 2013; Kurtz et al., 2013; Laxon et al., 2013; Webster et al., 2018). In contrast, thick snow over Arctic sea ice in the Atlantic sector north of Svalbard (snow thickness around 0.4 m or more) has been observed in the 1970s and since the 1990s (Rösel et al., 2018), but data are too sparse to detect trends.

In summary, over 1979–2019 Arctic SIA has decreased for all months, with the strongest decrease in summer (very high confidence). Decadal means for SIA decreased from the first to the last decade in that period from 6.23 to 3.76 million km² for September, and from 14.52 to 13.42 million km² for March. Arctic sea ice has become younger, thinner and faster moving (very high confidence). Snow thickness on sea ice has decreased in the western Arctic Ocean (medium confidence). Since the Younger Dryas at the end of the LDT, proxy indicators show that Arctic sea ice has fluctuated on multiple time scales with a decrease in sea ice coverage during the Early Holocene and an increase from the MH to the mid-15th century. Current pan-Arctic sea ice coverage levels (annual mean and late summer) are unprecedentedly low since 1850 (high confidence), and with medium confidence for late summer for at least the past 1 kyr.

The AR5 reported a small but significant increase in the total annual mean Antarctic SIE that was very likely in the range of 1.2–1.8% per decade between 1979 and 2012 (0.13–0.20 million km² per decade) (very high confidence), while SROCC reported that total Antarctic sea ice coverage exhibited no significant trend over the period of satellite observations (1979–2018) (high confidence). The SROCC noted that a significant positive trend in mean annual sea ice cover between 1979 and 2015 had not persisted, due to three consecutive years of below-average sea ice cover (2016–2018). The SROCC stated also that historical Antarctic sea ice data from different sources indicated a decrease in overall Antarctic sea ice cover since the early 1960s, but was too small to be separated from natural variability (high confidence).

There is onlylimited evidence from predominantly regional paleo proxies for the evolution of Southern Ocean sea ice before the instrumental record and estimates are not available for all proxy target periods (Section 9.3.2). Proxies from marine sediments for intervals preceding and following the MPWP indicate open water conditions with less sea ice than modern conditions (Taylor-Silva and Riesselman, 2018; Ishino and Suto, 2020). During the LGM, proxies indicate that austral winter sea ice coverage reached the polar ocean front (e.g., Nair et al., 2019). More recently, sea ice coverage appears to have fluctuated substantially throughout the Holocene (e.g., for the western Amundsen Sea, Lamping et al., 2020). At the beginning of the CE, regional summer sea ice coverage in the north-western Ross Sea was lower than today (Tesi et al., 2020). Crosta et al. (2021) suggest, based on different proxies, four different phases with 7–10 months periods of sea ice occurrence per year in the Antarctic region off Adelie Land during the CE, where each phase was several hundred years long.

More recent sea ice reconstructions are based on diverse sources including whaling records (de La Mare, 1997, 2009; Cotté and Guinet, 2007), old ship logbooks (Ackley et al., 2003; Edinburgh and Day, 2016), and ice core records (Curran et al., 2003; Abram et al., 2010; Sinclair et al., 2014), amongst other methods (e.g., Murphy et al., 2014). These reconstructions, in combination with recent satellite-based observations indicate: (i) a decrease in summer SIE across all Antarctic sectors since the early- to mid-20th century; (ii) a decrease in winter SIE in the East Antarctic and Amundsen-Bellingshausen Seas sectors starting in the 1960s; and (iii) small fluctuations in winter SIE in the Weddell Sea over the 20th century (Hobbs et al., 2016a, b). There are also ice-core indications that the pronounced Ross Sea increase dates back to the mid-1960s (Sinclair et al., 2014; Thomas and Abram, 2016). While there is reasonable broad-scale concurrence across these estimates, the uncertainties are large, there is considerable interannual variability, and reconstructions require further validation (Hobbs et al., 2016a, b). New reconstructions (Thomas et al., 2019) from Antarctic land ice cores show that SIE in the Ross Sea had increased between 1900 and 1990, while the Bellingshausen Sea had experienced a decline in SIE; this dipole pattern is consistent with satellite-based observations from 1979 to 2019 (Parkinson, 2019), but the recent rate of change then has been larger. Records of Antarctic SIE for the late 19th and early 20th centuries (Edinburgh and Day, 2016), show SIE comparable with the satellite era, although with marked spatial heterogeneity (e.g., Thomas et al., 2019).

Early Nimbus satellite visible and infrared imagery from the 1960s (Meier et al., 2013; Gallaher et al., 2014) indicate higher overall SIE compared to 1979–2013 (Hobbs et al., 2016a, b), but with large uncertainties and poorly quantified biases (NA SEM, 2017). The continuous satellite passive-microwave record shows that there was a modest increase in overall Antarctic SIA of 2.5% ± 0.2% per decade (1 standard error over 1979–2015; Comiso et al., 2017). For overall ice coverage and for this period, positive long-term trends were most pronounced during austral autumn advance (Maksym, 2019), being moderate in summer and winter, and lowest in spring (Holland, 2014; Turner et al., 2015; Hobbs et al., 2016a, b; Comiso et al., 2017). Since 2014, overall Antarctic SIE (and SIA) has exhibited major fluctuations from record-high to record-low satellite era extents (Massonnet et al., 2015; Reid and Massom, 2015; Reid et al., 2015; Comiso et al., 2017; Parkinson, 2019). After setting record-high extents each September from 2012 through 2014, Antarctic SIE (and SIA) dipped rapidly in mid-2016 and remained predominantly below average through 2019 (Reid et al., 2020). For the most recent decade of observations (2010–2019), the decadal means of three SIA products (Figure 2.20b) were 2.17 million km² for February and 15.75 million km² for September, respectively. The corresponding levels for the means for the first decade of recordings (1979–1988) were 2.04 million km² for February and 15.39 million km² for September indicating little overall change. Initial SIA data for 2020 (OSISAF) show SIA for September above, and for February slightly below the recent decadal means (Figure 2.20b). The 2020 September level (OSISAF) remains below the levels observed over 2012–2014.

In summary, Antarctic sea ice has experienced both increases and decreases in SIA over 1979–2019, and substantively lower levels since 2016, with only minor differences between decadal means of SIA for the first (for February 2.04 million km², for September 15.39 million km²) and last decades (for February 2.17 million km², for September 15.75 million km²) of satellite observations (high confidence). There remains low confidence in all aspects of Antarctic sea ice prior to the satellite era owing to a paucity of records that are highly regional in nature and often seemingly contradictory.

The AR5 concluded the volume of the Greenland Ice Sheet (GrIS) was reduced compared to present during periods of the past few million years that were globally warmer than present (high confidence). It reported that the GrIS had lost ice during the prior two decades (very high confidence), that the ice loss had occurred in several sectors, and that high rates of mass loss had both expanded to higher elevations (high confidence) and very likely accelerated since 1992. The SROCC concluded that it was extremely likely that ice loss increased through the early 21st century. The SROCC also found that summer melting rate had increased since the 1990s to a rate unprecedented over the last 350 years (very high confidence), being two to five times greater than the pre-industrial rates (medium confidence).

Details of the history of the GrIS fluctuations during warm interglacials continue to be elucidated. Oscillations over the past 7.5 Myr, including the Pliocene and through the glacial – interglacial cycles of the Pleistocene are not well-constrained, but most studies indicate that Greenland was at least partially glaciated over this time with extended periods when it was predominantly deglaciated (Bierman et al., 2016; Schaefer et al., 2016). Geological evidence and modelling studies suggest periods of glacial intensification during the Pliocene at 4.9 Myr, 4.0 Myr, 3.6 Myr and 3.3 Myr (De Schepper et al., 2014; Bierman et al., 2016; Bachem et al., 2017). Retreat of the GrIS occurred during the MPWP and GrIS extent was reduced compared to today with some studies suggesting that the ice sheet was limited to the highest elevations (De Schepper et al., 2014; Koenig et al., 2015; Haywood et al., 2016; Blake-Mizen et al., 2019). There is apparent glacial intensification following the MPWP, 2.75–2.72 Myr (Nielsen and Kuijpers, 2013; De Schepper et al., 2014; Blake-Mizen et al., 2019; Knutz et al., 2019). Several studies agree that during the LIG the total GrIS extent was likely less than present day (Section 9.4.1, Figure 9.17) with the total mass loss ranging from 0.3–6.2 m sea level equivalent (SLE), although timing and magnitude of this mass loss are not well constrained (Helsen et al., 2013; Stone et al., 2013; Vasskog et al., 2015; Goelzer et al., 2016; Sinclair et al., 2016; Yau et al., 2016; Clark et al., 2020). During the LGM, the GrIS reached a peak ice volume greater than present (2–5 m SLE), as revealed by the limited number of available geological records (Simpson et al., 2009; Lecavalier et al., 2014; Batchelor et al., 2019).

Recent studies of marine and lake sediments, glacier ice, and geomorphic features show that the GrIS retreated rapidly during the early Holocene but halted periodically, with a complex ice-margin chronology (Carlson et al., 2014; Larsen et al., 2014, 2015; Young and Briner, 2015; Briner et al., 2016; Young et al., 2020). It is probable that its total volume during 8–3 ka was smaller than today (Larsen et al., 2015; Young and Briner, 2015; Briner et al., 2016), but uncertainties exist regarding precisely when the minimum MH extent and volume was reached, due to uncertainties in reconstructions. The GrIS then re-advanced reaching its maximum extent in most places during 1450–1850 CE, although the timing and extent of this maximum differed by sector (Larsen et al., 2015; Briner et al., 2016).

Greenland-wide estimates of mass change based on direct observations were limited prior to 1992 at the time of AR5 (Kjeldsen et al., 2015). Combined records based on airborne observations, model-based estimates and geodetic approaches indicate an average mass loss of 75 ± 29.4 Gt yr^–1 for 1900–1983 (Kjeldsen et al., 2015). Integration of proxies and modelling indicates that the last time the rate of mass loss of the GrIS was plausibly similar to 20th century rates was during the early Holocene (Buizert et al., 2018; Briner et al., 2020).

Since AR5, a combination of remote sensing, in situ observations and modelling has provided new insights regarding surface processes and their contribution to recent GrIS mass changes (AMAP, 2017; van den Broeke et al., 2017; Bamber et al., 2018; Mouginot et al., 2019; IMBIE Consortium, 2020; Khan et al., 2020). Estimates of total ice loss during the post-1850 period (Kjeldsen et al., 2015) and recent observations show that the rate of loss has increased since the beginning of the 21st century (IMBIE Consortium, 2020; Sasgen et al., 2020; Velicogna et al., 2020) (Section 9.4.1.1 and Figures 2.24 and 9.17).

The GrIS lost 4890 [4140 to 5640] Gt (SLE 13.5 [11.4 to 15.6] mm) of ice between 1992 and 2020 (Section 9.4.1 and Figure 2.24; IMBIE Consortium, 2020). The ice sheet was close to mass balance in the 1990s, but increases in mass loss have occurred since (Bamber et al., 2018; WCRP Global Sea Level Budget Group, 2018; Mouginot et al., 2019; IMBIE Consortium, 2020). The rate of ice-sheet (including peripheral glaciers) mass loss rose from 120 [70 to 170] Gt yr^–1 (SLE 0.33 [0.18 to 0.47] mm yr^–1) in 1901–1990 to 330 [290 to 370] Gt yr^–1 (SLE 0.91 [0.79 to 1.02] mm yr^–1) for 2006–2018 (Section 9.4.1, Table 9.5).

In summary, the GrIS was smaller than present during the MPWP (medium confidence), LIG (high confidence) and the MH (high confidence). GrIS mass loss began following a peak volume attained during the 1450–1850 period and the rate of loss has increased substantially since the turn of the 21st century (high confidence).

The AR5 reported that there was high confidence that the Antarctic Ice Sheet (AIS) was losing mass. The average ice mass loss from Antarctica was 97 [58 to 135] Gt yr^–1 (GMSL equivalent of 0.27 [0.16 to 0.37] mm yr^–1) over 1993–2010, and 147 [74 to 221] Gt yr^–1 (0.41 [0.20 to 0.61] mm yr^–1) over 2005–2010. These assessments included the Antarctic peripheral glaciers. The AR5 reported with high confidence that the volume of the West Antarctic Ice Sheet (WAIS) was reduced during warm periods of the past few million years. The SROCC concluded that over 2006–2015, the AIS lost mass at an average rate of 155 ± 19 Gt yr^–1 (very high confidence). The SROCC also stated that it is virtually certain that the Antarctic Peninsula and WAIS combined have cumulatively lost mass since widespread measurements began in 1992, and that the rate of loss has increased since around 2006.

Process understanding and, to some extent, paleoclimate records show that changes in parts of the AIS can occur over multi-century time scales (<2kyr; Sections 9.4.2.3 and 9.6.2; e.g., Dowdeswell et al., 2020). Based on physical understanding, paleo evidence and numerical simulations, it is very likely that the AIS has been smaller than today during at least some past warm climates (such as MCO and LIG), in particular the WAIS (Figure 9.18; Golledge et al., 2014; de Boer et al., 2015; DeConto and Pollard, 2016; Levy et al., 2016). Results from sediment studies suggest a smaller AIS during the MPWP compared with current levels, with main differences in the WAIS (Section 9.6.2; SROCC, IPCC, 2019; Bertram et al., 2018; Shakun et al., 2018). Marine sediments indicate that during the Pleistocene repeated ungrounding and loss of large marine-based parts of the AIS occurred during interglacial periods, with at least seven transitions between floating and grounded ice in the Ross Sea during the last 780 kyr (McKay et al., 2012) and at least three reductions in ice volume in the Wilkes Basin during the last 500 kyr (Wilson et al., 2018). Proxies, modelling and process understanding (Rohling et al., 2019; Clark et al., 2020) indicate that the AIS was smaller during the LIG than present.

Geological evidence has been used to reconstruct Holocene glacial fluctuations of the ice sheet margin and lowerings of its surface, which occurred at different times in different places, as recently reviewed by Noble et al. (2020). In West Antarctica, marine sediments below the ice sheet (Kingslake et al., 2018) corroborate a previous glacial isostatic adjustment modelling study (Bradley et al., 2015), which suggests that ice had retreated behind the present grounding line prior to about 10 ka, and then readvanced. Geophysical imaging indicates a readvance in this area around 6 ± 2 ka (Wearing and Kingslake, 2019). Other studies from the region conclude that ice-sheet retreat and thinning was fastest from 9 to 8 ka (Johnson et al., 2014; McKay et al., 2016; Spector et al., 2017), or millennia later, during the MH (Hein et al., 2016; Johnson et al., 2019), with indications of a subsequent readvance (Venturelli et al., 2020). In East Antarctica, rapid ice-sheet thinning occurred between around 9 and 5 ka (Jones et al., 2015), consistent with previous work indicating that the ice sheet in many regions was at or close to its current position by 5 ka (Bentley et al., 2014). Overall, during the MH, the AIS was retreating, but remained more extensive than present, while some parts of the ice sheet might have been smaller than now (low confidence).

Improved estimates of surface mass balance (SMB) in Antarctica from 67 ice core records do not show any substantial changes in accumulation rates over most of Antarctica since 1200 CE (Frezzotti et al., 2013). The SMB growth rate in Antarctica is estimated to be 7.0 ± 0.1 Gt per decade between 1800 and 2010 and 14.0 ± 1.8 Gt per decade since 1900 (Thomas et al., 2017). For the period 1979–2000, an insignificant Antarctic-wide negative SMB trend has been estimated (Medley and Thomas, 2019). The Antarctic Ice Sheet lost 2670 [1800 to 3540]Gt (SLE 7.4 [5.0 to 9.8]mm) of ice between 1992 and 2020. The rate of ice-sheet (including peripheral glaciers) mass loss rose from 0 [–36 to +40] Gt yr^–1 (SLE 0.0 [–0.10 to 0.11] mm yr^–1) in 1901–1990 to 192 [145 to 239] Gt yr^–1 (SLE 0.54 [0.47 to 0.61] mm yr^–1) for 2006–2018 (Section 9.4.2, Figure 2.24, and Table 9.5). Within quantified uncertainties, this estimate agrees with other recent estimates (Rignot et al., 2019; B. Smith et al., 2020; Velicogna et al., 2020). There is therefore very high confidence that the AIS has been losing mass over 1992–2020 (Section 9.4.2.1 and Figure 2.24). Major contributions to recent AIS changes arise from West Antarctica and Wilkes Land in East Antarctica (Rignot et al., 2019). For the East Antarctic most studies suggest that the mass balance is not significantly different from zero (Bamber et al., 2018; IMBIE Consortium, 2018; Mohajerani et al., 2018; Rignot et al., 2019).

Figure 2. 24 | Cumulative Antarctic Ice Sheet (AIS) and Greenland Ice Sheet (GrIS) mass changes. Values shown are in gigatons and come from satellite-based measurements (IMBIE Consortium, 2018, 2020) for the period 1992–2020. The estimated uncertainties, very likely range, for the respective cumulative changes are shaded. Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

In summary, the AIS has lost mass between 1992 and 2020 (very high confidence), and there is medium confidence that this mass loss has increased. During the MPWP and LIG, the ice sheet was smaller than present (medium confidence). There is low confidence as to whether the total mass of the ice sheet was larger or smaller around 6 ka compared to now.

The AR5 concluded that there was no evidence of a trend in the AMOC during the period of instrumental observations. However, AR5 also stressed insufficient evidence to support a finding of change in the heat transport of the AMOC. SROCC assessed that there was emerging evidence in sustained observations, both in situ (2004–2017) and revealed from SST-based reconstructions, that the AMOC had weakened during the instrumental era relative to 1850–1900 (medium confidence), although there were insufficient data to quantify the magnitude of the weakening. SROCC also concluded with low confidence an increase of the Southern Ocean upper cell overturning circulation. SROCC also reported with medium confidence that the production of Antarctic Bottom Water had decreased since the 1950s consistent with a decreased lower cell overturning circulation, and potentially modulating the strength of the AMOC.

On multi-millennial timescales, proxy evidence indicates that the AMOC varied repeatedly in strength and vertical structure. During the last glacial period, particularly around the LGM, AMOC was estimated to be shallower than present, although there is continued debate about the magnitude of the shoaling (Lynch-Stieglitz et al., 2007; Gebbie, 2014), and whether this change was associated with a weaker overturning (Ritz et al., 2013; Menviel et al., 2017; Muglia et al., 2018). There are indications that substantial variations in AMOC were associated with abrupt climate changes during the glacial intervals, including Dansgaard-Oeschger and Heinrich events (14–70 ka) (McManus et al., 2004; Böhm et al., 2015; Henry et al., 2016; Lynch-Stieglitz, 2017). During these millennial-scale oscillations, weakened AMOC was associated with dramatic cooling in the NH and warming in the SH (Buizert et al., 2015; Henry et al., 2016), while hemispheric changes of opposite sign accompanied strengthened AMOC. After the final demise of the Laurentide ice sheet about 8 ka, the mean overall strength of AMOC has been relatively stable throughout the rest of the Holocene compared to the preceding 100 kyr (Hoffmann et al., 2018; Lippold et al., 2019). There are however indications of episodic variations in AMOC during the Holocene (Bianchi and McCave, 1999; Oppo et al., 2003; Thornalley et al., 2013; Ayache et al., 2018), and past interglacial intervals (Galaasen et al., 2014, 2020; Hayes et al., 2014; Mokeddem et al., 2014; H. Huang et al., 2020). Over the last 3 kyr, there are indications that AMOC variability was potentially linked to decreasing production of Labrador Sea Water (LSW), one of the water masses contributing to AMOC (Alonso-Garcia et al., 2017; Moffa-Sánchez and Hall, 2017; Moffa-Sánchez et al., 2019).

Numerous proxy records collectively imply that AMOC is currently at its weakest point in the past 1.6 ka (Rahmstorf et al., 2015; Caesar et al., 2018, 2021; Thibodeau et al., 2018; Thornalley et al., 2018). Caesar et al. (2021) analyse a compilation of various available indirect AMOC proxies from marine sediments, in situ-based reconstructions and terrestrial proxies, which show a decline beginning in the late 19th century and over the 20th century superimposed by large decadal variability in the second half of the 20th century. Indirect reconstructions of AMOC components based on coastal sea level records in the western North Atlantic (Ezer, 2013; McCarthy et al., 2015; Piecuch, 2020) show an AMOC decline since the late 1950s, with only a short period of recovery during the 1990s.

However, other studies highlight that proxy records do not show such clear signals (Moffa-Sánchez et al., 2019), and the use of SST- and coastal sea level-based proxies of AMOC places uncertainties on these results (Little et al., 2019; Jackson and Wood, 2020; Menary et al., 2020). For instance, SSTs are additionally influenced by atmospheric and non-AMOC related ocean variability (Josey et al., 2018; Keil et al., 2020; Menary et al., 2020), while sea level responds to a variety of factors (e.g., atmospheric pressure and local winds) independent of the AMOC (Woodworth et al., 2014; Piecuch and Ponte, 2015; Piecuch et al., 2016). Finally, large decadal variability is present in many reconstructions and obscures estimation of the long-term trend over the 20th century (Ezer, 2013; McCarthy et al., 2015; Yashayaev and Loder, 2016; Thornalley et al., 2018; Caesar et al., 2021). It is also noted that the proxy reported AMOC decline, beginning in the late 19th century, is not supported by model-based evidence (Sections 3.5.4.1 and 9.2.3.1).

Since the 1980s, multiple lines of observational evidence for AMOC change exist. Ship-based hydrographic estimates of AMOC as far back as the 1980s show no overall decline in AMOC strength (Fu et al., 2020; Worthington et al., 2021). Direct indications from in-situ observations report a –2.5 ± 1.4 Sv change between 1993 and 2010 across the OVIDE section, superimposed on large interannual to decadal variability (Mercier et al., 2015). At 41°N and 26°N, a decline of –3.1 ± 3.2 Sv per decade and –2.5 ± 2.1 Sv per decade respectively has been reported over 2004–2016 (Baringer et al., 2018; Smeed et al., 2018). However, Moat et al. (2020) report an increase in AMOC strength at 26°N over 2009–2018. Recent time series of moored observations at 11ºS (Hummels et al., 2015), 34°S (Meinen et al., 2018; Kersalé et al., 2020), and between 57 and 60ºN (Lozier et al., 2019) are currently too short to permit robust conclusions about changes. The directly observed AMOC weakening since 2004, while significant, is over too short a period to assess whether it is part of a longer term trend or dominated by decadal‐scale internal variability (Smeed et al., 2014; Collins et al., 2019; Moat et al., 2020). Notably an increase and subsequent decline in the 1990s is present in estimates of AMOC and associated heat transport constructed from reanalyses or auxiliary data (Section 9.2.3.1; Frajka-Williams, 2015; Jackson et al., 2016; Trenberth and Fasullo, 2017; Jackson and Wood, 2020).

Repeated full depth in situ measurements report that deep convection – a major driver for AMOC – has recently returned to the Labrador Sea, particularly in 2015 (Yashayaev and Loder, 2016; Rhein et al., 2017), and to the Irminger Sea (de Jong and de Steur, 2016; Gladyshev et al., 2016; de Jong et al., 2018) following an extended period with weak convection since 2000. An associated strengthening of the outflow from the Labrador Sea has not been observed (Zantopp et al., 2017; Lozier et al., 2019), while strengthening of the AMOC is tentative (Desbruyères et al., 2019; Moat et al., 2020). A long-term increase of the upper overturning cell in the Southern Ocean since the 1990s can be assessed with low confidence, and there is medium confidence of a decrease in Antarctic bottom water (AABW) volume and circulation, which has potential implications for the strength of the AMOC (Section 9.2.3.2).

In summary, proxy-based reconstructions suggest that the AMOC was relatively stable during the past 8 kyr (medium confidence), with a weakening beginning since the late 19th century (medium confidence), but due to a lack of direct observations, confidence in an overall decline of AMOC during the 20th centuryis low. From the mid-2000s to mid-2010s, the directly observed weakening in AMOC (high confidence) cannot be distinguished between decadal-scale variability or a long-term trend (high confidence).

Both AR5 and SROCC reported that western boundary currents (WBCs) have undergone an intensification, warming and poleward expansion, except for the Gulf Stream and the Kuroshio, but did not provide confidence statements. The AR5 reported with medium to high confidence intensification of the North Pacific subpolar gyre, the South Pacific subtropical gyre, and the subtropical cells, along with an expansion of the North Pacific subtropical gyre since the 1990s. It was pointed out that these changes are likely predominantly due to interannual-to-decadal variability, and in the case of the subtropical cells represent a reversal of earlier multi-decadal changes. SROCC concluded that it was unlikely that there has been a statistically significant net southward movement of the mean Antarctic Circumpolar Current (ACC) position over the past 20 years, in contrast to AR5, where this change had been assessed with medium confidence.

The intensity of the Kuroshio current system in the north-west Pacific varied in conjunction with the glaciation cycles over the last 1 Myr, with some limited glacial-interglacial variability in position (Jian et al., 2000; Gallagher et al., 2015). The Agulhas current has strengthened substantially during the warming associated with deglaciations of the past 1 Myr (Peeters et al., 2004; Bard and Rickaby, 2009; Martínez-Méndez et al., 2010; Marino et al., 2013; Ballalai et al., 2019). According to sediment core analyses, the Agulhas leakage varied by about 10 Sv during major climatic transitions over the past 640 kyr (Caley et al., 2014). Available data suggests that there was relatively little change in the net flow of the ACC during the LGM, with no consensus on the sign of changes (McCave et al., 2013; Lamy et al., 2015; Lynch-Stieglitz et al., 2016), except at one location at the northern edge of the Drake Passage where a 40% decrease of transport had been reported (Lamy et al., 2015). Longer time series from the northern entrance to Drake Passage suggest a consistent transport variability of 6–16% through glacial climate cycles, with higher current speeds during interglacial times and reduced current speeds during glacial intervals (Toyos et al., 2020). Inferred variability in the size and strength of the North Atlantic subpolar gyre was substantial, and included rapid changes on millennial time scales during both interglacial and glacial intervals over the last 150 kyr (Born and Levermann, 2010; Mokeddem et al., 2014; Irvalı et al., 2016; Mokeddem and McManus, 2016). North Atlantic – Arctic exchange has also varied in the past, with indications of an increasing inflow of Atlantic waters into the Arctic during the late Holocene (Ślubowska et al., 2005) with an acceleration to the recent inflow that is now the largest of the past 2 kyr (Spielhagen et al., 2011).

A latitudinal shift of subtropical/subpolar gyres on the order of 0.1 ± 0.04° per decade is derived by an indirect method using remote sensing data during 1993–2018 (Yang et al., 2020). Direct observations show a systematic poleward migration of WBCs (Wu et al., 2012; Yang et al., 2016, 2020; Bisagni et al., 2017). However, they do not support an intensification of WBCs, with a weakening, broadening, or little change reported for the Kuroshio (Y.-L. Wang et al., 2016; Wang and Wu, 2018; Collins et al., 2019), Gulf Stream (McCarthy et al., 2018; Collins et al., 2019; Dong et al., 2019; Andres et al., 2020), Agulhas (Beal and Elipot, 2016; Elipot and Beal, 2018) and East Australian (Sloyan and O’Kane, 2015) currents. The Gulf Stream has recently reversed a long-term poleward migration (Bisagni et al., 2017). Multidecadal variability of the strength and position of WBCs (Hsin, 2015; Bisagni et al., 2017; McCarthy et al., 2018) and short records from direct observations obscure the detection of any long-term trends (Yang et al., 2020).

The Pacific to Arctic exchange at the Bering Strait plays a minor role in the total Arctic exchange with the global ocean, which has increased from 0.8 Sv to 1.0 Sv over 1990–2015 (Woodgate, 2018). For Atlantic-Arctic exchange, major branches of Atlantic Water inflow from the North Atlantic into the Arctic across the Greenland–Scotland Ridge have remained stable since the mid-1990s (Berx et al., 2013; Hansen et al., 2015; Jochumsen et al., 2017; Østerhus et al., 2019), with only the smaller pathway of Atlantic Water north of Iceland showing a strengthening trend during 1993–2018 (Casanova-Masjoan et al., 2020), but with associated heat transport strengthening through the 1990s (Rossby et al., 2020; Tsubouchi et al., 2021). The Arctic outflow remained broadly stable from the mid-1990s to the mid 2010s (Østerhus et al., 2019).

The heat and mass transport of the Indonesian throughflow (ITF) shows substantial variability at seasonal to decadal time scales (Zhuang et al., 2013; Q.-Y. Liu et al., 2015; Susanto and Song, 2015; Feng et al., 2017, 2018; M. Li et al., 2018; Sprintall et al., 2019; Xie et al., 2019). Q.-Y. Liu et al. (2015) reported an increasing trend in the ITF geostrophic transport of 1 Sv per decade over 1984–2013, consistent with direct estimates (Sprintall et al., 2014), and results from reanalyses (M. Li et al., 2018), and this appears to be linked to multi-decadal scale variability rather than a long-term trend (Kosaka and Xie, 2013; England et al., 2014; Lee et al., 2015).

Southern Ocean circulation changes are assessed in SROCC (Meredith et al., 2019), and are confirmed and synthesized in Section 9.2.3.2 which shows that there is no indication of ACC transport change, and that it is unlikely that the mean meridional position of the ACC has moved southward in recent decades.

In summary, over the past 3–4 decades, the WBC strength is highly variable (high confidence), and WBCs and subtropical gyres have shifted poleward since 1993 (medium confidence). Net Arctic Ocean volume exchanges with the other ocean basins remained stable over the mid-1990s to the mid-2010s (high confidence). There is high confidence that the ITF shows strong multi-decadal scale variability since the 1980s.

SROCC pointed out that long-term global observations of many key ocean variables, including phytoplankton, have not reached the density and accuracy necessary for detecting change. But SROCC noted the good comparability between short time-scale single-sensor ocean-colour products and a longer time-scale, climate-quality time series of multi-sensor, inter-sensor-bias-corrected, and error-characterized, global data on chlorophyll-a concentration in the surface layers of the ocean. With respect to oligotrophic gyres, AR5 WGII concluded that the oligotrophic subtropical gyres of the Atlantic and Pacific Oceans are expanding and that they indicate declining phytoplankton stocks in these waters (limited evidence, low agreement). With respect to distributions of marine organisms, AR5 WGII reported range shifts of benthic, pelagic, and demersal species and communities (high confidence), though the shifts were not uniform.

Phytoplankton are responsible for marine primary production through photosynthesis; they are a major player in the ocean carbon cycle. They have a high metabolic rate and respond fast to changes in environmental conditions (light, temperature, nutrients, mixing), and as such, serve as a key indicator for change in marine ecosystems. Concentration of chlorophyll-a, the major photosynthetic pigment in all phytoplankton, is often used as a measure of phytoplankton biomass. As primary producers, they are also food for organisms at higher trophic levels. The multi-sensor time series of chlorophyll-a concentration has now been updated (Sathyendranath et al., 2019) to cover 1998–2018. Figure 2.31 shows that global trends in chlorophyll-a for the last two decades are insignificant over large areas of the global ocean (von Schuckmann et al., 2019), but some regions exhibit significant trends, with positive trends in parts of the Arctic and the Antarctic waters (>3% yr^–1), and both negative and positive trends (within ± 3% yr^–1) in parts of the tropics, subtropics and temperate waters. The interannual variability in chlorophyll-a data in many regions is strongly tied to indices of climate variability (Section 2.4 and Annex IV) and changes in total concentration are typically associated with changes in phytoplankton community structure (e.g., Brewin et al., 2012; Racault et al., 2017b). Variability in community structure related to El Niño has, in turn, been linked to variability in fisheries, for example in the catch of anchovy (Engraulis ringens) in the Humboldt current ecosystem (Jackson et al., 2011).

Figure 2. 31 | Phytoplankton dynamics in the ocean. (a) Climatology of chlorophyll-a concentration derived from ocean-colour data (1998–2018); (b) Linear trends in chlorophyll concentration. Trends are calculated using OLS regression with significance assessed following AR(1) adjustment after Santer et al. (2008) (‘×’ marks denote non-significant changes). (c) Histogram of linear trends in chlorophyll concentration, after area weighting and with per-pixel uncertainty estimates based on comparison with in situ data. Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

Since AR5 WGII, analysis of a longer time series of ocean-colour data (1998–2012) has shown (Aiken et al., 2017) that the expansion of the low nutrient part of the North Atlantic oligotrophic gyre was significant, at 0.27 × 10⁶km² per decade, but that the rate was much lower than that reported earlier by Polovina et al. (2008). Furthermore, Aiken et al. (2017) reported no significant trend in the oligotrophic area of the South Atlantic Gyre. With the time series extended to 2016, von Schuckmann et al. (2018) reported that since 2007, there was a general decreasing trend in the areas of the North and South Pacific oligotrophic Gyres, while the North and South Atlantic oligotrophic Gyres remained stable, with little change in area, consistent with Aiken et al . (2017). The changing sign of trends in the areal extent of the oligotrophic gyres with increase in the length of the time series raises the possibility that these changes arise from interannual to multi-decadal variability. The time series of ocean-colour data is too short to discern any trend that might be superimposed on such variability.

Similarly, there is limited consistent and long-term information on large-scale distributions of marine organisms at higher trophic levels. But there are increased indications since AR5 and SROCC that the distributions of various higher trophic-level organisms are shifting both polewards and to deeper levels (Edwards et al., 2016; Haug et al., 2017; Atkinson et al., 2019; Lenoir et al., 2020; Pinsky et al., 2020), mostly consistent with changes in temperature. However observations also show a smaller set of counter-intuitive migrations towards warmer and shallower waters, which could be related to changes in phenology and in larval transport by currents (Fuchs et al., 2020). There are also strengthening indications of greater representation by species with warm-water affinity in marine communities, consistent with expectations under observed warming (Burrows et al., 2019). There are indications that pre-1850 CE plankton communities are different from their modern counterparts globally (Jonkers et al., 2019). Indicators of geographical distributions of species (mostly from coastal waters) suggest that the rates at which some species are leaving or arriving at an ecosystem are variable, leading to changes in community composition (Blowes et al., 2019), with likely greater representation of warm-water species in some locations (Burrows et al., 2019).

In summary, there is high confidence that the latitudinal and depth limits of the distribution of various organisms in the marine biome are changing. There is medium confidence that there are differences in the responses of individual species relative to each other, such that the species compositions of ecosystems are changing. There is medium confidence that chlorophyll concentration in the surface shows weak negative and positive trends in parts of low and mid latitudes, and weak positive trends in some high-latitude areas. There is medium confidence that the large-scale distribution of the oligotrophic gyre provinces is subject to significant inter-annual variations, but low confidence in the long-term trends in the areal extent of these provinces because of insufficient length of direct observations.

SROCC expressed low confidence in satellite-based estimates of trends in marine primary production, citing insufficient length of the time series and lack of corroborating in situ measurements and independent validation time series. The report also cites significant mismatches in absolute values and decadal trends in primary production when different satellite-based products are compared.

Recent model-based results with assimilation of satellite data (Gregg and Rousseaux, 2019), show global annual mean marine primary production of around 38 (±1.13) PgC yr^–1 over 1998–2015. This new result lies towards the low end of values reported in earlier, satellite-based, studies (range 36.5–67 PgC yr^–1, reported in Sathyendranath et al. (2020)). Reconciling the results of Gregg and Rousseaux (2019) with earlier satellite-based studies leads to a mean of 47 (±7.8) PgC yr^–1. There is a strong correlation between interannual regional variability in marine primary production and climate variability (Racault et al., 2017b; Gregg and Rousseaux, 2019). The increase in primary production in the Arctic has been associated with retreating sea ice and with increases in nutrient supply and chlorophyll concentration (Lewis et al., 2020). Gregg and Rousseaux (2019) reported a decreasing trend in marine primary production, of –0.8 PgC (–2.1%) per decade globally. There is low confidence in this trend because of the small number of studies and the short length of the time series (<20 years).

In conclusion, there is low confidence because of the small number of recent studies and the insufficient length of the time series analysed that marine primary production is 47 (± 7.8) PgC yr^–1. A small decrease in productivity is evident globally for the period 1998–2015, but regional changes are larger and of opposing signs (low confidence).

Phenology is the study of the timing of important events in the annual life cycle of organisms (plants or animals; see also Annex VII: Glossary). The AR5 WGII noted that the timing of various seasonal biological events in the ocean had advanced by more than four days per decade over the previous 50-year period and concluded that there was high confidence in observed changes in the phenological metrics of marine organisms. The AR5 WGII further reported that, of those observations that showed a response, 81% of changes in phenology, distribution and abundance were consistent with anticipated responses to climate warming according to theoretical expectations, corroborated by updates in SROCC. The consequent current and future impacts on interactions between species, including competition and predator-prey dynamics, were noted with high confidence.

There are additional indications that phenological metrics related to different species are changing, but not always in a similar manner. For example, many seabirds are breeding earlier, while others are breeding later (Sydeman et al., 2015). Planktonic organisms in the North Atlantic are also responding differently to each other when subjected to the same environmental changes (Edwards and Richardson, 2004). Furthermore, different factors could be responsible for triggering phenological responses in different stages in the life cycle of a single organism (Koeller et al., 2009). The shift in the distribution of many benthic invertebrates on the North-west Atlantic shelf, including some commercially important shellfish, could be explained by phenology and larval transport, and the shift and contraction of species range have been associated with higher mortality (Fuchs et al., 2020). Changes in phytoplankton phenological indicators globally (Racault et al., 2012; Sapiano et al., 2012) have been linked to indicators of climate variability, such as the multivariate ENSO Index (Racault et al., 2017a), with responses varying across ecological provinces of the ocean (Longhurst, 2007).

Phenological links between multiple components of an ecosystem have to be maintained intact, to retain system integrity. Since all higher pelagic organisms depend on phytoplankton for their food, either directly or indirectly, a match favours survival, and a mismatch is antagonistic to survival. Match represents synchronicity in the phenological events of both prey and predator. There are indications from ocean-colour data used in conjunction with fisheries data that the survival rate of various larger marine organisms depends on phenological metrics related to the seasonality of phytoplankton growth. Such links have been demonstrated, for example, for haddock (Melanogrammus aeglefinus) in the North-west Atlantic (Platt et al., 2003); northern shrimp in the North Atlantic (Koeller et al., 2009; Ouellet et al., 2011); sardine (Sardinella aurita) off the Ivory coast (Kassi et al., 2018); cod (Gadus morhua) and haddock (Melanogrammus aeglefinus) larvae in the North-West Atlantic (Trzcinski et al., 2013); and oil sardine (Sardinella longiceps) off the south-west coast of India. Borstad et al. (2011) showed that fledgling production rate of rhinoceros auklets (Cerorhinca monocerata) on a remote island in coastal north-eastern Pacific was related to seasonal values of chlorophyll-a biomass in the vicinity of the island.

In summary, new in situ data as well as satellite observations strengthen AR5 and SROCC findings that various phenological metrics for many species of marine organisms have changed in the last half century (high confidence), though many regions and many species of marine organisms remain under-sampled or even unsampled. The changes vary with location and with species (high confidence). There is a strong dependence of survival in higher trophic-level organisms (fish, exploited invertebrates, birds) on the availability of food at various stages in their life cycle, which in turn depends on phenologies of both (high confidence). There is a gap in our understanding of how the varied responses of marine organisms to climate change, from a phenological perspective, might threaten the stability and integrity of entire ecosystems.

The AR5 WGII briefly discussed large-scale changes in the length of the growing season but made no confidence statement about observed trends. However, AR5 did conclude with high confidence that warming contributed to an overall spring advancement in the NH.

Recent in situ analyses document increases in the length of the thermal growing season (i.e., the period of the year when temperatures are warm enough to support growth) over much of the extratropical land surface since at least the mid-20th century. Over the NH as a whole, an increase of about 2.0 days per decade is evident for 1951–2018 (Dunn et al., 2020), with slightly larger increases north of 45°N (Barichivich et al., 2013). Over North America, a rise of about 1.3 days per decade is apparent in the United States for 1900–2014 (Kukal and Irmak, 2018), with larger increases after 1980 (McCabe et al., 2015); likewise, all ecozones in Canada experienced increases from 1950–2010 (Pedlar et al., 2015). Growing season length in China increased by at least 1.0 days per decade since 1960 (Xia et al., 2018) and by several days per decade in South Korea since 1970 (Jung et al., 2015). In general, changes in phenological indicators are consistent with the increase in growing season length documented by instrumental data (Parmesan and Hanley, 2015). Several long-term, site-specific records illustrate the unusualness of recent phenological changes relative to interannual variability; for example, peak bloom dates for cherry blossoms in Kyoto, Japan have occurred progressively earlier in the growing season in recent decades, as have grape harvest dates in Beaune, France (Figure 2.32).

Figure 2.32 | Phenological indicators of changes in growing season. (a) Cherry blossom peak bloom in Kyoto, Japan; (b) grape harvest in Beaune, France; (c) spring phenology index in eastern China; (d) full flower of Piedmont species in Philadelphia, USA; (e) grape harvest in Central Victoria, Australia; (f) start of growing season in Tibetan Plateau, China. Red lines depict the 25-year moving average (top row) or the nine-year moving average (middle and bottom rows) with the minimum roughness boundary constraint of Mann (2004). Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

Changes in the length of the photosynthetically active growing season (derived from the Normalized Difference Vegetation Index (NDVI)) are also evident over many land areas since the early 1980s. Increases of about 2.0 days per decade are apparent north of 45°N since the early 1980s (centred over mid-latitude Eurasia and north-eastern North America), with indications of a reversal to a decline in season length starting in the early 2000s (Barichivich et al., 2013; Zhao et al., 2015; Garonna et al., 2016; Q. Liu et al., 2016; Park et al., 2016). Satellite-based records suggest that most NH regions have experienced both an earlier start and a later end to the growing season, a finding supported by ground-based data (Piao et al., 2020). A number of studies also capture increases in growing season length over the Canadian Arctic (W. Chen et al., 2016), Fennoscandia (Høgda et al., 2013), most of Europe (Garonna et al., 2014), and parts of sub-Saharan Africa (Vrieling et al., 2013).

The general consistency between in situ and satellite estimates over the NH is noteworthy given that many factors independently contribute to uncertainty in observed changes. For example, there is no universally accepted definition of growing season length across in situ analyses; some define the growing season as the period based on a temperature threshold (e.g., 5°C) whereas others use the frost-free period. Spatial and temporal coverage can also affect conclusions based upon in situ studies (Donat et al., 2013b). For satellite analyses, uncertainties can be related to the satellite datasets themselves (e.g., satellite drift, sensor differences, calibration uncertainties, atmospheric effects); and to the methods for determining phenological metrics (e.g., start, end, and length of season; S. Wang et al., 2016).

In summary, based on multiple independent analyses of in situ, satellite, and phenological data, there is high confidence that the length of the growing season has increased over much of the extratropical NH since at least the mid-20th century.

The AR5 WGII concluded that many terrestrial species have shifted their geographic ranges in recent decades (high confidence). Similarly, SRCCL assessed that many land species have experienced range size and location changes as well as altered abundances over recent decades (high confidence). SROCC noted that species composition and abundance have markedly changed in high mountain ecosystems in recent decades (very high confidence).

Paleoclimate reconstructions document large-scale biome shifts from the deep past through the Holocene (e.g., Hoogakker et al., 2016). The northernmost location of the treeline is a representative indicator in this regard (Figure 2.34). During the MPWP, boreal forest extended to the Arctic coast, with the northernmost treeline being about 4° to 10° latitude further north than at present; temperate forests and grasslands were also shifted poleward (with reduced tundra extent), while savannahs and woodlands were more expansive in Africa and Australia at the expense of deserts (Cross-Chapter Box 2.4, Figure 1b; Salzmann et al., 2008, 2013; Sniderman et al., 2016; Andrae et al., 2018). During the LGM, tundra and steppe expanded whereas forests were globally reduced in extent (Prentice et al., 2000; Binney et al., 2017), the northern treeline being about 17° to 23° latitude south of its present-day location in most areas. During the LDT, pervasive ecosystem transformations occurred in response to warming and other climatic changes (Nolan et al., 2018; Fordham et al., 2020). By the MH, North Africa had experienced a widespread conversion from grasslands to desert (Hoelzmann et al., 1998; Prentice et al., 2000; Sha et al., 2019), and the northernmost treeline had shifted poleward again to about 1° to 3° latitude north of its current location (MacDonald et al., 2000; Binney et al., 2009; Williams et al., 2011). Over the past half century, there has been an increase in the spatial synchrony of annual tree growth across all continents that is unprecedented during the past millennium (Manzanedo et al., 2020). Elevated rates of vegetation change in the Holocene are consistent with climate variability (Shuman et al., 2019), intensified human land use (Fyfe et al., 2015; Marquer et al., 2017), and resulting increased ecosystem novelty (Finsinger et al., 2017; K.D. Burke et al., 2019).

Long-term ecological records capture extensive range shifts during the 20th and early 21st centuries (Lenoir and Svenning, 2015; Pecl et al., 2017). Research has been most extensive for North America and western Eurasia, with fewer studies for central Africa, eastern Asia, South America, Greenland, and Antarctica (Lenoir and Svenning, 2015). Most documented changes are toward cooler conditions – that is, poleward and upslope (Lenoir et al., 2008; Harsch et al., 2009; Elmendorf et al., 2015; Parmesan and Hanley, 2015; Evans and Brown, 2017). Notably, a large, quasi-global analysis (Chen et al., 2011) estimated that many insect, bird, and plant species had shifted by 17 (±3) km per decade toward higher latitudes and 11 (±2) m per decade toward higher elevations since the mid-20th century, with changes in both the leading and trailing edges of species ranges (Rumpf et al., 2018). Over the past century, long-term ecological surveys also show that species turnover (i.e., the total number of gains and losses of species within an area) has significantly increased across a broad array of ecosystems (Dornelas et al., 2014, 2019), including undisturbed montane areas worldwide (Gibson-Reinemer et al., 2015). Despite global losses to biodiversity, however, most local assemblages have experienced a change in biodiversity rather than a systematic loss (Pimm et al., 2014). With increased species turnover, the novelty of contemporary communities relative to historical baselines has risen (Hobbs et al., 2009; Radeloff et al., 2015) due to greater spatial homogenization, mixtures of exotic and native species, altered disturbance regimes, and legacies of current or historic land use (Olden and Rooney, 2006; Schulte et al., 2007; Thompson et al., 2013; Goring et al., 2016). In general, terrestrial species have had lower rates of turnover than marine species (Dornelas et al., 2018; Blowes et al., 2019).

There are exceptions to the general pattern of poleward/upslope migration. For some species, various biotic and abiotic factors (such as precipitation and land use) supersede the physiological effects of temperature (Vanderwal et al., 2013; Gibson-Reinemer and Rahel, 2015; Ordonez et al., 2016; Scheffers et al., 2016; Lenoir et al., 2020). For other species, poleward migration is slower than expectations from the observed temperature increases. Trees are one such example because of their long lifespan and gradual maturity (Renwick and Rocca, 2015); in fact, poleward advance is only evident at about half of the sites in a large global dataset of treeline dynamics for 1900-present (Harsch et al., 2009). Furthermore, the northernmost extent of treeline at present (roughly 73°N) is actually somewhat south of its location in the MH (MacDonald et al., 2008) despite an expanding growing season in the extratropical NH since the mid-20th century (Section 2.3.4.3.1).

Consistent with species range shifts, SRCCL noted that there have been changes in the geographical distribution of climate zones. Poleward shifts in temperate and continental climates are evident across the globe over 1950–2010, with decreases in the area (and increases in the average elevation) of polar climates (Chan and Wu, 2015). Zonal changes towards higher latitudes in winter plant hardiness regions are apparent since the 1970s over the central and eastern USA, with elevational changes also being important in the western USA (Daly et al., 2012). A clear northward shift in winter plant hardiness zones is detectable across western Canada since 1930, with somewhat lesser changes in the south-eastern part of the country (McKenney et al., 2014). A northward migration of agro-climate zones is also evident over Europe since the mid-1970s (Ceglar et al., 2019). In addition, a shift toward more arid climate zones is apparent in some areas, such as the Asian monsoon region (Son and Bae, 2015) as well as parts of South America and Africa (Spinoni et al., 2015).

In summary, there is very high confidence that many terrestrial species have shifted their geographic ranges poleward and/or upslope over the past century, with increased rates of species turnover. There is high confidence that the geographical distribution of climate zones has shifted in many parts of the world.

The AR5 WGII briefly discussed changes in global vegetation greenness derived from satellite proxies for photosynthetic activity. Observed trends varied in their strength and consistency, and AR5 thus made no confidence statement on observed changes. The SRCCL subsequently concluded that greening had increased globally over the past 2–3 decades (high confidence).

Vegetation index data derived from AVHRR and MODIS depicts increases in aspects of vegetation greenness (i.e., green leaf area and/or mass) over the past four decades (Piao et al., 2020). NDVI increased globally from the early 1980s through the early 2010s (Y. Liu et al., 2015a). N. Pan et al. (2018) found NDVI increases over about 70% of the Earth’s vegetated surface through 2013, and Osborne et al. (2018) noted strong upward changes in NDVI in the circumpolar Arctic through 2016. Globally integrated Leaf Area Index (LAI) also rose from the early 1980s through at least the early 2010s (Zhu et al., 2016; Forzieri et al., 2017; Jiang et al., 2017; Xiao et al., 2017) and probably through near-present; for example, C. Chen et al. (2019) documented an LAI increase over one-third of the global vegetated area from 2000–2017. Although less frequently analysed for temporal trends, Fraction of Absorbed Photosynthetically Active Radiation (FAPAR) likewise increased over many global land areas (particularly China, India, and eastern Europe) in the past two decades (Figure 2.33; Forkel et al., 2014; Gobron, 2018; Keenan and Riley, 2018). There are also documented changes in specific vegetation types, such as a 7% rise in global tree cover for 1982–2016 (Song et al., 2018) and an expansion of shrub extent in the Arctic tundra over 1982–2017 (Myers-Smith et al., 2020). The increased greening is largely consistent with CO₂ fertilization at the global scale, with other changes being noteworthy at the regional level (Piao et al., 2020); examples include agricultural intensification in China and India (X. Chen et al., 2019; Gao et al., 2019) and temperature increases in the northern high latitudes (Kong et al., 2017; Keenan and Riley, 2018) and in other areas such as the Loess Plateau in central China (Y. Wang et al., 2018). Notably, some areas (such as parts of Amazonia, central Asia, and the Congo basin) have experienced browning (i.e., decreases in green leaf area and/or mass) (Hoogakker et al., 2015; Gottschalk et al., 2016; Anderson et al., 2019). Because rates of browning have exceeded rates of greening in some regions since the late 1990s, the increase in global greening has been somewhat slower in the last two decades (T.-Y. Pan et al., 2018).

Figure 2. 33 | Satellite-based trends in fraction of absorbed photosynthetically active radiation (per decade) for 1998–2019. Trends are calculated using OLS regression with significance assessed following AR(1) adjustment after Santer et al. (2008); ‘×’ marks denote non-significant trend). Unvegetated areas such as barren deserts (grey) and ice sheets (white) have no trend in FAPAR. Further details on data sources and processing are available in the chapter data table (Table 2.SM.1).

Open figure

Global-scale linear trends differ substantially across products for the same periods and trend metrics used (Jiang et al., 2017). Several factors contribute to this large span in estimated changes. Remotely sensed vegetation products vary in their spatial and temporal completeness as well as resolution and are sensitive to contamination from atmospheric composition, clouds, snow cover, and anisotropy, as well as orbital changes and sensor degradations (de Jong et al., 2012; Zhu et al., 2016; Jiang et al., 2017; Xiao et al., 2017; N. Pan et al., 2018). Ground-based measurements suitable for calibration and validation are scarce before 2000 (Xiao et al., 2017), and the recalibration of satellite records (e.g., as in from MODIS Collection 5 to 6) can affect trends (Piao et al., 2020). It is possible that the increase in greenness over 2000–2015 is larger than the increase in gross primary production (based on flux tower measurements and MODIS Collection 6 data) (L. Zhang et al., 2018). Land use changes and altered disturbance regimes (e.g., floods, fires, diseases) may mask large-scale signals (Franklin et al., 2016). In addition, there is a plethora of models for the identification of phenological metrics from satellite data as well as a variety of statistical techniques for analysing historical changes (S. Wang et al., 2016).

In summary, there is high confidence that vegetation greenness (i.e., green leaf area and/or mass) has increased globally since the early 1980s. However, there is low confidence in the magnitude of this increase owing to the large range in available estimates.