Since AR5, evidence of changes in the temperature of lakes and rivers has continued to increase. Global warming rates for lake surface waters were estimated as 0.21°C–0.45°C per decade between 1970 and 2010, exceeding sea-surface temperature (SST) trends of 0.09°C per decade between 1980 and 2017 (robust evidence, high agreement ) (Figure 2.2; (Schneider and Hook, 2010; Kraemer et al., 2015; O’Reilly et al., 2015; Woolway et al., 2020b). Warming of lake surface water temperatures was variable within regions (O’Reilly et al., 2015) but more homogeneous than deep-water temperature changes (Pilla et al., 2020). Because temperature trends in lakes can vary vertically, horizontally and seasonally, complex changes have occurred in the amount of habitat available to aquatic organisms at particular depths and temperatures (Kraemer et al., 2021).

Changes in river water temperatures ranged from −1.21°C to +1.076°C per decade between 1901 and 2010 (medium evidence, medium agreement ) (Hari et al., 2006; Kaushal et al., 2010; Jurgelėnaitė et al., 2012; Li et al., 2012; Latkovska and Apsīte, 2016; Marszelewski and Pius, 2016). The more rapid increase in surface water temperature in lakes and rivers in regions with cold winters (O’Reilly et al., 2015) can, in part, be attributed to the amplified warming in polar and high-latitude regions (robust evidence, high agreement ) (Screen and Simmonds, 2010; Stuecker et al., 2018).

Shifts in thermal regime: Since AR5, the trend that lake waters mix less frequently continues (Butcher et al., 2015; Adrian et al., 2016; Richardson et al., 2017; Woolway et al., 2017). This results from greater warming of surface temperatures relative to deep-water temperatures, and the loss of ice during winter which prevents inverse thermal stratification in north temperate lakes (robust evidence, high agreement ) (Adrian et al., 2009; Winslow et al., 2015; Adrian et al., 2016; Schwefel et al., 2016; Richardson et al., 2017).

Oxygen availability: increased water temperature and reduced mixing cause a decrease in dissolved oxygen. In 400 lakes, dissolved oxygen in surface and deep waters declined by 4.1 and 16.8%, respectively, between 1980 and 2017 (Jane et al., 2021). The deepest water layers are expected to experience an increase in hypoxic conditions by >25% due to fewer complete mixing events, with strong repercussions for nutrient dynamics and the loss of thermal habitat (robust evidence, high agreement ) (Straile et al., 2010; Zhang et al., 2015; Schwefel et al., 2016).

Depending on how the intensification of the global water cycle affects individual lake water budgets, the amount of water stored in specific lakes may increase, decrease or have no substantial cumulative effect (Notaro et al., 2015; Pekel et al., 2016; Rodell et al., 2018; Busker et al., 2019; Woolway et al., 2020b). The magnitude of hydrological changes that can be assuredly attributed to climate change remains uncertain (Hegerl et al., 2015; Gronewold and Rood, 2019; Kraemer et al., 2020). Attribution of water storage variation in lakes due to climate change is facilitated when such variations occur coherently across broad geographic regions and long time scales, preferably absent of other anthropogenic hydrological influences (Watras et al., 2014; Kraemer et al., 2020). There is increasing awareness that climate change contributes to the loss of small temporary ponds which cover a greater global area than lakes (Bagella et al., 2016).

Lakes fed by glacial melt water are growing in response to climate change and glacier retreat (robust evidence, high agreement ) (Shugar et al., 2020). Water storage increases on the Tibetan Plateau (Figure 2.3a) have been attributed to changes in glacier melt, permafrost thaw, precipitation and runoff, in part as a result of climate change (Huang et al., 2011; Meng et al., 2019; Wang et al., 2020a). High confidence in attribution of these trends to climate change is supported by long-term ground survey data and observations from the Gravity Recovery and Climate Experiment (GRACE) satellite mission (Ma et al., 2010; Rodell et al., 2018; Kraemer et al., 2020).

In the Arctic, lake area has increased in regions with continuous permafrost, and decreased in regions where permafrost is thinner and discontinuous (robust evidence, high agreement ) (See Chapter 4) (Smith et al., 2005; Andresen and Lougheed, 2015; Nitze et al., 2018; Mekonnen et al., 2021).

Analysis of river flows from 7250 observatories around the world covering the years 1971–2010 and identified spatially complex patterns, with reductions in northeastern Brazil, southern Australia and the Mediterranean, and increases in northern Europe (medium evidence, medium agreement ) (Gudmundsson et al., 2021). More than half of global rivers undergo periodic drying that reduces river connectivity (medium evidence, medium agreement ). Increased frequency and intensity of droughts may cause perennial rivers to become intermittent and intermittent rivers to disappear (medium evidence, medium agreement ), threatening freshwater fish in habitats already characterised by heat and droughts (Datry et al., 2016; Schneider et al., 2017; Jaric et al., 2019). In high-altitude/latitude streams, reduced glacier and snowpack extent, earlier snowmelt and altered precipitation patterns, attributed to climate change, have increased flow intermittency (Siebers et al., 2019; Gudmundsson et al., 2021). Patterns in flow regimes can be directly linked to a variety of processes shaping freshwater biodiversity, so any climate change-induced changes in flow regimes and river connectivity are expected to alter species composition as well as having societal impacts (See Chapter 3 in (IPCC, 2018b)) (Bunn and Arthington, 2002; Thomson et al., 2012; Chessman, 2015; Kakouei et al., 2018).

Studies since AR5 have confirmed ongoing and accelerating loss of lake and river ice in the Northern Hemisphere (robust evidence, high agreement ) (Figure 2.4). In recent decades, systems have been freezing later in winter and thawing earlier in spring, reducing ice duration by >2 weeks per year and leading to an increasing numbers of years with a loss of perennial ice cover, intermittent ice cover or even an absence of ice (Adrian et al., 2009; Kirillin et al., 2012; Paquette et al., 2015; Adrian et al., 2016; Park et al., 2016; Roberts et al., 2017; Sharma et al., 2019). The global extent of river ice declined by 25% between 1984 and 2018 (Yang et al., 2020). This trend has been more pronounced at higher latitudes, consistent with enhanced polar warming (large geographic coverage) (Du et al., 2017). Empirical long-term and remote-sensing data gathered in an increasingly large number of freshwater systems supports very high confidence in attributing these trends to climate change. For the decline of glaciers, snow and permafrost, see Chapter 4 (this report) and the Special Report on the Ocean and Cryosphere in a Changing Climate (IPCC, 2019b).

Since AR5, numerous drastic short-term responses have been observed in lakes and rivers, to both expected seasonal extreme events and unexpected supra-seasonal extremes extending over multiple seasons. Consequences for ecosystem functioning are not well understood (Bogan et al., 2015; Death et al., 2015; Stockwell et al., 2020). Increasing frequencies of severe floods and droughts attributed to climate change are major threats for river ecosystems (Peters et al., 2016; Alfieri et al., 2017). While extreme floods cause massive physical disturbance, moderate floods can have positive effects, providing woody debris that contributes to habitat complexity and diversity, flushing fine sediments, dissolving organic carbon and providing important food sources from terrestrial origins (Peters et al., 2016; Talbot et al., 2018). Droughts reduce river habitat diversity and connectivity, threatening aquatic species, especially in deserts and arid regions (Bogan et al., 2015; Death et al., 2015; Ledger and Milner, 2015; Jaric et al., 2019).

Rivers already under stress from human activities such as urban development and farming on floodplains are prone to reduced resilience to future extreme events (medium confidence) (Woodward et al., 2016; Talbot et al., 2018). Thus, the potential for floods to become catastrophic for ecosystem services is exacerbated by LULCC (Peters et al., 2016; Talbot et al., 2018). However, biota can recover rapidly from extreme flood events if river geomorphology is not greatly altered. If instream habitat is strongly affected, recovery, if it occurs, takes much longer, resulting in a decline of biodiversity (medium confidence) (Thorp et al., 2010; Death et al., 2015; Poff et al., 2018).

However, not all extreme events will have a biological impact, depending, in particular, on the timing, magnitude and frequency of events and the antecedent conditions (Bailey and van de Pol, 2016; Stockwell et al., 2020; Jennings et al., 2021; Thayne et al., 2021). For instance, an extreme wind event may have little impact on phytoplankton in a lake that was fully mixed prior to the event. Conversely, the effects of a storm on phytoplankton communities may compound when lakes have not yet recovered from a previous storm or if periods of drought alternate with periods of intense precipitation (limited evidence) (Leonard et al., 2014; Stockwell et al., 2020).

In summary, extreme events (heat waves, storms and loss of ice) affect lakes in terms of water temperature, water level, light, oxygen concentrations and nutrient dynamics, which, in turn, affect primary production, fish communities and GHG emissions (high confidence). These impacts are modified by levels of solar radiation, wind speed and precipitation (Woolway et al., 2020a). Droughts have a negative impact on water quality in streams and lakes by increasing water temperature, salinity, the frequency of algal blooms and contaminant concentrations, and reducing concentrations of nutrients and dissolved oxygen (medium confidence) (Peters et al., 2016; Alfieri et al., 2017; Woolway et al., 2020a). Understanding how these pressures subsequently cascade through freshwater ecosystems will be essential for future projections of their resistance and resilience towards extreme events (Leonard et al., 2014; Stockwell et al., 2020). See Table SM2.1 for specific examples of observed changes.

Given the strength of relationship between past GSAT and warming trends at lake surfaces (Figure 2.2; Section 2.3.3.1) and projected increases in heat waves, surface water temperatures are projected to continue to increase (Woolway et al., 2021). Mean May to October lake surface temperatures in 46,557 European lakes were projected to be 2.9°C, 4.5°C and 6.5°C warmer by 2081–2099 compared to the historic period (1981–1999) under RCP2.0, RCP6.0 and RCP8.5, respectively (Woolway et al., 2020a). Under RCP2.6, the average intensity of lake heat waves increases from 3.7°C to 4.0°C and the average duration from 7.7 to 27.0 days, relative to the historic period (1970–1999). For RCP8.5, warming increases to 5.4°C and duration increases dramatically to 95.5 days (medium confidence) (Woolway et al., 2021).

Worldwide alterations in lake mixing regimes in response to climate change are projected (Kirillin, 2010). Most prominently, monomictic lakes—undergoing one mixing event in most years—will become permanently stratified, while lakes that are currently dimictic—mixing twice per year—will become monomictic by 2080–2100 (medium confidence) (Woolway and Merchant, 2019). Nevertheless, predicting mixing behaviour remains an important challenge and attribution to climate change remains difficult (Schwefel et al., 2016; Bruce et al., 2018).

Under climate projections of 3.2°C warming, 4.6% of the ice-covered lakes in the Northern Hemisphere could switch to intermittent winter ice cover (Figure 2.4a; (Sharma et al., 2019). Unfrozen and warmer lakes lose more water to evaporation (Wang et al., 2018b). By 2100, global annual lake evaporation will increase by 16%, relative to 2006–2015, under RCP8.5 (Woolway et al., 2020b). Moreover, melting of ice decreases the ratio of sensible to latent heat flux, thus channelling more energy into evaporation (medium confidence) (Wang et al., 2018b). In the periods 2009–2029 and 2080–2100, average duration of river ice is projected to decline by 7.3 and 16.7 days under RCP4.5 and RCP8.5, respectively (Figure 2.4b; Yang et al., 2020).

Projections of lake water storage are limited by the absence of reliable, long-term, homogenous and spatially resolved hydrologic observations (Hegerl et al., 2015). This uncertainty is reflected in the widely divergent projections in response to future climate changes in individual lakes (Angel and Kunkel, 2010; MacKay and Seglenieks, 2012; Malsy et al., 2012; Notaro et al., 2015) . Selecting models that perform well when comparing hindcasted to observed past water storage variation often does little to reduce water storage projection uncertainty (Angel and Kunkel, 2010). This wide range of potential changes complicates lake management. For information on observed and projected changes in the global water cycle and hydrological regimes for streams, lakes, wetland, groundwater and their implications on water quality and societies, see Chapter 4, this report, and (Douville et al., 2021). For the role of weather and climate extremes on the global water cycle, see (Seneviratne et al., 2021).

In summary, with ongoing climate warming and an increase in the frequency and intensity of extreme events, observed increases in water temperature, losses of ice and shifts in thermal regime are projected to continue (high confidence).

Poleward and upward range shifts were already attributable to climate warming with high confidence in AR5. Publication of observed range shifts in accord with climate change have accelerated since AR5 and strengthened attribution. Ongoing latitudinal and elevational range shifts driven by regional climate trends are now well-established globally across many groups of organisms, and attributable to climate change with very high confidence due to very high consistency across a now very large body of species and studies and an in-depth understanding of mechanisms underlying physiological and ecological responses to climate drivers (Table 2.2; Table 2.3, Table SM2.1) (Pöyry et al., 2009; Chen et al., 2011; Grewe et al., 2013; Gibson-Reinemer and Rahel, 2015; MacLean and Beissinger, 2017; Pacifici et al., 2017; Anderegg et al., 2019). Range shifts stem from local population extinctions along warm-range boundaries (Anderegg et al., 2019) as well as from the colonisation of new regions at cold-range boundaries (Ralston et al., 2017).

Many studies since AR4 have tended not to be designed as attribution studies, particularly recent large-scale, multi-species meta-analyses. That is to say, all the data available was included in such studies (from both undisturbed and highly degraded lands and including very short-term data sets of <20 years), with little attempt to design the studies to differentiate the effects of climate change from those of other potential confounding variables. These studies tended to find greater lag and a lower proportion of species changing in the directions expected from climate change, with the authors concluding that LULCC, particularly habitat loss and fragmentation, was impeding wild species from effectively tracking climate change (Lenoir and Svenning, 2015; Rumpf et al., 2019; Lenoir et al., 2020).

Attribution is strong for species and species-interactions for which there is a robust mechanistic understanding of the role of climate on biological processes (high confidence). Unprecedented outbreaks of spruce beetles occurring from Alaska to Utah in the 1990s were attributed to warm weather that, in Alaska, facilitated a halving of the insect’s life cycle from two years to one (Logan et al., 2003). Milder winters and warmer growing seasons were likewise implicated in poleward expansions and increasing outbreaks of several forest pests (Weed et al., 2013), leading to the current prediction that 41% of major insect pest species will further increase their damage as climate warms, and only 4% will reduce their impacts, while the rest will show mixed responses (Lehmann et al., 2020).

During their range shifts, forest pests remain climate-sensitive. For example, the distribution of the western spruce budworm is limited at its warm range edges by the adverse effects of mild winters on overwinter survival, and at its cool range by the ability to arrive at a cold-resistant stage before winter arrives (Régnière and Nealis, 2019). We might therefore expect tree mortality from insect outbreaks to be most severe at sites climatically less suitable for the plants, where plants would be under more stress. However, (Jaime et al., 2019), using separate species distribution models (SDMs) (MaxEnt) for the insects and plants, found that observed mortality of Scots pine from bark beetles was highest at sites that were most climatically suitable for the trees as well as for the insects. In a study of tree mortality in California, bark beetles selectively killed highly stressed fir trees, but killed pines according to their size irrespective of stress status (Stephenson et al., 2019).

Range shifts in a poleward and upward direction, following expected trajectories according to local and regional climate trends, are strongly occurring in freshwater fish populations in North America (Lynch et al., 2016b), Europe (Comte and Grenouillet, 2013; Gozlan et al., 2019) and Central Asia (Gozlan et al., 2019) (medium evidence, high agreement). Cold-water fish, such as coregonids and smelt have been negatively affected at the equatorial borders of their distributions (Jeppesen et al., 2012). Upward elevational range shifts in rivers and streams have been observed. Systematic shifts towards higher elevation and upstream were found for 32 stream-fish species in France following regional variation in climate change (Comte and Grenouillet, 2013). Bull trout (Salvelinus confluentus) in Idaho (USA), were estimated to have lost 11–20% (8–16% per decade) of the headwater stream lengths necessary for cold-water spawning and early juvenile rearing, with the largest losses occurring in the coldest habitats (Isaak et al., 2010). Range contractions of the same species have been found in the Rocky Mountains watershed (Eby et al., 2014). Likewise, the distribution of the stonefly Zapada glacier, endemic to the alpine streams of the Glacier National Park in Montana (USA), has been reduced over several decades by an upstream retreat to higher, cooler sites as water temperatures have increased and glacial masses have decreased (Giersch et al., 2015).

The melting of glaciers has led to a change in water discharge associated with community turnover in glacier-fed streams (Cauvy-Fraunié and Dangles, 2019). For instance, glacier-obligate macro-invertebrates have started disappearing when glacial cover drops below approximately 50% (robust evidence, high agreement ), reviewed in (Hotaling et al., 2017). For freshwater invertebrates, no meaningful trends have been detected in geographic extent or population size for most species (Gozlan et al., 2019).

An invasive freshwater cyanobacterium in lakes, Cylindrospermopsis raciborskii, originating from the tropics, has spread to temperate zones over the last few decades due to the climate change-induced earlier increase of water temperature in spring (Wiedner et al., 2007), aided by a competitive advantage in eutrophic systems (Ekvall et al., 2013; Urrutia-Cordero et al., 2016).

Disappearances of local populations within a species range are more frequent and better documented than whole species’ extinctions, and attribution to climate change is possible for sites with minimal confounding non-climatic stressors. Changes of temperature extremes are often more important to these local extinction rates than changes of mean annual temperature (high confidence) (see Sections 2.3.1, 2.3.2, 2.3.3.5, 2.4.2.6, Cross-Chapter Box EXTREMES in this chapter) (Parmesan et al., 2013). A global meta-analysis of 236 species of birds, mammals, amphibians, fish, invertebrates and plants across 132 independent studies found that changes in population abundances were strongly related to temperature variability globally, and significantly related to precipiation variability in lower latitudes (Pearce-Higgins et al., 2015). In a global study of 538 diverse plant and animal species, sites with local extinctions were associated with smaller changes of mean annual temperature but larger and faster changes of hottest yearly temperatures than sites where populations persisted (Román-Palacios and Wiens, 2020). Near warm range limits, 44% of species had suffered local extinctions. In both temperate and tropical regions, sites with local extinction had greater increases in maximum temperatures than those without: a Tmax increase of 0.456°C and 0.316°C versus a Tmean increase of 0.153°C and 0.061°C for temperate (n= 505 sites) and tropical (n= 76 sites), respectively (P < 0.001) (Román-Palacios and Wiens, 2020).

Wiens (2016) assumed that population extinctions were primarily driven by climate change when they occurred at elevational or latitudinal ‘warm edge’ range limits, and were at relatively undisturbed sites stated by the authors to be under increasing climatic stress. By this criterion, climate-caused local extinctions were widespread among plants and animals globally, detected in 47% of 976 species examined. The percentage of species suffering these extinctions was higher in the Tropics (55%) than in temperate habitats (39%), higher in freshwater (74%) than in marine (51%) or terrestrial (46%) habitats, and higher in animals (50%) than in plants (39%). The difference between plants and animals varied with latitude; in the temperate zone, a much higher proportion of animals than plants suffered range-limit extinctions (38.6% of 207 animal species vs. 8.6% of 105 plants, P < 0.0001) while at tropical sites, local extinction rates were (nonsignificantly) higher in plants (59% of 155 species) than in animals (52% of 349 species), the reverse of their temperate-zone relationship. Rates varied across animal groups from 35% in mammals, to 43% in birds, 56% in insects and 59% in fish (Wiens, 2016).

Freshwater population extinctions are mainly due to habitat loss, the introduction of alien species, pollution, over-harvesting (Gozlan et al., 2019; IPBES, 2019) and climate change-induced epidemic diseases (Pounds et al., 2006)(see Section 2.4.2.7.1). Climate warming, particularly through the intensification and severity of droughts, contributes to the disappearance of small ponds which hold rare and endemic species (Bagella et al., 2016). Systematic data on the extent and biology of small ponds is, however, lacking on the global scale. Extreme heat waves can lead to large local fish kills in lakes (see Section 2.3.3.5), when water temperature and oxygen concentrations surpass critical thresholds and threatening cold-water fish and amphibians (Thompson et al., 2012). Evidence of a local extinction of some invertebrate species with a 1.4°C–1.7°C rise in mean annual stream winter temperature from 1981 to 2005 was reported in Abrahams et al. (2013). Population declines of specialist species in glacier-fed streams, such as the non-biting midge Diamesa davisi (Chironomidae), can be attributed to climate-change-driven glacier retreat (Cauvy-Fraunié and Dangles, 2019), and the flatworm Crenobia alpina (Planariidae) has been reported as locally extinct in the Welsh Llyn Brianne river (Durance and Ormerod, 2010; Larsen et al., 2018).

Many high montane possums in Australia have low physiological tolerance to heat waves, with death occuring due to heat-driven dehydration at temperatures exceeding 29°C–30°C for >4–5 h over several days (Meade et al., 2018; Turner, 2020). Major declines have been recorded for several species, population extinctions have occured at lower elevations since the early 2000s, and the white sub-species of the lemuroid ringtail possum (Hemibelideus lemuroides) in Queensland, Australia, disappeared after heat waves in 2005 (high confidence): intensive censuses found only 2 individuals in 2009 (Chandler, 2014; Weber et al., 2021).

Two terrestrial and freshwater species have become extinct in the wild, with climate change implicated as a key driver. The cloud forest-restricted golden toad (Incilius periglenes) was extinct by 1990 in a nature preserve in Costa Rica, driven by successive extreme droughts. This occurred in the absence of chytridiomycosis infection, caused by the fungal pathogen Bd, verified during field censuses of golden toad populations in the process of extinction as well as genetic analyses of museum specimens, although Bd was present in other frog species in the region (medium evidence, high agreement ) (Pounds et al., 1999; Pounds et al., 2006; Puschendorf et al., 2006; Richards-Hrdlicka, 2013). The interaction between expansion of chytrid fungus globally and local climate change is implicated in the extinction of a wide range of tropical amphibians (high confidence) (see Section 2.4.2.7.1 Case Study 2 Chytrid fungus and climate change).

The BC melomys (Melomys rubicola), the only mammal endemic to the Great Barrier Reef, inhabited a small (5-hectare) low-lying (<3-m-high) cay in the Torres Strait Islands, Australia. Recorded as having a population size of several hundred in 1978, this mammal has not been seen since 2009 and was declared extinct in 2016 (Gynther et al., 2016). SLR and documented increases in storm surge and in tropical cyclones, driven by climate change, led to multiple inundations of the island in the 2000s. Between 1998 and 2014, herbacious vegetation, the food resource for the BC melomys, declined by 97% in area (from 2.2 down to 0.065 hectares), and from 11 plant species down to two (Gynther et al., 2016; Watson, 2016; Woinarski, 2016; Woinarski et al., 2017). The island was unihabited with few non-climatic threats, providing high confidence in the attribution of extinction of the BC melomys to climate change-driven increases in the frequency and duration of island inundation (Turner and Batianoff, 2007; Woinarski et al., 2014; Gynther et al., 2016; Watson, 2016; Woinarski et al., 2017).

In the IUCN Red List (IUCN, 2019), 16.2% of terrestrial and freshwater species (n= 3,777 species) that are listed as endangered, critically endangered or extinct in the wild (n= 23,251 species) list climate change or severe weather as one of their threats.

In summary, local population extinctions caused by climate-change-driven increases in extreme weather and climate events have been widespread among plants and animals (very high confidence), and the first clear documentations of entire species driven extinct by recent climate change is emerging (medium confidence).

Since AR5, the number of studies of changes in phenology (timing of biological events) has increased substantially, aided by advances in remote sensing (Piao et al., 2019). Phenological studies have documented particularly consistent conclusions on responses of plants and animals to warming, including the advancement of spring events and the lengthening of growing seasons in temperate regions (via a combination of advancement of spring events and, to a lesser extent, the retardation of autumn events) (robust evidence, high agreement ) (Table 2.2, Table 2.3, Table SM2.1) (Menzel et al., 2020). In the Tropics, by contrast, changes in precipitation have more strongly influenced animal phenology than have temperature changes (Cohen et al., 2018). A meta-analysis compared observed phenological advances in birds with expectations due to warming local climates, and concluded that the observed advances fell short of what was expected and that a substantial phenological climate debt had been generated (Radchuk et al., 2019).

Taxonomic groups have differed in their responses (Parmesan, 2007; Thackeray et al., 2010), and a few have completely diverged from general trends. For example, seabirds continue to breed with their pre-climate-change phenologies (Keogan et al., 2018). Newer reviews and analyses reveal differences in responses across continents and time intervals (Piao et al., 2019). Mean advance in days per decade for plants was 5.5 in China and 3.0–4.2 in Europe, but only 0.9 in North America (Piao et al., 2019). Mean values for the retardation of autumn leaf fall, which can be more influenced by photoperiod and less by temperature than spring leaf-out, were 0.36 days per decade in Europe (Menzel et al., 2020), 2.6 days per decade in China and around 3 days per decade in the USA (medium evidence, high agreement ) (Piao et al., 2019).

The rapid rate of the advancement of spring events in the 1990s slowed down in the 2000s, and stalled or even reversed in some regions (Menzel et al., 2020). Wang et al. (2019) noted, from remote sensing, that during the ‘global warming hiatus’ from 1998 to 2012, there were no global trends in either spring green-up or autumn colouring. Annual crops, the timing of which is determined by farmers, were an exception. When natural systems were advancing fast prior to 1998, farmers advanced more slowly, but during the natural ‘hiatus’, farmed crops advanced faster than wild plants and cultivated trees (Menzel et al., 2020). In a long (67 years) European time series (Menzel et al., 2020), autumn leaf colouring showed delays attributed to winter and spring warming in 57% of observations (mean delay of 0.36 days per decade); spring and summer phenologies advanced in 89% of wild plants, despite decreased winter chilling, with around 60% of trends significant and ‘strongly attributable’ to winter and spring warming; and the growing season lengthened in 84% of cases (mean lengthening 0.26 days yr -1) (Table 2.2).

Table 2.2 | Global fingerprints of climate change impacts across wild species. (Updated from (Parmesan and Hanley, 2015). For each study for which data were made available, a response for an individual species or functional group was classified as (1) no response (i.e., no significant change in the measured trait over time), (2) if a significant change was found, the response was classified as either consistent or not consistent with expectations according to local or regional climate trends. Percentages are approximate and estimated for the studies as a whole. Individual analyses within the studies may differ. The specific metrics of climate change analysed for associations with biological change vary somewhat across studies, but most use changes in local or regional temperatures (e.g., mean monthly T or mean annual T), with some using precipitation metrics (e.g., total annual rainfall). For example, a consistent response would be poleward range shifts in areas that are warming. Probability (P) of getting the observed ratio of consistent-to-not consistent responses by chance was <10–13 for (Parmesan and Yohe, 2003; Root et al., 2003; Root et al., 2005; Poloczanska et al., 2013) and <0.001 for Rosenzweig et al. (2008). The last collumn distinguishes studies that were designed for attribution to climate change (e.g. by analysing only long-term data from relatively undisturbed habitats (see section 2.1.3 and 2.4.1)(Parmesan et al., 2013; Cramer et al., 2014) from those that analysed all available data, including data from areas highly-impacted by non-climate drivers (e.g. LULCC).

Changes in freshwater systems are consistent with changes in terrestrial systems: earlier development of phytoplankton and zooplankton and earlier spawning by fish in spring as well as extension of the growing season are occurring (robust evidence, high agreement ) (Adrian et al., 2009; De Senerpont Domis et al., 2013; Adrian et al., 2016; Thackeray et al., 2016). Phenological changes in lakes have been related to rising water temperatures, reduced ice cover and prolonged thermal stratification (increasing evidence and agreement since AR5; very high confidence). Crozier and Hutchings (2014) reviewed the phenological changes in fish and documented that changes in the timing of migration and reproduction, age at maturity, age at juvenile migration, growth, survival and fecundity were associated primarily with changes in temperature. The median return time of Atlantic salmon to rivers in Newfoundland and Labrador advanced by 12–21 days over the past decades, associated with overall warmer conditions (Dempson et al., 2017).

Early meta-analyses tested the straightforward hypotheses that warming should shift timing earlier and ranges polewards. Once these trends had been established (IPCC, 2014b; Parmesan and Hanley, 2015), exceptions to them became a focus of study. For example, in northern regions of the Northern Hemisphere, the spring flowering of some plants was delayed instead of being advanced as to be expected with warming (Cook et al., 2012a; Cook et al., 2012b; Legave et al., 2015). These turned out to be species requiring vernalisation (winter chilling) to speed their development in spring (Ettinger et al., 2020). For these plants, phenological changes result from the combined effects of advancement caused by spring warming and retardation caused by winter warming. Incorporating this level of complexity into analyses revealed that a greater proportion of species was responding to climate change than estimated according to the simple expectation that warming would always cause advancement (92% responding versus 72% from earlier analyses) (Cook et al., 2012b).

Animal species can show vernalisation equivalent to that in plants (Stålhandske et al., 2017). However, a semi-global meta-analysis of terrestrial animals failed to detect the delaying effects of warming winters (Cohen et al., 2018). The same animal-based meta-analysis contrasted phenological changes in temperate-zone animals, which are principally explained by changes in temperature, with those at lower latitudes, which tend to follow changes in precipitation (Cohen et al., 2018).

Vitasse et al. (2018), working with alpine trees, found that phenological delay with increasing elevation had declined from 34 days/1000 m in 1960 to 22 days/1000 m in 2016, greatly reducing the differences in timing between trees growing at different elevations. This reduction was greatest after warmer winters, suggesting that winter warming is a principal cause of the overall trend.

Lian et al. (2020) observed that earlier spring leaf-out in the Northern Hemisphere is causing increases in evapotranspiration that are not fully compensated by increased precipitation. The consequence is a greater soil moisture deficit in summer, expected to exacerbate impacts of heat waves as well as drought stress. In Arctic freshwater ecosystems, Heim et al. (2015) demonstrated the importance of seasonal cues for fish migration, which can be impacted by climate change due to reduced stream connectivity and fragmentation, earlier peak flows and increased evapotranspiration.

Precipitation has also been implicated in exceptions to the rule that ranges should be shifting to higher elevations. In dry climates, increases in precipitation accompanying climate warming can facilitate downslope range shifts (Tingley et al., 2012).

Multiple responses can co-occur. Hällfors et al. (2021), in a study of 289 Lepidoptera in Finland, found that, with warming, 45% had either shifted their range northward or advanced their flight season. The 15% of species that did both (shifting northward by 113.1 km and advancing their flight period by 2.7 days per decade, on average, over a 20-year period) had the largest population increases, and the 40% of species that showed no response had the largest population declines.

Impacts on species physiology in terrestrial and freshwater systems have been observed, and attributed to climate change (medium confidence). These include changes in tolerance to high temperatures (Healy and Schulte, 2012; Gunderson and Stillman, 2015; Deery et al., 2021), increased metabolic costs of living at elevated temperatures (Scheffers et al., 2016) and shifts in sex ratios in species with temperature-dependent sex determination. For example, warmer temperatures have driven the masculinisation of lizard populations (Schwanz and Janzen, 2008; Schwanz, 2016; Edmands, 2021) and the feminisation of turtle populations (Telemeco et al., 2009). Skewed sex ratios can lead to mate shortages, reduced population growth, reduced adaptive potential and increased extinction risk, because genetic diversity decreases as fewer individuals mate and heterozygosity is lost (Mitchell and Janzen, 2010; Edmands, 2021).

Behavioural plasticity (flexibility) such as nest-site selection can provide a partial buffer from the effects of increasing temperature by placing the individual in a slightly cooler microclimate, but there are environmental and physical limits to this plasticity (medium confidence) (Refsnider and Janzen, 2016; Telemeco et al., 2017). Plasticity in heat tolerance (e.g., due to reversible acclimation or acclimatisation) can also potentially compensate for rising temperatures (Angilletta Jr, 2009), but ectotherms have relatively low acclimation in thermal tolerance and acclimation is expected to only slightly reduce the risk of overheating in even the most plastic taxa (low confidence) (Gunderson and Stillman, 2015).

Geographic variation in thermal tolerance plasticity is expected to influence the vulnerability and range shifts of species in response to climate change (Gunderson and Stillman, 2015; Sun et al., 2021). In many ectotherms, plasticity in thermal tolerance increases polewards, as thermal seasonality increases (Chown et al., 2004), contributing to higher vulnerability to warming in tropical organisms (low confidence) (Huey et al., 2009; Campos et al., 2021). Some species have evolved extreme upper thermal limits at the expense of plasticity, reflecting an evolutionary trade-off between these traits (Angilletta et al., 2003; Stillman, 2003). The most heat-tolerant species, such as those from extreme environments, may therefore be at a greater risk of warming because of an inability to physiologically adjust to thermal change (low confidence) (Bozinovic et al., 2011; Overgaard et al., 2014; Magozzi and Calosi, 2015).

Physiological changes have observable impacts on morphology, such as changes in body size (and length of appendages), and colour changes in butterflies, dragonflies and birds (medium confidence) (Galeotti et al., 2009; Karell et al., 2011). However, trends are not always linear or consistent across realms, taxonomic groups or geographic regions (Gotanda et al., 2015). Some morphological changes arise in response to environmental changes, rather than as the result of genetic adaptation or selection for an optimal body type. For example, dietary changes associated with climate change have led to changes in chipmunk skull morphology (Walsh et al., 2016).

Decreased body size has been suggested as a general response of species to climate change in freshwater species, given the temperature-related constraints of metabolism with larger size. Reduced body size in response to global warming has been documented for freshwater bacteria, plankton and fish, as well as a shift towards smaller species (Daufresne et al., 2009; Winder et al., 2009; Jeppesen et al., 2010; Crozier and Hutchings, 2014; Jeppesen et al., 2014; Farmer et al., 2015; Rasconi et al., 2015; Woodward et al., 2016). However, the lack of systematic empirical evidence in fresh waters, and confounding effects such as interactions between temperature, nutrient availability and predation, limit generalisations in attributing observed body size changes to climate change (low confidence) (Pomati et al., 2020 Nutrients).

Evidence is weak for a consistent reduction in body size across taxonomic groups in terrestrial animals (low confidence) (Siepielski et al., 2019). Decreased body size in warmer climates (as higher surface area-to-volume ratios maximise heat loss) is expected, based on biogeographic patterns such as Bergmann’s rule, but both increases and decreases have been documented in mammals, birds, lizards and invertebrates and were attributed to climate change (Teplitsky and Millien, 2014; Gotanda et al., 2015; Gardner et al., 2019; Hill et al., 2021). Contrasting patterns (increased body size) may be due to short-term modifications in selection pressures (e.g., changes to predation and competition), variation in life histories or as a result of interactions with climate variables other than temperature (e.g., changes to food availability along with rainfall changes) and other disturbances (Yom-Tov and Yom-Tov, 2004; Gardner et al., 2019; Wilson et al., 2019) or use of different body size measurements (linear vs. volumetric dimensions) (Salewski et al., 2014).

Several lines of evidence suggest the evolution of melanism in response to climate change (low confidence), with colour changes associated with thermoregulation being demonstrated in butterflies (Zeuss et al., 2014; MacLean et al., 2016; MacLean et al., 2019a), beetles (de Jong and Brakefield, 1998; Brakefield and de Jong, 2011; Zvereva et al., 2019), dragonflies (Zeuss et al., 2014) and phasmids (Nosil et al., 2018). Such changes may represent decreased phenotypic diversity and, potentially, genetic diversity (low confidence), but the consequences of climate change for the genetic structure and diversity of populations have not been widely assessed (Pauls et al., 2013). Simplistically, the thermal melanism hypothesis suggests that lighter (higher-reflectance) individuals should be fitter and therefore be selected for in a warmer climate (Clusella-Trullas et al., 2007). However, several biotic (e.g., thermoregulatory requirements, predator avoidance and signalling) and abiotic (e.g., UV, moisture and inter-annual variability) factors interact to influence changes in colour, making attribution to climate change across species and broad geographic regions difficult (Kingsolver and Buckley, 2015; Stuart-Fox et al., 2017; Clusella-Trullas and Nielsen, 2020).

Interactions between morphological changes and changes in phenology may facilitate or constrain adaptation to climate change (medium confidence) (Hedrick et al., 2021). For example, advancing phenology in migratory species may impose selection on morphological traits (e.g., wing length) to increase migration speed. If advancing spring phenology results in earlier breeding, this may offset the effect of rising temperatures in the breeding range and reduce the effect of increasing temperature on body size (Zimova et al., 2021). A study of 52 species of North American migratory birds, based on >70,000 specimens, showed that spring migration phenology has advanced over the past 40 years, concurrent with widespread shifts in morphology (reduced body size and increased wing length), perhaps to compensate for the increased metabolic cost of flight as body size decreases (Weeks et al., 2020).

A lack of understanding of physiological constraints and mechanisms remains a barrier to predicting many of the ecological effects of climate change (Bozinovic et al., 2011; Vázquez et al., 2017; González-Tokman et al., 2020). Many behavioural, morphological and physiological responses are highly species- and context-specific, making generalisations difficult (Bodensteiner et al., 2021). Recent advances in mechanistic understanding (from experiments), in process-based modelling (including micro-climates and developmental processes) (Carter and Janzen, 2021) and in the sophistication of niche models (Kearney et al., 2009) have improved projections, but comprehensive tests of geographic patterns and processes in thermal tolerance and plasticity are still lacking, with studies limited to a few phylogenetically restricted analyses showing mixed results (Gunderson and Stillman, 2015). Improved understanding of the mechanistic basis for observed geographic patterns in thermal tolerance and plasticity is needed to identify the physiological limits of species, the potential for adaptation and the presence of evolutionary trade-offs, which will strongly influence population declines, species range shifts, invasive interactions and the success of conservation interventions (Cooke et al., 2021; Ryan and Gunderson, 2021).

Assessment of changes in diseases of terrestrial and freshwater wild organisms was scarce in WGII AR4, AR5, IPCC SR1.5 and IPCC SRCCL. Further, most emerging infectious diseases (EIDs) are zoonoses, that is, they are transmissible between humans and animals, and are climate sensitive (Woolhouse et al., 2001; Woolhouse and Gowtage-Sequeria, 2005; McIntyre et al., 2017; Salyer et al., 2017). AR4 found weak-to-moderate evidence that disease vectors and their diseases had changed their distributions in concert with climate change, but attribution studies were lacking (Smith et al., 2014). In AR5, WGII AR5 Chapter 11, geographic expansion of a few VBDs to higher latitudes and elevations were detected and associated with regional climate trends, but the non-climatic drivers were not well assessed, leading to a medium confidence in attribution (Smith et al., 2014)). Here, we build on previous assessments by focussing on changes in the population dynamics and geographic distribution of diseases in wild animals as well as diseases in humans and domestic animals that are harboured, amplified and transmitted by wild animal reservoir hosts and vectors.

Increased disease incidence is correlated with regional climatic changes, as expected from a basic understanding of underlying biology and relationships between temperature, precipitation, and disease ecology (robust evidence, high agreement ) (Norwegian Polar Institute, 2009; Tersago et al., 2009; Tabachnick, 2010; Paz, 2015; Dewage et al., 2019; Deksne et al., 2020; Shocket et al., 2020; Couper et al., 2021). Whether increases in diseases in wild and domestic animals correspond to an increased risk of disease in nearby human populations is complicated by the potential buffering effects of the local medical system, access to health care and the socioeconomic status, education, behaviours and general health of the human population (see also Chapter 7 and Cross-Chapter Box ILLNESS in this chapter).

Previous sections document the tendency of species to retain their climate envelopes by some combination of range shift and phenological change (very high confidence). However, this tracking of climate change can be incomplete, causing species or populations to experience hotter conditions than those to which they are adapted, and thereby incur ‘climate debts’ (section 2.4.2.3.1) (Devictor et al., 2012). The importance of population-level debt is illustrated by a study in which the estimated debt values were correlated with population dynamic trends in a North American migratory songbird, the yellow warbler, Setophaga petechia. Populations that were genetic outliers for their local climate space had larger population declines (greater debt) than those with genotypes closer to the average values for that particular climate space. Debt values were estimated from genomic analyses independent of the population trends, and were distributed across the species’ range in a mosaic, not simply concentrated at range margins, rendering the results robust to being confounded by broad-scale geographical trends (Bay et al., 2018). Soroye et al. (2020) found similar results for 66 species of bumble-bees across Europe and North America, with declines in abundances spread throughout species’ ranges, but being greatest where populations already near their climate limits were being pushed beyond their climatic tolerances with climate change {2.4.2.3.1}.

In the absence of evolutionary constraints, climate debts can be cancelled by genetically based increases in thermal tolerance and the ability to perform in high ambient temperatures. In species already showing local adaptation to climate, populations currently living at relatively cool sites should be able to evolve to adopt the traits of populations currently at warmer sites as their local experience of climate changes (Singer, 2017; Socolar et al., 2017).

An increasing number of studies document evolutionary responses to climate change in populations not at their warm range limits (Franks and Hoffmann, 2012). Organisms with short generation times should have a higher capacity to genetically track climate change than species with long generation times, such as mammals (Boutin and Lane, 2014). Indeed, observed evolutionary impacts have been mainly documented in insects, especially at expanding range margins (Chuang and Peterson, 2016) where evolutionary changes have increased dispersal ability (Thomas et al., 2001) and decreased host specialisation (Bridle et al., 2014; Lancaster, 2020) (medium evidence, medium agreement).

Away from range margins, individual populations experiencing regional warming have evolved diverse traits related to climate adaptation. For example, pitcher-plant mosquitoes (Wyeomyia smithii) in Pacific Northwest America have evolved to wait for shorter days before initiating diapause. This adaptation to lengthening summers enables them to delay overwintering until later and add an extra generation each year (Bradshaw and Holzapfel, 2001). Among 26 populations of Drosophila subobscura studied on three continents, 22 experienced climate warming across two or more decades, and 21 of these 22 showed increasing frequency of the chromosome inversion characteristic of populations adapted to hot climates (robust evidence, high agreement) (Balanya et al., 2006).

However, for populations already at their warm range limits, their ability to track climate change in situ would require evolving to survive and reproduce outside their species’ historical climate envelope: abilities of wild species to do this is not supported by experimental or observational evidence (medium evidence, high agreement ) (Singer, 2017). Whether or not they can depends on the level of ‘niche conservatism’ operating at the species level (Lavergne et al., 2010). If a species whose range limits are determined by climate finds itself completely outside of its traditional climate envelope, extinction is expected in the absence of ‘evolutionary rescue’ (Bell and Gonzalez, 2009; Bell et al., 2019).

To investigate the evolutionary potential of a species to survive in a novel climate entirely outside its traditional climate envelope, experiments have been carried out on ectotherms testing thermal performance, thermal tolerance and their evolvabilities (Castaneda et al., 2019; Xue et al., 2019). Tests of thermal performance have been complicated, as both long-term acclimation and trans-generational effects occur (Sgro et al., 2016). However, the results to date have been consistent: despite widespread local adaptation to climate across species’ ranges, substantial constraints exist regarding the evolution of greater stress tolerance (e.g., high temperatures and drought) at warm range limits (medium evidence, high agreement ) (Hoffmann and Sgro, 2011; MacLean et al., 2019b).

For example, as temperature was experimentally increased, the amount of genetic variance in the fitness of Drosophila melanogaster decreased; in hot environments, flies had low evolvability (Kristensen et al., 2015). The hypothesis that heat-stress tolerance is evolutionarily constrained is further supported by experiments in which 22Drosophila spp. drawn from tropical and temperate climes were subjected to extremes of heat and cold. They differed, as expected, in cold tolerances, but not in heat tolerances nor in temperatures at which optimal performances were observed (MacLean et al., 2019b).

Plasticity (flexibility) in acclimating to thermal regimes helps organisms adapt to environmental change. The form and extent of plasticity can vary among populations experiencing different climates (Kelly, 2019) and generate phenotypic values outside the prior range for the species, but plasticity itself has not yet been observed to evolve in response to climate change (Kelly, 2019).

Relevant genetic changes in nature (e.g., affecting heat tolerance) have not yet been shown to alter the boundaries of existing genetic variation for any species. Further, a recent global analysis of 91 species found, on average, a 5.4–6.5% decline in genetic diversity within populations since the start of the Industrial Revolution, with much larger declines for island species (27.6–30.9% reductions) (Leigh et al., 2019). In Leigh et al. (2019), genetic declines were documented in both common and already endangered species of fish, mammals, birds, insects, amphibians and reptiles. These declines in genetic diversity, though not caused by climate change, decrease the abilities of wild species to adapt to climate change via evolutionary responses. Evolutionary rescue of entire species has not yet been observed in nature, nor is it expected, based on experimental and theoretical studies (medium evidence, high agreement ).

Hybridisation between closely related species has increased in recent decades as one species shifts its range boundaries and positions itself more closely to the other. Hybrids between polar bears and brown bears have been documented in northern Canada (Kelly et al., 2010). In North American rivers, hybridisation between invasive rainbow trout and native cutthroat trout has increased in frequency as the rainbow trout has expanded into warming waters (Muhlfeld et al., 2014). Whether climate-changed induced hybridisations can generate novel climate adaptations remains to be seen.

In summary, with our present knowledge, evolution is not expected to be sufficient to prevent the extinction of whole species if a species’ climate space disappears within the region they inhabit (high confidence).

Attribution for biome (major vegetation form of an ecosystem) shifts is complex because of their extensive, sometimes continental, spatial scale (Whittaker, 1975; Olson et al., 2001; Woodward et al., 2004). Therefore, non-climatic factors strongly influence biome spatial distributions (Ellis and Ramankutty, 2008).

The most robust attribution studies use data from many species, individual locations with minimal confounding factors, particularly observed recent LULCC, and scale up by analysing multiple locations across a large zone between biomes, providing multiple lines of evidence (Hegerl et al., 2010; Parmesan et al., 2013). Multivariate statistical analyses aid attribution studies by allowing the assessment of relative weights among multiple factors, including variables related to climate change (Gonzalez et al., 2012). However, drivers often have strong, significant interactions with one another, complicating quantitative assessment of the strength of individual drivers (Parmesan et al., 2013). In these cases, manipulative experiments are critical in assessing attribution to the drivers of climate change.

Certain biomes exhibit a relatively stronger relationship to climate; for example, Arctic tundra generally has a distinct ecotone with boreal conifer forest (Whittaker, 1975). In these areas, attribution of biome shifts to climate change are relatively straightforward, if human LULCC is minimal. However, other biomes, such as many grassland systems, are not in equilibrium with climate (Bond et al., 2005). In these systems, their evolutionary history (Keeley et al., 2011; Strömberg, 2011; Charles-Dominique et al., 2016), distribution, structure and function have been shaped by climate and natural disturbances, such as fire and herbivory (Staver et al., 2011; Lehmann et al., 2014; Pausas, 2015; Bakker et al., 2016; Malhi et al., 2016). Disturbance variability is an inherent characteristic of grassland systems, and suitable ‘control’ conditions are seldom available in nature. Furthermore, due to the integral role of disturbance, these biomes have been widely affected by long-term and widespread shifts in grazing regimes, large-scale losses of mega-herbivores and fire suppression policies (Archibald et al., 2013; Malhi et al., 2016; Hempson et al., 2017). It is necessary to conduct climate change attribution on a case-by-case basis for grasslands; such assessments are complex as direct climate change impacts from either inherent variation within disturbance regimes or directional changes in background disturbances are difficult to separate (detailed in Sections 2.4.3.2.1; 2.4.3.2.2; 2.4.3.5). Confidence in assessments is increased when the observed trends are supported by a mechanistic understanding of responses identified by physiological studies, manipulative field experiments, greenhouse studies and lab experiments (Table SM2.1).

Divergent responses to anthropogenic climate change are occurring within and across arid regions, depending on time period, location, detection methodology and vegetation type (see Cross-Chapter Paper 3). Emerging shifts in ecosystem structure, functioning and biodiversity are supported by evidence from modelled impacts of projected climate and CO2 levels. While observed responsiveness of arid vegetation productivity to rising atmospheric CO2 (Fensholt et al., 2012) may offset risks from reduced water availability (Fang et al., 2017), climate- and CO2-driven changes are key risks in arid regions, interacting with habitat degradation, wildfires and invasive species (Hurlbert et al., 2019).

Widespread vegetation greening, as projected in AR4, is occurring in arid shrublands (Zhang et al., 2019a; Maestre et al., 2021) as a result of increases in leaf area, woody cover and herbaceous production at desert–grassland interfaces (Gonsamo et al., 2021). Plant productivity in arid regions has increased (Fensholt et al., 2012) because of improved water-use efficiency associated with elevated CO2 (Norby and Zak, 2011; Donohue et al., 2013; Burrell et al., 2020; Gonsamo et al., 2021) (medium evidence, high agreement ), altered rainfall seasonality and amount (Rohde et al., 2019; Zhang et al., 2019a ) (robust evidence, high agreement ), increases in temperature (Ratajczak et al., 2014; Wilcox et al., 2018) (robust evidence, high agreement ) and heavy grazing (robust evidence, high agreement ), with the relative importance differing across locations (Donohue et al., 2013; Caracciolo et al., 2016; Archer et al., 2017; Hoffmann et al., 2019b; Rohde et al., 2019). Woody-plant encroachment into arid shrublands is occurring with high confidence in North America (Caracciolo et al., 2016; Archer et al., 2017) and southern Africa (du Toit and O’Connor, 2014; Ward et al., 2014; Masubelele et al., 2015a; Hoffman et al., 2019; Rohde et al., 2019), and with low confidence in central Asia (Li et al., 2015). In North America, sagebrush steppe changes have been attributed to increases in temperature and earlier snowpack melt (USGCRP, 2017; Mote et al., 2018; Snyder et al., 2019).

Non-native grasses are invading the sagebrush steppes (cold deserts) in North America (Chambers et al., 2014) attributed to warming (Bradley et al., 2016; Hufft and Zelikova, 2016). In the eastern semi-desert (Karoo) of South Africa, annual rainfall increases and a rainfall seasonality shift (du Toit and O’Connor, 2014) are increasing grassiness as arid grasslands expand into semi-desert shrublands (du Toit et al., 2015; Masubelele et al., 2015b; Masubelele et al., 2015a) causing fire in areas seldom burned historically (Coates et al., 2016).

Interactions of drought, warming and land management have caused vegetation mortality (see Section 2.4.4.3) and reduced vegetation cover in shrublands, as projected by AR4 (Burrell et al., 2020). Increased heat and drought are causing the health and abundance of succulent species to decline (Musil et al., 2009; Schmiedel et al., 2012; Aragón-Gastélum et al., 2014; Koźmińska et al., 2019). Hot droughts, in particular, have been shown to reduce population resilience (Koźmińska et al., 2019).

Since AR5 (Settele et al. (2014), all five Mediterranean-type ecosystems (MTEs) of the world have experienced extreme droughts within the past decade, with South Africa and California reporting their worst on record (robust evidence, high agreement ) (Diffenbaugh et al., 2015; Williams et al., 2015a; Garreaud et al., 2017; Otto et al., 2018; Sousa et al., 2018). Climate change is causing these droughts to become more frequent and severe (medium evidence, medium agreement ) (AghaKouchak et al., 2014; Garreaud et al., 2017; Otto et al., 2018; Seneviratne et al., 2021).

MTEs show a range of direct responses to various forms of water deficit, but have also been affected by increasing fire activity linked to drought (Abatzoglou and Williams, 2016), and interactions between drought or extreme weather and fire affecting post-fire ecosystem recovery (Slingsby et al., 2017). Responses include shifts in functional composition (Acácio et al., 2017; Syphard et al., 2019a), decline of vegetation health (Hope et al., 2014; Asner et al., 2016a), decline or loss of characteristic species (White et al., 2016; Stephenson et al., 2019), shifts in composition towards more drought- or heat-adapted species and declining diversity (see also section 2.4.4.3) (Slingsby et al., 2017.; Harrison et al., 2018).

Declines in plant health and increased mortality in MTEs associated with drought have been widely documented (robust evidence, high agreement ) (Section 2.4.4.3). Remote-sensing studies show drought-associated mortality in post-fire vegetation regrowth in the Fynbos of South Africa (Slingsby et al., 2020b), reduced canopy health in forests within MTE zones of South Africa (Hope et al., 2014) and declines in canopy water content in the forests of California (Asner et al., 2016a). Several studies reported climate-associated responses of dominant or charismatic species. High mortality in the Clanwilliam cedar tree between 1931 and 2013 occurred at lower, hotter elevations in the Fynbos of South Africa (White et al., 2016). Drought reduced growth and increased mortality of the holm oak, Quercus ilex, on the Iberian Peninsula of Spain (Natalini et al. (2016). Portuguese shrublands experienced losses of many deciduous and evergreen oak species, and an increasing dominance of pyrophytic xeric trees (Acácio et al., 2017). The 2012–2015 drought in California caused high-canopy foliage dieback of the giant sequoia (Sequoiadendron giganteum) (Stephenson et al., 2019), increased the dominance of oaks relative to pines as a result of the increased water deficit, and led to large-scale tree mortality due to interactions of drought and insect pest outbreaks (McIntyre et al., 2015; Fettig et al., 2019).

Species distribution or community composition changes have contributed to declines in diversity and/or shifts towards more drought- or heat-adapted species (medium evidence, high agreement ). Two conifer species (Pinus longaeva and P. flexilis) shifted upslope 19 m from 1950 to 2016 in the Great Basin, USA, (Smithers et al., 2018). Reduced winter precipitation caused native annual forbs to recede, resulting in long-lasting and potentially unidirectional reductions in diversity in a Californian grassland (Harrison et al., 2018). More frequent extreme hot and dry weather between 1966 and 2010 caused a decline in diversity during the post-fire regeneration phase in the Fynbos of South Africa (Slingsby et al., 2017), resulting in shifts towards species with higher temperature preferences (Slingsby et al., 2017). In Italy, Del Vecchio et al. (2015) observed increases in plant cover and thermophilic species in coastal foredune habitats between 1989 and 2012.

In southern California, USA, areas of forest and woody shrublands are shifting to grasslands, driven by a combination of climate and land use factors such as increased drought, fire ignition frequency and increases in nitrogen deposition (robust evidence, high agreement ) (Jacobsen and Pratt, 2018; Park et al., 2018; Park and Jenerette, 2019; Syphard et al., 2019b).

The effects of climate change on heat, fuel and wildfire ignition limits show spatial and temporal variation globally (see Section 2.3.6.1), but there have been a number of observed impacts on MTEs (medium evidence, high agreement ). Climate change caused increases in fuel aridity and the area of land burned by wildfires across the western USA from 1985 to 2015 (Abatzoglou and Williams, 2016). Local and global climatic variability led to a 4-year decrease in the average fire return time in the Fynbos, South Africa, when comparing fires recorded in 1951–1975 and 1976–2000 (Wilson et al., 2010). In Chile, González et al. (2018) reported a significant increase in the number, size, duration and simultaneity of large fires during the 2010–2015 ‘megadrought’ when compared to the 1990–2009 baseline.

Savannas consist of co-existing trees and grasses in tropical and temperate regions (Archibald et al., 2019). The global trend of woody encroachment reported in AR5 (Settele et al., 2014) is continuing (robust evidence, high agreement , very high confidence) (see Table SM2.1), with increases occurring in temperate savannas in North America (10–20% per decade) and tropical savannas in South America (8% per decade), Africa (2.4% per decade) and Australia (1% per decade) (O’Connor et al., 2014; Espírito-Santo et al., 2016; Skowno et al., 2017; Stevens et al., 2017; McNicol et al., 2018; Venter et al., 2018; Rosan et al., 2019). Additionally, the forest expansion into mesic savannas reported in AR5 (Settele et al., 2014) is continuing in Africa, South America and Southeast Asia (Marimon et al., 2014; Keenan et al., 2015; Baccini et al., 2017; Ondei et al., 2017; Stevens et al., 2017; Aleman et al., 2018; Rosan et al., 2019). Extreme high rainfall anomalies have also contributed to an increase in herbaceous and foliar production in the Sahel (Brandt et al., 2019; Zhang et al., 2019a).

New studies since AR5, using multiple study designs (experimental manipulations in lab and field, meta-analyses and modelling), attribute climate change increases in woody cover to elevated atmospheric CO2 (Donohue et al., 2013; Nackley et al., 2018; Quirk et al., 2019) and increased rainfall amount and intensity (robust evidence, high agreement ) (Venter et al., 2018; Xu et al., 2018b; Zhang et al., 2019a). Direct quantification of climate-change drivers is confounded with local LUC such as fire suppression (Archibald, 2016; Venter et al., 2018), heavy grazing (du Toit and O’Connor, 2014; Archer et al., 2017), removal of native browsers and, specifically, loss of mega-herbivores in Africa (medium evidence, medium agreement ) (Asner et al., 2016b; Daskin et al., 2016; Stevens et al., 2016; Davies et al., 2018). The relative importance of the climate- and non-climate-related causes of woody plants varies between regions, but there is general consensus that the impacts of climate change, specifically, increasing rainfall and rising CO2, are frequent and strong contributing factors of woody-cover increase (robust evidence, high agreement ).

Extensive woody-cover increases in non-forested biomes is reducing grazing potential (Smit and Prins, 2015) as well as changing the carbon stored per unit of land area (González-Roglich et al., 2014; Puttock et al., 2014; Pellegrini et al., 2016; Mureva et al., 2018) and the hydrological characteristics (Honda and Durigan, 2016; Schreiner-McGraw et al., 2020). Woody-cover encroachment also reduces biodiversity by threatening fauna and flora adapted to open ecosystems (Ratajczak et al., 2012; Smit and Prins, 2015; Pellegrini et al., 2016; Andersen and Steidl, 2019).

The global extent of grasslands is declining significantly because of climate change (medium confidence). In temperate and boreal zones, where about half of tree lines are shifting, they are overwhelmingly expanding poleward and upward, with an accompanying loss of montane and boreal grassland (robust evidence, high agreement ) whereas tropical tree lines have been generally stable (medium evidence, medium agreement ) (Harsch et al., 2009; Rehm and Feeley, 2015; Silva et al., 2016; Andela et al., 2017; Song et al., 2018; Aide et al., 2019; Gibson and Newman, 2019). The Eurasian steppes experienced a 1% increase in woody cover per decade since 2000 (Liu et al., 2021) and inner Mongolian grasslands in China experienced broad encroachment as well (Chen et al., 2015). Climatic drivers of woody expansion in temperature-limited grasslands, particularly alpine grasslands, are most frequently attributed to warming (robust evidence, high agreement , high confidence) (D’Odorico et al., 2012; Hagedorn et al., 2014), an increase in water and nutrient availability from thawing permafrost (medium evidence, high agreement ) (Zhou et al., 2015b; Silva et al., 2016) and rising CO2 (medium evidence, medium agreement ) (Frank et al., 2015; Aide et al., 2019). Interactions of LULCCs such as land abandonment, grazing management shifts and fire suppression with climate change are contributing factors (Liu et al., 2021)

Remote sensing shows overall increasing trends in both the annual maximum Normalized Difference Vegetation Index (NDVI) and annual mean NDVI in global grassland ecosystems between 1982 and 2011 (Gao et al., 2016). Multiple lines of evidence indicate that changes in grassland productivity are positively correlated with increases in mean annual precipitation (Hoover et al., 2014; Brookshire and Weaver, 2015; Gang et al., 2015; Gao et al., 2016; Wilcox et al., 2017; Wan et al., 2018). Increasing temperatures positively impact grassland production and biomass, especially in temperature-limited regions (Piao et al., 2014; Gao et al., 2016). However, it is expected that grasslands in hot areas will decrease production as temperatures increase (limited evidence, low agreement ) (Gang et al., 2015). Nevertheless, grassland responses to warming and drought are being ameliorated by increasing CO2 and associated improved water-use efficiency (Roy et al., 2016). For example, in a cool temperate grassland experiment, warming led to a longer growing season and elevated CO2 further extended growing by conserving water, which enabled most species to remain active longer (medium evidence, medium agreement ) (Reyes-Fox et al., 2014).

Overall declines of tropical forest cover (Kohl et al., 2015; Liu et al., 2015; Baccini et al., 2017; Harris et al., 2021), with declines more than triple the gains (Harris et al., 2021) have been driven primarily by deforestation and land conversion (robust evidence, high agreement ) (Lewis et al., 2015; Curtis et al., 2018; Assis et al., 2019). In opposition to this general trend, expansion of tropical forest cover into savannas and grasslands has occurred in Africa, South America and Australia (Marimon et al., 2014; Baccini et al., 2017; Ondei et al., 2017; Stevens et al., 2017; Aleman et al., 2018; Staver, 2018; Rosan et al., 2019).

Specific examples of climate change-driven range shifts of tropical deciduous forests upslope into alpine grasslands have been documented in the Americas (Chacón-Moreno et al., 2021; Jiménez-García et al., 2021) and Asia (Sigdel et al., 2018). However, tree line behaviours are diverse. A study in Nepal recorded that the tree line fomed by Abies spectabilis had been stable for more than a century, while the upper limit of large shrubs (Rhododendron campanulatum) had been advancing (Mainali et al., 2020). In both the Andes (Harsch et al., 2009) and Himalayas (Singh et al., 2021), most tree lines have been stable, leading (Rehm and Feeley, 2015) to postulate a ‘grass ceiling’ that has been difficult for trees to penetrate. The tree line shifts that have occurred are probably driven by interactions between changing land use (e.g., fire suppression) and climate changes such as increased rainfall, warming and elevated CO2 (via CO2 fertilisation or increases in water-use efficiency) (medium evidence, medium agreement ) (Cernusak et al., 2013; Huang et al., 2013; Van Der Sleen et al., 2015; Yang et al., 2016).

Increases in productivity of tropical forests (Gatti et al., 2014; Brienen et al., 2015; Baccini et al., 2017), Africa and southeast Asia (Qie et al., 2017) have been attributed to elevated CO2 (robust evidence, medium agreement ) (Ballantyne et al., 2012; Brienen et al., 2015; Sitch et al., 2015; Yang et al., 2016; Mitchard, 2018). The rates of these increases have been slowing down in the central Amazon (Brienen et al., 2015; de Meira Junior et al., 2020) and Southeast Asia (Qie et al., 2017). In contrast, the carbon sink (and hence the rate of biomass gain) in intact African forests was stable until 2010 and has only recently started to decline, indicating asynchronous carbon sink saturation in Amazonia and Africa, the difference being driven by rates of tree mortality (Hubau et al., 2020). At the global level, Hubau et al. (2020) argue that the carbon sink associated with intact tropical forests peaked in the 1990s and is now in decline.

Declines in productivity are most strongly associated with warming (Sullivan et al., 2020), reduced growth rates during droughts (Bennett et al., 2015; Bonai et al., 2016; Corlett, 2016), drought-related mortality (Brando et al., 2014; Zhou et al., 2014; Brienen et al., 2015; Corlett, 2016; McDowell et al., 2018), fire (Liu et al., 2017) and cloud-induced radiation limitation (robust evidence, high agreement ) (Deb Burman et al., 2020). Increases in the frequency and severity of droughts and shorter tree residence times due to increases in growth rates caused by elevated CO2 may be additional interactive factors increasing tree mortality (Malhi et al., 2014; Brienen et al., 2015). Vulnerability to drought varies between tree species and sizes, with large, older trees at the highest risk of mortality (McDowell et al., 2018; Meakem et al., 2018). Mortality risk also varies between forest types, with seasonal rainforests appearing to be the most vulnerable to drought (Corlett, 2016).

Lianas (long-stemmed woody vines) generally negatively impact trees, significantly reducing the growth of heavily infested trees (Reis et al., 2020). Lianas would benefit from climate change and disturbance (LingZi et al., 2014; Hodgkins et al., 2018). The extent of their suitable niche can increase (Taylor and Kumar, 2016), thereby decreasing forest biomass accumulation (robust evidence, high agreement ) (van der Heijden et al., 2013; Fauset et al., 2015; Estrada-Villegas et al., 2020).

Climate change continues to degrade forests by reducing resilience to pests and diseases, increasing species invasion, facilitating pathogen spread (Malhi et al., 2014; Deb et al., 2018) and intensifying fire risk and potential dieback (Lapola et al., 2018; Marengo et al., 2018). Drought, temperature increases and forest fragmentation interact to increase the prevalence of fires in tropical forests (robust evidence, high agreement ). Warming increases water stress in trees (Corlett, 2016) and, together with forest fragmentation, dramatically increases the desiccation of forest canopies—resulting in deforestation that then leads to even hotter and drier regional climates (Malhi et al., 2014; Lewis et al., 2015). Warming and drought increase the invasion of grasses into forest edges and increase fire risk (robust evidence, high agreement ) (Brando et al., 2014; Balch et al., 2015; Lewis et al., 2015). Droughts and fires additively increase mortality and, consequently, reduce canopy cover and above-ground biomass (Cross-Chapter Paper 7) (Brando et al., 2014, 2020; Balch et al., 2015; Lewis et al., 2015).

The AR5 found increased tree mortality, wildfire and plant phenology changes in boreal and temperate forests (Settele et al., 2014). Expanding on these conclusions, this assessment, using analyses of causal factors, attributes the following observed changes in boreal and temperate forests in the 20th and 21st centuries to anthropogenic climate change: upslope and poleward biome shifts at sites in Asia, Europe and North America (Section 2.4.3.2.1); range shifts of plants (Section 2.4.2.1); earlier blooming and leafing of plants (Section 2.4.2.4); poleward shifts in tree-feeding insects (Section 2.4.2.1); increases in insect pest outbreaks (Section 2.4.4.3.3); increases in the area burned by wildfire in western North America (Section 2.4.4.2.1); increased drought-induced tree mortality in western North America (Section 2.4.4.3.1); and thawing of the permafrost that underlies extensive areas of boreal forest (Section 2.4.3.9)(Section 2.3.2.5 in (Gulev et al., 2021)). Atmospheric CO2 from anthropogenic sources has also increased net primary productivity (NPP) (Section 2.4.4.5.1). In summary, anthropogenic climate change has caused substantial changes in temperate and boreal forest ecosystems, including biome shifts and increases in wildfire, insect pest outbreaks and tree mortality, at a global mean surface temperature (GMST) increase of 0.9°C above the pre-industrial period (robust evidence, high agreement ).

Other changes detected in boreal forests and consistent with, but not formally attributed to, climate change, include increased wildfire in Siberia (Section 2.4.4.2.3), long-lasting smouldering below-ground fires in Canada and the USA (Scholten et al., 2021), tree mortality in Europe (Section 2.4.4.3.3) and post-fire shifts of boreal conifer to deciduous broadleaf tree species in Alaska (Mack et al., 2021). From 1930 to 1960, boreal forest growth became limited more by precipitation than temperature in the Northern Hemisphere (Babst et al., 2019).

For some vegetation, changes in land use and management have exerted more influence than climate change. These include upslope and poleward forest shifts in Europe following the abandonment of timber harvesting or livestock grazing (Section 2.4.3.2.2), changes in wildfire in Europe affected by fire suppression, fire prevention and agricultural abandonment (Section 2.4.4.2.3), and forest species composition changes in Scotland due to nitrogen deposition from air pollution (Hester et al., 2019). Remote sensing suggests that the area of temperate and boreal forests increased in Asia and Europe between 1982 and 2016 (Song et al., 2018) and in Canada between 1984 and 2015 (Guindon et al., 2018), but forest plantations and regrowth are probable drivers (Song et al., 2018).

Globally, peatland ecosystems store approximately 25% (600 ± 100 GtC) of the world’s soil organic carbon (Yu et al., 2010; Page et al., 2011; Hugelius et al., 2020) and 10% of the world’s freshwater resources (Joosten and Clarke, 2002), despite only occupying 3% of the global land area (Xu et al., 2018a). The long-term role of northern peatlands in the carbon cycle was mentioned for the first time in IPCC AR4 (IPCC, 2007), while SR1.5 briefly mentioned the combined effects of changes in climate and land use on peatlands (IPCC, 2018b). New evidence confirms that climate change, including extreme weather events (e.g., droughts; Section 8.3.1.6), permafrost degradation (Section 2.3.2.5), SLR (Section 2.3.3.3) and fire (Section 5.4.3.2) (Henman and Poulter, 2008; Kirwan and Mudd, 2012; Turetsky et al., 2015; Page and Hooijer, 2016; Swindles et al., 2019; Hoyt et al., 2020; Hugelius et al., 2020; Jovani-Sancho et al., 2021; Veraverbeke et al., 2021), superimposed on anthropogenic disturbances (e.g., draining for agriculture or mining; Section 5.2.1.1), has led to rapid losses of peatland carbon across the world (robust evidence, high agreement ) (Page et al., 2011; Leifeld et al., 2019; Hoyt et al., 2020; Turetsky et al., 2020; Loisel et al., 2021). Other essential peatland ecosystem services, such as water storage and biodiversity, are also being lost worldwide (robust evidence, high agreement ) (Bonn et al., 2014; Martin-Ortega et al., 2014; Tiemeyer et al., 2017).

The switch from carbon sink to carbon source in peatlands globally is mainly attributable to changes in the depth of the water table, regardless of management or status (robust evidence, high agreement ) (Lafleur et al., 2005; Dommain et al., 2011; Lund et al., 2012; Cobb et al., 2017; Evans et al., 2021; Novita et al., 2021). Across the temperate and tropical biomes, extensive drainage and deforestation have caused widespread water table draw-downs and/or peat subsidence, as well as high CO2 emissions (medium evidence, high agreement ). Climate change is compounding these impacts (medium evidence, medium agreement ). For example, in Indonesia, the highest emissions from drained tropical peatlands were reported in the extremely dry year of the 1997 El Niño (810–2570 TgC yr -1) (Page et al., 2002) and the 2015 fire season (380 TgC yr -1) (Field et al., 2016). These prolonged dry seasons have also led to tree die-offs and fires, which are relatively new phenomena at these latitudes (medium evidence, high agreement ) (Cole et al., 2015; Mezbahuddin et al., 2015; Fanin and van der Werf, 2017; Taufik et al., 2017; Cole et al., 2019). Low soil moisture contributes to increased fire propagation (Section 12.4.2.2) (Dadap et al., 2019; Canadell et al., 2021), causing long-lasting fires responsible for smoke and haze pollution (robust evidence, high agreement ) (Ballhorn et al., 2009; Page et al., 2009; Gaveau et al., 2014; Huijnen et al., 2016; Page and Hooijer, 2016; Hu et al., 2018; Vadrevu et al., 2019; Niwa et al., 2021). Increases in fires and smoke lead to habitat loss and negatively impact regional faunal populations (limited evidence, high agreement ) (Neoh et al., 2015; Erb et al., 2018b; Thornton et al., 2018).

In large, lowland tropical peatland basins that are less impacted by anthropogenic activities (i.e., the Amazon and Congo river basins), the direct impact of climate change is that of a decreased carbon sink (limited evidence, medium agreement ) (Roucoux et al., 2013; Gallego-Sala et al., 2018; Wang et al., 2018a; Dargie et al., 2019; Ribeiro et al., 2021). As for the temperate and boreal regions, climatic drying also tends to promote peat oxidation and carbon loss to the atmosphere (medium evidence, medium agreement ) (Section 2.3.1.3.4) (Helbig et al., 2020; Zhang et al., 2020). In Europe, increasing mean annual temperatures in the Baltic, Scandinavia, and continental Europe (Section 12.4.5.1) have led to widespread lowering of peatland water tables at intact sites (Swindles et al., 2019), desiccation and die-off of sphagnum moss (Bragazza, 2008; Lees et al., 2019) and increased intensity and frequency of fires, resulting in a rapid carbon loss (Davies et al., 2013; Veraverbeke et al., 2021). Nevertheless, longer growing seasons and warmer, wetter climates have increased carbon accumulation and promoted thick deposits regionally, as reported for some North American sites (limited evidence, medium agreement ) (Cai and Yu, 2011; Shiller et al., 2014; Ott and Chimner, 2016).

In high-latitude peatlands, the net effect of climate change on the permafrost peatland carbon sink capacity remains uncertain (Abbott et al., 2016; McGuire et al., 2018b; Laamrani et al., 2020; Loisel et al., 2021; Sim et al., 2021; Väliranta et al., 2021). Increasing air temperatures have been linked to permafrost degradation and altered hydrological regimes (2.3.3.2; Figure 2.4a; 2.4.3.9; Box 5.1), which have led to rapid changes in plant communities and bio-geochemical cycling (robust evidence, high agreement ) (Liljedahl et al., 2016; Swindles et al., 2016; Voigt et al., 2017; Zhang et al., 2017b; Voigt et al., 2020; Sim et al., 2021). In many instances, permafrost degradation triggers thermokarst land subsidence associated with local wetting (robust evidence, high agreement ) (Jones et al., 2013; Borge et al., 2017; Olvmo et al., 2020; Olefeldt et al., 2021). Permafrost thaw in peatland-rich landscapes can also cause local drying through increased hydrological connectivity and runoff (Connon et al., 2014). In the first decades following thaw, increases in methane, CO2 and nitrous oxide emissions have been recorded from peatland sites, depending on surface moisture conditions (Schuur et al., 2009; O’Donnell et al., 2012; Elberling et al., 2013; Matveev et al., 2016; Euskirchen et al., 2020; Hugelius et al., 2020). Conversely, some evidence suggests increased peat accumulation after thaw (Jones et al., 2013; Estop-Aragonés et al., 2018; Väliranta et al., 2021). There is also a need to consider the impact of wildfire on permafrost thaw, due to its effect on soil temperature regime (Gibson et al., 2018), as fire intensity and frequency have increased across the boreal and Arctic biomes (limited evidence, high agreement ) (Kasischke et al., 2010; Scholten et al., 2021).

The CO2 emissions from degrading peatlands is contributing to climate change in a positive feedback loop (robust evidence, high agreement). At mid-latitudes, widespread anthropogenic disturbance led to large historical GHG emissions and current legacy emissions of 0.15 PgC yr -1 between 1990 and 2000 (limited evidence, high agreement ) (Maljanen et al., 2010; Tiemeyer et al., 2016; Drexler et al., 2018; Qiu et al., 2021). About 80 million hectares of peatland have been converted to agriculture, equivalent to 72 PgC emissions in 850–2010 CE (Leifeld et al., 2019; Qiu et al., 2021). In Southeast Asia (SEA), an estimated 20–25 Mha of peatlands have been converted to agriculture with carbon currently being lost at a rate of ~155 ± 30 MtC yr −1 (Miettinen et al., 2016; Leifeld et al., 2019; Hoyt et al., 2020). Extensive deforestation and drainage have caused widespread peat subsidence and large CO2 emissions at a current average of ~10 ± 2 tonnes ha -1 yr -1, excluding fires (Hoyt et al., 2020), with values estimated from point subsidence measurements being as high as 30–90 tonnes CO2 ha −1 yr −1 locally (robust evidence, high agreement ) (Wösten et al., 1997; Matysek et al., 2018; Swails et al., 2018; Evans et al., 2019; Conchedda and Tubiello, 2020; Anshari et al., 2021). On average, at the global scale, increases in GHG emissions from peatlands have primarily come from the compounded effects of LUC, drought and fire, with additional emissions from some thawing-permafrost peatlands (robust evidence, high agreement ).

Warming at high latitudes, documented in both AR4 and AR5, is leading to earlier snow and sea ice melt and longer growing seasons (IPCC, 2021a) which are continuing to alter tundra plant communities (medium evidence, high agreement ) (Post et al., 2009; Gauthier et al., 2013). Woody encroachment and increases in vegetation productivity, observed in both AR4 and AR5, are widespread and continuing. Both experiments and monitoring indicate that climate warming is causing increases in shrub, grass and sedge abundance, density, frequency, and height, with decreases in mosses and/or lichens (robust evidence, high agreement ) (Myers-Smith et al., 2011; Bjorkman et al., 2018; Bjorkman et al., 2019). Shrub growth is climate-sensitive and is greater in years with warmer growing seasons (Myers-Smith et al., 2015 ). Plant species that prefer warmer conditions are increasing (Elmendorf et al., 2015; Bjorkman et al., 2018), plant cover is increasing and bare ground is decreasing in long-term monitoring plots (Bjorkman et al., 2019; Myers-Smith et al., 2019). Animals such as moose, beavers and songbirds may already be responding to these vegetation changes by expanding their ranges northward or upslope into shrub tundra (Boelman et al., 2015; Tape et al., 2016a; Tape et al., 2016b; Tape et al., 2018).

In addition to direct warming, indirect effects of climate change, first found in AR4 and AR5, continue, such as thawed permafrost, altered hydrology and enhanced nutrient cycling, and these processes are causing pronounced vegetation changes (medium evidence, medium agreement ) (Schuur et al., 2009; Natali et al., 2012). Soil moisture status influences temperature sensitivity of plant growth and canopy heights (Myers-Smith et al., 2015 ; Ackerman et al., 2017; Bjorkman et al., 2018). In tundra ecosystems, permafrost thawing can decouple below-ground plant growth dynamics from above-ground dynamics, with below-ground root growth continuing until soils re-freeze in autumn (Cross-Chapter Paper 6) (Iversen et al., 2015; Blume-Werry et al., 2016; Radville et al., 2016).

In boreal coniferous areas, there has been an increase in the transporting of terrestrial-derived dissolved organic carbon (DOC) into rivers and lakes, which has caused increased opacity and a shift toward a brown colour (browning). There was little assessment of this in AR5. This process is driven by climate change, and stems from hydrological intensification, greening of the Northern Hemisphere and degradation of carbon sinks in peatlands (robust evidence, high agreement ) (Solomon et al., 2015; Catalán et al., 2016; de Wit et al., 2016; Finstad et al., 2016; Creed et al., 2018; Hayden et al., 2019). These factors enhance terrestrial productivity, alter vegetation communities and affect the hydrological control of the production and transport of DOC (Weyhenmeyer et al., 2016). Non-climate-related drivers of browning are: declining atmospheric sulphur deposition, forestry practices and LULCCs (see Table SM2.1 for detail).

Browning creates a positive feedback to climate by absorbing photosynthetically active radiation, which accelerates upper water (epilimnetic) warming (Solomon et al., 2015). Browning of lakes leads to shallower and more stable thermoclines, and thus overall deep water cooling (Solomon et al., 2015; Williamson et al., 2015), and can provoke a transition of the seasonal mixing regime from a mixed lake (polymictic) to one that is seasonally stratified (Kirillin and Shatwell, 2016).

The ecological responses of browning are a concomitant effect of climate change and nutrient status. Results from long-term, large-scale lake experiments have been variable, showing both strong synergistic effects (Urrutia-Cordero et al., 2016) and no significant effects of browning on plankton community food webs (Rasconi et al., 2015). Browning has driven a shift from auto- to heterotrophic/mixotrophic-based production (Urrutia-Cordero et al., 2017) and supports heterotrophic metabolism of the bacterial community (Zwart et al., 2016). Browning may also accelerate primary production through the input of nutrients associated with dissolved organic matter (DOM) in nutrient-poor lakes and increase cyanobacteria, which cope better with low light intensities (Huisman et al., 2018) and toxin levels (Urrutia-Cordero et al., 2016). However, the synergistic impacts of browning and climate change on aquatic communities depends on regional precipitation patterns (Weyhenmeyer et al., 2016), watershed type (de Wit et al., 2016) and the length of the food chain (Hansson et al., 2013). Quantitative attribution of browning to climate change remains difficult (medium evidence, medium agreement ).

In summary, new studies since AR5 have explicitly estimated the effects of warming and browning on freshwaters in boreal areas, with complex positive and negative repercussions on water temperature profiles (lower vs. upper water) (high confidence) and primary production (medium confidence).

Methods for projecting the impacts of climate change on biodiversity can be classified into three types: (1) statistical models such as SDMs (Elith and Leathwick, 2009); (2) mechanistic or process-based models (Chuine and Régnière, 2017) and (3) trait-based models (Pacifici et al., 2015). It is only recently that models have been developed looking at lower levels of warming like 1.5°C (Hoegh-Guldberg et al., 2018; Warren et al., 2018).

SDMs or niche-based models assess potential geographic areas of suitable climate for the species in current conditions and then project them into future conditions (Trisurat, 2018; Vieira et al., 2018). There are limitations in all models and it is critical that modellers understand the assumptions, proper parameterization and limitations of each model technique, including differences between climate models, emission scenarios or RCPs and baselines (Araujo et al., 2019). Several systems automate the development of SDMs, including R-packages (Beaumont et al., 2016; Hallgren et al., 2016), and other model types (Foden et al., 2019) and aid in the use of climate model data (Suggitt et al., 2017), including allowing for connectivity constraints (Peterson et al., 2013). Buisson et al. (2010) found most variation in model outputs stems from differences in design, followed by general circulation models (GCMs).

Mechanistic approaches, also known as process-based models, project the responses of species to climate changes by explicitly incorporating known biological processes, thresholds and interactions (Morin and Thuiller, 2009; Maino et al., 2016). Mechanistic models are able to accommodate a broad range of mechanisms of climate change impacts and include species-specific characteristics such as dispersal distances, longevity, fecundity, genetic evolution and phenotypic plasticity. However, sufficient knowledge is available for only a few well-studied species. Species’ traits have been used to more broadly estimate potential climate change impacts (Foden et al., 2013; Cizauskas et al., 2017).

Most models are on a large scale (20–50 km), and so cannot capture micro-climatic refugia generated by diversities of slope aspect, elevation or shade (Suggitt et al., 2015; Suggitt et al., 2018). In analysing records of 430 climate-threatened and range-declining species in England, (Suggett et al., 2015; Suggitt et al., 2018) showed that topographic diversity reduced population declines most strongly in areas experiencing the most local warming and in the species most sensitive to warming. Under these circumstances, topographic diversity reduced the risk of population extinction by 22% for plants and 9% for insects.

None of the modelling techniques are predictions of the future, they are rather projections of possible futures. To date, only a few studies have validated model performance against observations, but the studies that have been conducted do generally validate models using either SDMs or process-based models (Johnston et al., 2013; Fordham et al., 2018). SDMs should be considered as hypotheses of what a future world might look like if the climate projections came to pass. Suggestions have been made on how to start bringing more biotic interactions into SDMs (Early and Keith, 2019), but limited basic ecological understanding of interactions, along with limits on computation and funding, constrains how far and how fast these modelling techniques can advance.

Table 2.3 | Assessing uncertainty in detection and attribution of observed changes in terrestrial and freshwater species and ecosystems to climate change. The lines of evidence used to support given uncertainty statements, including confidence statements, of the attribution of key conclusions on observed biological changes to climate change and increased atmospheric CO2. Icons represent lines of evidence. This is a summary table that is fully detailed in Table SM2.1.

In order to add realism and reliability to risk assessments at the species and community levels, non-modelling approaches, based on known biological traits or processes as well as on expert opinion (Camac et al., 2021), are used to temper model outputs with ground-based validation. Trait-based assessment approaches use species’ biological characteristics as predictors of sensitivity, adaptive capacity and extinction risk due to climate change. Climate exposure can be estimated using GIS-based modelling, statistical programs or expert judgment (Chin et al., 2010). These trait-based approaches are widely applied to predict responses of biodiversity to climate change because they do not require modelling expertise or detailed distibutional data (Pacifici et al., 2015; Willis et al., 2015). Most of these methods have not been independently validated and do not allow direct comparison of vulnerability and risk across taxonomic groups.

Some studies have combined two or three approaches for the assessment of climate change risk to biodiversity, in order to capture the advantages of each and avoid their limitations. Warren et al. (2013) used combinations of SDMs and trait-based approaches to estimate the proportions of species losing their climatically suitable ranges under the various future scenarios of climate and dispersal rate. Similarly, spatial projections of exposure to climate change were combined with traits to assess the vulnerability of sub-Saharan amphibians (Garcia et al., 2014). Laurance et al. (2012) combined 31 functional groups of species and 21 potential drivers of environmental change, in order to assess both the ecological integrity and threats to protected tropical areas on a global scale. Keith et al. (2014) used a combination of three approaches (SDMs–trait–mechanistic) to determine how long before extinction a species would become eligible for listing as threatened, based on the IUCN Red List criteria.

Multiple studies predict increases in disease incidence or geographic and phenological changes of pathogens, vectors and reservoir host species due to climate change with or without other non-climatic variables (González et al., 2010; Moo-Llanes et al., 2013; Roy-Dufresne et al., 2013; Liu-Helmersson et al., 2014; Laporta et al., 2015; Ryan et al., 2015; Haydock et al., 2016; Hoover and Barker, 2016; Prist et al., 2017; Blum and Hotez, 2018; Dumic and Severnini, 2018; Hundessa et al., 2018; Ryan et al., 2019; Ryan et al., 2021). However, models predicting changes in infectious disease risk are complex and sometimes produce conflicting results and lack consensus (Caminade et al., 2014; Giesen et al., 2020). For example, malaria is projected to increase in some regions of Africa, Asia and South America by the end of the 21st century if public health interventions are not sufficient, but is also forecasted to decrease in some higher-risk areas (Cross-Chapter Box Illness in this chapter) (Peterson, 2009; Caminade et al., 2014; Ryan et al., 2015; Khormi and Kumar, 2016; Leedale et al., 2016; Murdock et al., 2016; Endo and Eltahir, 2020; Mordecai et al., 2020).

While malaria risk is predicted to decrease in some lowland tropical areas as temperatures become too hot for vector or parasite development, other warm-adapted diseases, like dengue and Zika, transmitted by A. aegypti, are predicted to increase (Cross-Chapter Box Illness in this chapter, chapter 7) (Ryan et al., 2019; Ryan et al., 2021). In more temperate regions, arboviruses and other VBDs with wider thermal breadths, such as West Nile fever, Ross River fever and Lyme disease, are predicted to increase with climate warming (Ogden et al., 2008; Leighton et al., 2012; Shocket et al., 2018; Shocket et al., 2020; Couper et al., 2021). Drought can exacerbate these effects of temperature (Paull et al., 2017).

A global analysis of 7346 wildlife populations and 2021 host–parasite combinations found that organisms adapted to cool and mild climates are likely to experience increased risks of outbreaks along with climate warming, while warm-adapted organisms may experience a lower disease risk, providing further support for predictions that climate change will increase the transmission of infectious diseases at higher latitudes across a taxonomically diverse array of pathogens (robust evidence, high agreement ) (Cohen et al., 2020). A study examining the future risk of arboviruses (chikungunya, dengue, yellow fever and Zika viruses) spread by A. aegypti and A. albopictus projected increased disease risk due to interactions of multiple variables, including increased human connectivity, urbanisation and climate change (Kraemer et al., 2019), although vector species’ ranges will broaden only slightly (Campbell et al., 2015).

In sum, climate change is expected to expand and redistribute the burden of vector-borne and other environmentally transmitted diseases of wild animals, domesticated animals and humans, by shifting many regions toward the thermal optima of VBD transmission for multiple parasites, thereby increasing risk of transmission, while pushing temperatures above optimal and towards upper thermal limits for other vectors and pathogens, thus decreasing their transmission (high confidence) (see also chapter 7) (Mordecai et al., 2019; Mordecai et al., 2020). These effects are mediated by other human impacts such as LUC, mobility, socioeconomic conditions and vector and pathogen control measures (Parham et al., 2015; Tjaden et al., 2018).

Shifts in terrestrial biome and changes in ecosystem processes in response to climate change are most frequently projected with dynamic global vegetation models (DGVMs) or land-surface models that form part of ESMs, which use gridded climate variables, atmospheric CO2 concentration and information on soil properties as input variables. Since AR5, most DGVMs have been upgraded to capture carbon–nitrogen cycle interactions (e.g., (Le Quéré et al., 2018), many also include a representation of wildfire and fire–vegetation interactions (Rabin et al., 2017) and a small number now also account for land management (e.g., wood removal from forests and crop fertilisation harvest of irrigation (Arneth et al., 2017). Other forms of disturbance, such as tree mortality, in response to, for example, episodic weather extremes or insect pest outbreaks, are relatively poorly represented or not at all, although they demonstrably impact calculated carbon cycling (Pugh et al., 2019a). Simulated biome shifts are generally in agreement in projecting broad patterns on a global scale but vary greatly regarding the simulated trends in historical and future carbon uptake or losses, both regionally and globally (Chang et al., 2017; Canadell et al., 2021).

Similar to other models, models to project large-scale changes in vegetation and ecosystem processes have to deal with structural uncertainty (associated with the choice and the representation of processes in models), input-data uncertainty (associated with variability in initial conditions and parameter values) and error propagation (associated with coupling models) (Rounsevell et al., 2019). The IPBES methodological assessment report on scenarios and models of biodiversity and ecosystem services provides a comprehensive overview over the relevant issues (Ferrier et al., 2016).

In order to assess performance, most models have been individually evaluated against a range of observations. Moreover, in the annual updates of the global carbon budgets, a model has to meet a small set of basic criteria to have its output included (Le Quéré et al., 2018). More systematic benchmarking approaches have also been proposed that utilise a range of different datasets (Kelley et al., 2013; Chang et al., 2017) to assess multiple simulated processes. These methods, in principle, facilitate assigning quality scores to models based on their overall performance (Kelley et al., 2013). So far, this scoring does not yet allow a clear quality ranking of models, since individual DGVMs tend to score well for some variables and badly for others. A recent comparison of global fire–vegetation model outputs was also able to clearly identify outliers when using a formalised benchmarking and scoring approach (Hantson et al., 2020). However, benchmarking does not address sources of uncertainty and it would be advisable to perform ‘perturbed-physics’ experiments, in which multiple model parameters are varied in parallel more frequently as a means to test parameter-value uncertainty (Wramneby et al., 2008; Booth et al., 2012; Lienert and Joos, 2018).

Species diversity impacts ecosystem functioning and hence ecosystem services (Hooper et al., 2012; Mokany et al., 2016). So far, however, integrated modelling of ecosystem processes and biodiversity across multiple trophic levels and food webs is in its infancy (Harfoot et al., 2014). Whether or not the enhanced integration of state, function and functional diversity across multiple trophic levels in models will markedly alter projections of how ecosystems respond to climate change thus remains an open research question.

Beyond dynamic simulation of biome shifts and carbon cycling, which are important aspects of climate regulation, DGVMs can also provide information on a number of variables closely linked to other ecosystem services such as water availability, air quality or food provisioning (Krause et al., 2017; Rabin et al., 2020). However, they are not intended to provide a comprehensive assessment of ecosystem services. For these, other approaches applied but, to date, these are mostly applied on regional scales and are only weakly dynamic (Ferrier et al., 2016).

Climate change and the associated change in atmospheric CO2 levels already exacerbate other human-caused impacts on the structure and composition of land and freshwater ecosystems, such as LULCC, nitrogen deposition and pollution. The relative importance of these drivers for ecosystems over the coming decades will likely differ between biomes, but climate change and atmospheric CO2 will be pervasive unless there is a rapid lowering of fossil-fuel emissions and warming (high confidence) (Pereira et al., 2010; Warren, 2011; Ostberg et al., 2013; Davies-Barnard et al., 2015; Pecl et al., 2017; Ostberg et al., 2018). Global vegetation and ESMs agree on climate change-driven shifts of biome boundaries of potentially hundreds of kilometres over this century, combined with several substantial alterations that take place within biomes (e.g., changes in phenology, canopy structure and functional diversity, etc.). Large discrepancies exist between models and between scenarios regarding the region and the speed of change (Gonzalez et al., 2010; Pereira et al., 2010; Pecl et al., 2017), but robust understanding is emerging in that the degree of impact increases in high-emission and high-warming scenarios (high confidence) (Figure 2.9).

Substantial changes in vegetation structure and ecosystem processes are already happening (see Section 2.4). Many of these observations have already been projected to take place as early as at least IPCC AR3 (Rosenzweig et al., 2007), and can they now be increasingly tested for their robustness with observational evidence. These multiple changes in response to warming (and changes in precipitation and increasing atmospheric CO2 levels that go hand-in-hand with warming) are further expected for already relatively small additional temperature increases. In particular, in cold (boreal and tundra) regions, as well as in dry regions (high confidence), alterations of 2–47% of the areal extent of terrestrial ecosystems in scenarios of <2°C warming above pre-industrial levels have been projected, increasing drastically with higher-warming scenarios (Warren, 2011; Wårlind et al., 2014). More recent work, applying also probabilistic methods, confirm the risk of drastic changes in vegetation cover (e.g., forest to non-forest or vice versa) at the end of the 21st century even for approximately 2°C warming scenarios, especially in tundra, and also in tropical forest and savannah regions, with more subtle changes (within a given biome type) likely to occur in all regions (Ostberg et al., 2013; Ostberg et al., 2018). Model studies have found 5–20% of terrestrial ecosystems affected by warming of around 2°C–3°C, increasing to above one-third at a warming of 4°C–5°C (Ostberg et al., 2013; Warszawski et al., 2013).

In general, vegetation types are projected to be moving into their ‘neighbouring’ climates, depending on whether temperature or precipitation is expected to be the predominant factor and how vegetation interacts with the increasing CO2 levels in the atmosphere (Wårlind et al., 2014; Scheiter et al., 2015; Schimel et al., 2015; Huntzinger et al., 2017). For instance, boreal or temperate forest vegetation is simulated to migrate polewards, closed tropical (moist) forest is expected to transition towards dry tropical forest types, while climate-driven degradation might expand arid vegetation cover (Sections 2.5.2.2–2.5.2.9). However, ‘novel ecosystems’, that is, communities with no current or historical equivalent because of the novel combinations of abiotic conditions under climate change, are expected to be increasingly common in the future (medium confidence), although the regions where these novel ecosystems might emerge are still disputed (Reu et al., 2014; Radeloff et al., 2015; Ordonez et al., 2016). The possibility of these novel ecosystems and the communities that live within them are a challenge for current modelling of ecosystem shifts, and new approaches to conservation will be required that are designed to adapt to rapid changes in species composition and the ensuing challenges.

Shifts in arid system structure and functioning that have been observed to date (Section 2.4.3.3) are projected to continue (medium confidence). These include widespread woody plant encroachment, notably in savanna systems in Africa, Australia and South America, and are attributed to interactions of LULCC, climate change and CO2 fertilisation effects (Fensholt et al., 2012; Fang et al., 2017; Stevens et al., 2017). Arid Mongolian steppe grassland did not respond to experimentally elevated CO2 (Song et al., 2019). Woody encroachment is projected to continue or not reverse in North American drylands (Caracciolo et al., 2016) and southern African arid ecosystems (Moncrieff et al., 2014b). Dryland woody encroachment may increase carbon stocks, depending on emissions scenario (Martens et al., 2021), but reduce soil water and biodiversity of grassland-dependent species diversity (Archer et al., 2017). Warm season (C4) grass expansion into arid shrublands risks sudden ecosystem transformation due to introduced wildfire (Bradley et al., 2016), a risk anticipated for grass-invaded desert ecosystems of Australia and the southwestern USA (Horn and St. Clair, 2017). Novel fire regimes in grassy shrublands have enhanced grass cover locally in the southern African Nama-Karoo (du Toit et al., 2015).

Range retractions are projected for endemic plants in southern Africa (Young et al., 2016) and dry woodlands in Morocco (Alba-Sánchez et al., 2015). Increasing thermal stress is projected to increase woody plant mortality in the Sonoran Desert ecosystems (Munson et al., 2016) and facilitate perennial grass replacement by xeric shrubs in the southwestern USA (Bestelmeyer et al., 2018). Ecological effects may occur rapidly when extreme events compound long-term trends (Hoover et al., 2015), but evolve more slowly as opportunity costs accumulate due to warming (Cross-Chapter Paper 3) (Cunningham et al., 2021).

The regions containing MTEs all show high confidence in projected increases in the intensity and frequency of hot extremes and decreases in the intensity and frequency of cold extremes, and medium confidence in increasing ecological drought due to increased evapotranspiration (in all regions) and reduced rainfall (excluding California, USA, where model agreement is low) (see WGI Chapter 11). Projections also show a robust increase in the intensity and frequency of heavy precipitation in the event of ≥2°C warming for MTEs in South Africa, the Mediterranean Basin and California, USA, but are less clear for Australia and Chile (Seneviratne et al., 2021).

MTEs are characterised by the distinctive seasonal timing of precipitation and temperature, and the disruption of this regime is likely to be critical for their maintenance. Unfortunately, projections of changes in rainfall seasonality have received less attention and are far more uncertain than many other aspects of climate change (Pascale et al., 2016; Breinl et al., 2020), thus limiting our ability to predict the ecological consequences of climate change in MTEs. Responses to experimental manipulation of rainfall seasonality show the potential for shifts in plant functional composition and diversity loss, but results vary with soil type (van Blerk et al., 2021).

Unfortunately, global- and regional-scale dynamic vegetation models show a poor performance for large areas of MTEs, because they do not characterise shrub and Crassulacean acid metabolism (CAM)-photosynthetic plant functional types well (Moncrieff et al., 2015). Furthermore, the grain of these models is too coarse for quantifying impacts to many vegetation formations which are patchy or of limited extent (e.g., small stands of trees). There is high confidence that observations of high mortality in trees and other growth forms, reduced reproductive and recruitment success, range shifts, community shifts towards more thermophilic species and type conversions are set to continue, due to either direct climate impacts through drought and other extreme weather events or to their interaction with factors like fire and pathogens (Sections 2.4.3.5, 2.4.3.6; 2.4.3.7; 2.4.4.2; 2.4.4.3; 2.5.2.5, 2.5.2.6, 2.5.2.7, 2.5.4).

Fire is a key driver across most MTEs due to summer-dry conditions. Climate projections for the MTEs translate into high confidence that periods of low fuel moisture will become more severe and prolonged, and that episodes of extreme fire weather will become more frequent and severe (see (Douville et al., 2021; Seneviratne et al., 2021)). This will lead to the birth of novel fire regimes in MTEs, characterised by an increase in the probability of greater burned area and extreme wildfire events (e.g., megafires), with associated loss of human life and property, long-term impacts on ecosystems and acceleration of the possible loss of resilience and capacity to recover (Abatzoglou and Williams, 2016; González et al., 2018; Boer et al., 2020; Moreira et al., 2020; Nolan et al., 2020; Duane et al., 2021; Gallagher et al., 2021).

Fire is virtually certain to have additional impacts through compound events (see Section 11.8 in (Seneviratne et al., 2021)). Extreme post-fire weather is extremely likely to continue to impact diversity (Slingsby et al., 2017), retard vegetation regrowth (Slingsby et al., 2020a) and accelerate vegetation shifts (Batllori et al., 2019). Any increases in the intensity and frequency of heavy precipitation are highly likely to compromise soil stability in recently burnt areas (Morán-Ordóñez et al., 2020). The impacts of fire often depend on interactions with non-climatic factors such as habitat fragmentation (Slingsby et al., 2020b) and management (Steel et al., 2015) or the spread of flammable exotic plantation forestry and invasive species (Kraaij et al., 2018; McWethy et al., 2018). Managing these factors provides opportunities for adaptation and mitigation (Moreira et al., 2020). (See sections 2.4.4.2 and 2.5.3.2).

Human adaptation and mitigation responses to climate change may create additional threats to MTEs. MTEs have dry summers by definition, posing a challenge for the year-round supply of water to growing human populations and agriculture. With recent major droughts in all MTEs (Section 2.4.3.6), there is increasing reliance on groundwater for the bulk of the water supply (Kaiser and Macleod, 2018). The majority of groundwater systems have exceeded or are rapidly approaching their environmental flow limits (de Graaf et al., 2019), threatening human populations and ecosystems that depend on these systems for their persistence through unfavourable climatic conditions (McLaughlin et al., 2017). Similarly, much of the MTEs are open shrublands and grasslands and proposed extensive tree-planting to sequester atmospheric CO2 could result in a loss of biodiversity and threaten water security (Doblas-Miranda et al., 2017; Bond et al., 2019).

Worldwide, woody cover is increasing in savannas (Buitenwerf et al., 2012; Donohue et al., 2013; Stevens et al., 2017), as a result of interactions of elevated CO2 and altered fire and herbivory impacts, some of which stems from LULCC (high confidence) (see Section 2.4.3.5; Cross-Chapter Paper 3.2) (Venter et al., 2018; Wu et al., 2021). In some regions, altered climate may also contribute (Cross-Chapter Paper 3.2). Elevated CO2 benefits plants with C3 photosynthesis (often woody plants), more than C4 species (Moncrieff et al., 2014a; Scheiter et al., 2015; Knorr et al., 2016a). Increases in woody vegetation in grassy ecosystems could provide some carbon increase (medium confidence) (Zhou et al., 2017; Mureva et al., 2018), but is expected to decrease biodiversity (Smit and Prins, 2015; Abreu et al., 2017; Andersen and Steidl, 2019) and water availability (Honda and Durigan, 2016; Stafford et al., 2017) and alter ecosystem services like grazing and wood provision (high confidence) (Anadón et al., 2014b).

The relative importance of climate, disturbance (e.g., fire/herbivory) and plant feedbacks in shaping present and future savanna distribution varies between continents (Lehmann et al., 2014), which makes projections of changing the biome extent challenging (Moncrieff et al., 2016). It has been shown that simulation studies that do not account for CO2 interactions but only consider climate change impacts do not realistically capture the future distribution of savannas (high confidence) (Higgins and Scheiter, 2012; Moncrieff et al., 2016; Scheiter et al., 2020). Due to the continued strong effect of CO2 on tree and shrub-to-grass ratios in future, models suggest a loss of savanna extent and conversion into closed canopy forest/thicket and an expansion of savanna-type vegetation into arid grasslands (Wårlind et al., 2014; Moncrieff et al., 2016). In arid savannas and their interface to grasslands, survival of woody vegetation (which may be stimulated to grow by increasing CO2) will depend on their capacity to survive potentially more severe and frequent droughts (Sankaran and Staver, 2019). Across a range of models, for RCP4.5 future climate change and CO2 concentrations, savanna expanse declines by around 50% (converting to closed canopy systems) by 2070 in Africa and South America, 25% in Asia and with small changes in Australia (Moncrieff et al., 2016; Kumar et al., 2021). Future fire-spread is expected to be reduced with increased woody dominance (Scheiter et al., 2015; Knorr et al., 2016b; Scheiter et al., 2020), feeding back to further increase tree-to-grass ratios (high confidence).

Like the tropical forest biome, savannas are at a high risk, given the projected climate changes in combination with LULCC (see Cross-Chapter Paper 3). About 50% of the Brazilian Cerrado has been converted to agricultural land and pastures (Lehman and Parr, 2016), and African savannas have been proposed to follow a similar tropical agricultural revolution pathway to enhance agronomic prosperity (Ryan et al., 2016). In fact, indirect climate change impacts arising from mitigation efforts may be particularly perilous to savannas; extensive tree-planting to restore ecosystems and remove CO2 from the atmosphere, as pledged, for example, under the African Forest Restoration Initiative, could lead to carbon losses and the loss of biodiversity as well as damage the water balance if trees are planted on what was naturally grassland or savanna (Box 2.2; FAQ 2.6) (Bond et al., 2019).

Key factors affecting the future distribution of tropical humid and dry forests are amounts and seasonalities of precipitation, increased temperatures, prolonged droughts and droughted-moderated fires (robust evidence, high agreement ) (Bonai et al., 2016; Corlett, 2016; Lyra et al., 2017; Anderson et al., 2018; da Silva et al., 2018 ; Fontes et al., 2018; O’Connell et al., 2018; Aguirre-Gutiérrez et al., 2019; Bartlett et al., 2019; Brando et al., 2019; Stan and Sanchez-Azofeifa, 2019). The probability of severe drought is projected to quadruple in natural areas in Brazil with >2°C warming (Barbosa and Lakshmi Kumar, 2016; Marengo et al., 2020). Most multi-model studies assuming rapid economic growth/business-as-usual scenarios (A2, A1B and RCP8.5) show an increase in future woody biomass and areas of woody cover towards the end of the 21st century in temperate regions (Boit et al., 2016; Nabuurs et al., 2017) and tropical forests in East Africa (Ross et al., 2021) but a decrease in the remaining tropical regions (Anadón et al., 2014a; Boit et al., 2016; Lyra et al., 2017; Nabuurs et al., 2017; Maia et al., 2020). Terrestrial species are predicted to shift to cooler temperatures and higher elevations (Pecl et al., 2017). Tropical species are more susceptible to climate warming than temperate species (Rehm and Feeley, 2016; Sentinella et al., 2020). This susceptibility will be exacerbated by road-building increasing the ease of access into forests (Brinck et al., 2017; Taubert et al., 2018; Bovendorp et al., 2019; Senior et al., 2019). Furthermore, most tropical cloud forest species are unable to invade grasslands and this will increase the risk of extinctions in tropical cloud forests (Rehm and Feeley, 2015).

SLR as the result of climate change is likely to influence mangroves in all regions, with greater impact on North and Central America, Asia, Australia and East Africa than on West Africa and South America (robust evidence, high agreement ) (Alongi, 2015; Ward et al., 2016). On a small scale, mangroves are potentially moving landward (Di Nitto et al., 2014), while on a large scale they will continue to expand poleward (Alongi, 2015).

Most simulations predict a significant geographical shift of transition areas between tropical forests and savanna in the tropical and subtropical Americas and Himalayas (Anadón et al., 2014a; Rashid et al., 2015). Forest dieback, as postulated for the Amazon region, does not occur in the majority of simulations (Malhi et al., 2009; Poulter et al., 2010; Rammig et al., 2010; Higgins and Scheiter, 2012; Huntingford et al., 2013; Davies-Barnard et al., 2015; Sakschewski et al., 2016; Wu et al., 2016a). Model projections of future biodiversity in tropical forests are rare. Arguably, species are most vulnerable to climate change effects at higher altitudes or at the dry end of tropical forest occurrence (medium evidence, medium agreement ) (Krupnick, 2013; Nobre et al., 2016; Trisurat, 2018). Tropical lowlands are expected to lose plant species as temperatures rise above species’ heat tolerance, but could also generate novel communities of heat-tolerant species (robust evidence, high agreement ) (Colwell et al., 2008; Trisurat et al., 2009; Trisurat et al., 2011; Krupnick, 2013; Zomer et al., 2014a; Zomer et al., 2014b; Sullivan et al., 2020; Pomoim et al., 2021).

Statistical models that correlate data on species abundance with information on human pressures, such as LUCs (Srichaichana et al., 2019), population density (Leclère et al., 2020) and hunting (Mockrin et al., 2011), found in tropical and subtropical forests that birds, invertebrates, mammals and reptiles show a decline in their probability of presence with declining forest cover, particularly pronounced in forest specialists or species with narrow ranges (Newbold et al., 2014). Different soil fauna groups showed different responses in abundance and diversity to climate change conditions (Coyle et al., 2017; Facey et al., 2017) but these responses can impact decomposition rates and biogeochemical cycles (medium evidence, low agreement ).

Invasive plant species are predicted to expand upward by 500–1,500 m in the western Himalayas (Thapa et al., 2018), and by 6–35% yr -1 from the current extent in South America (robust evidence, high agreement ) (Bhattarai and Cronin, 2014). Global assessment (Wang et al., 2017) also revealed that ecoregions of high-elevation tropical forests and subtropical coniferous forests have a high risk of invasive plant expansion in low-CO2- emission scenarios, with negative impacts on ecosystem functioning and local livelihoods (Shrestha et al., 2019).

The impact of unsustainable land use on tropical forests continues in all regions (see Cross-Chapter Paper 7). Projected climate changes will not only impact biodiversity but also the livelihoods of affected people (robust evidence, high agreement ). Increased drought drives crop failures that cause local communities to expand their agricultural area by further clearing native forests (Desbureaux and Damania, 2018). Climate change is projected to enlarge the area of suitability for booming tree crops such as oil palm, acacia, Eucalyptus and rubber (Koninck et al., 2011; Cramb et al., 2015; Nath, 2016; Hurni et al., 2017; Li et al., 2017; Varkkey et al., 2018). An increase of 8% in the area of rubber plantations in Yunnan Province, China, between 2002–2010 to higher altitudes due to decreased environmental limits, has potentially increased pressure on the remaining biodiversity both within and outside of protected areas (Zomer et al., 2014a). As a consequence, the suitable area for mammals is projected to be reduced by 47.7% (RCP2.6) and 67.7% (RCP8.5) by 2070, with large variability depending on the different species (Cross-Chapter Paper 7) (Brodie, 2016). To minimize these potential threats, the Yunnan provincial government has identified suitable areas for the establishment of national parks, including the Asian Elephant National Park since 2006. And the government of China developed a national park system in 2013 across the country.

As in the Arctic, warming substantially exceeding the global average has already been observed in the northern parts of the temperate and boreal forest zone (Gauthier et al., 2015), and is projected to continue (see Cross-Chapter Paper 6 and (Lee et al., 2021)). As a consequence, boreal tree species are expected to move northwards (or in mountainous regions, upwards) into regions dominated by tundra, unless constrained by edaphic features, and temperate species are projected to grow in regions currently occupied by southern boreal forest (high confidence). In both biomes, deciduous trees are simulated to grow increasingly in regions currently dominated by conifers (Wårlind et al., 2014; Boulanger et al., 2017). These simulation results have been supported by observational examples. In eastern Siberia, fire disturbance of larch-dominated forest was followed by recovery to birch-dominated forest (Stuenzi and Schaepman-Strub, 2020). In Alberta, lodgepole pine (Pinus contorta) lost its dominant status after attacks by mountain pine beetles (Dendroctonus ponderosae) caused the canopy to switch to non-pine conifers and broadleaf trees (Axelson et al., 2018). In contrast to the examples above, some boreal forests have proven resilient to disturbances including recent unprecedented insect outbreaks (Campbell et al., 2019a; Prendin et al., 2020).

Reforestation, either natural or anthropogenic, leads to summer cooling and winter warming of the ground, while forest thinning or removal by fire has reverse effects, deepening the upper layer that is free of permafrost (Stuenzi et al., 2021a). Interactions between permafrost and vegetation are important. For example, trees in the east Siberian taiga obtained water mostly from rain in wet summers and permafrost melt water in dry summers (Sugimoto et al., 2002), suggesting that these forests will be particularly vulnerable to the combination of drought with the retraction of permafrost further underground due to climate warming.

The overall effect of climate change on the extent of northern peatlands is still debated (limited evidence, low agreement ). It is expected that climate change will drive the expansion of high-latitude peatlands poleward of their present distribution due to warming, permafrost degradation and glacier retreat, which could provide new land and conditions favourable for peat development (limited evidence, medium agreement ) (Zhang et al., 2017b), as seen during the last de-glacial warming (robust evidence, high agreement ) (MacDonald et al., 2006; Jones and Yu, 2010; Ratcliffe et al., 2018). Peatland area loss (shrinking) near the southern limit of their current distribution or in areas where the climate becomes unsuitable is also expected (medium evidence, medium agreement ) (Section 2.3.4.3.2) (Finkelstein and Cowling, 2011 Gallego-Sala and Prentice, 2013; Schneider et al., 2016; Müller and Joos, 2020) (Müller and Joos, 2021), but they could persist if moisture is maintained via their capacity to self-regulate. In western Canada, a study suggests that peatlands may persist until 2100, even though the climate will be less suitable (Schneider et al., 2016). Simulations suggest that climate change-driven increases in temperature and atmospheric CO2 could drive reductions in the northern peatland area up to 18% (SSP1–2.6), 41% (SSP2–4.5) and 61% (SSP5–8.5) by 2300 (Müller and Joos, 2020). This is in contrast with the findings of northern peatland persistence and expansion under RCP2.6 and RCP6.0 scenarios in 1861–2099 by another modelling study (Qiu et al., 2020). In the Tropics, the only available study suggests peatland area will increase until 2300, mainly due to increases in precipitation and the CO2 fertilisation effect (Müller and Joos, 2020; Müller and Joos, 2021).

The combination of changes in climate and land use represents a substantial risk to peatland carbon stocks, but full assessment is impeded because peatlands are yet to be included in ESMs (limited evidence, high agreement ) (Loisel et al., 2021). It is expected that the carbon balance of peatlands globally will switch from sink to source in the near future (2020–2100), mainly because tropical peatland emissions, together with those from climate change-driven permafrost thaw, will likely surpass the carbon gain expected from climate change-driven enhanced plant productivity in northern high latitudes (Gallego-Sala et al., 2018; Chaudhary et al., 2020; Turetsky et al., 2020; Loisel et al., 2021) which are mainly caused by groundwater drawdown (robust evidence, medium agreement ) (Hirano et al., 2014; Brouns et al., 2015; Cobb et al., 2017; Itoh et al., 2017; Evans et al., 2021). The overall northern peatland carbon sink has been simulated to persist for at least 300 years under RCP2.6, but not under RCP8.5 (Qiu et al., 2020).

Increases in the extent, severity and duration of fires are expected in all peatland regions in the future due to temperature increases (Section 4.3.1.1), changes in precipitation patterns (Section 4.3.1.2) and increases in ignition sources (e.g., lightning) (Section 5.4.3.2), with associated rapid carbon losses to the atmosphere (medium evidence, high agreement ) (Dadap et al., 2019; Chen et al., 2021a; Nelson et al., 2021). For example, drought has been linked to fires in Southeast Asian peatlands (Field et al., 2009) and there are predicted decreases in mean summer precipitation (10–30%) for high and low RCPs, particularly over the Indonesian region, by the mid and late 21st century (Section 12.4.2.2) (Tangang et al., 2020; Taufik et al., 2020). During wet years, the fire probability in Indonesian peatlands also significantly increases (by 15–40%) when temperatures in July to October surpass 0.5°C anomalies compared to the 1995–2015 baseline (Fernandes et al., 2017). Overall, current evidence suggests that peat carbon losses via fire have the potential to be equal to, or greater than, losses due to human peatland drainage and disturbance (limited evidence, high agreement ) (Turetsky et al., 2015).

Regarding permafrost peatlands, studies differ, with some projecting a net loss and others a net gain of carbon (medium evidence, low agreement ) (Estop-Aragonés et al., 2018; Hugelius et al., 2020; Loisel et al., 2021; Väliranta et al., 2021). In some permafrost peatlands, prolonged and warmer growing seasons due to climate change (Section 2.3.4.3.1), along with increases in nitrogen deposition since 1850 (Lamarque et al., 2013), are promoting plant primary productivity. Other studies indicate that increased nitrogen-mediated sequestration could be exceeded by increased decomposition due to climate change-driven warming and fire (medium evidence, low agreement ) (Natali et al., 2012; Vonk et al., 2015; Keuper et al., 2017; Burd et al., 2018; Estop-Aragonés et al., 2018; Gallego-Sala et al., 2018; Serikova et al., 2018; Wild et al., 2019; Chaudhary et al., 2020; Hugelius et al., 2020).

Any climate change- or human-driven degradation of peatlands will also entail losses in water storage (limited evidence, high agreement ) (Wooster et al., 2012; Hirano et al., 2015; Cole et al., 2019; Taufik et al., 2019) and biodiversity (Harrison, 2013; Lampela et al., 2017; Renou-Wilson et al., 2019). The environmental archive contained in peat that preserves records of vegetation, hydrology, climate change, pollution and/or human disturbances is also being lost as peatlands degrade (Kasischke and Turetsky, 2006; MacDonald et al., 2006; Turunen, 2008; Field et al., 2009; Flannigan et al., 2009; Jones and Yu, 2010; Kasischke et al., 2010; Peterson et al., 2010; Finkelstein and Cowling, 2011; Rooney et al., 2012; Gallego-Sala and Colin Prentice, 2013; Lamarque et al., 2013; Hirano et al., 2014; Brouns et al., 2015; Turetsky et al., 2015; Miettinen et al., 2016; Schneider et al., 2016; Cobb et al., 2017; Fernandes et al., 2017; Itoh et al., 2017; Gallego-Sala et al., 2018; Greiser and Joosten, 2018; Ratcliffe et al., 2018; Dadap et al., 2019; Leifeld et al., 2019; Chaudhary et al., 2020; Hoyt et al., 2020; Müller and Joos, 2020; Qiu et al., 2020; Tangang et al., 2020; Taufik et al., 2020; Turetsky et al., 2020; Chen et al., 2021a; Evans et al., 2021; Loisel et al., 2021; Nelson et al., 2021; Qiu et al., 2021).

For boreal–tundra systems, AR5 projected the transformation of species composition, land cover and permafrost extent, decreasing albedo and increasing GHG emissions (medium confidence). SR1.5 classified tundra and boreal forests as particularly vulnerable to degradation and encroachment by woody shrubs (high confidence). The SROCC projected climate-related changes to arctic hydrology, wildfires and abrupt thaw (high confidence) and the broad disappearance of arctic near-surface permafrost this century, with important consequences for global climate (very high confidence). Chapter 2 of AR6 has focused on new key findings about observed and projected changes in tundra vegetation and related hydrology, with implications for feedbacks to the climate system.

Due to the rapid warming at high northern latitudes, the Arctic tundra is one of the terrestrial biomes where climate change impacts are already clearly visible (Settele et al., 2014; Uboni et al., 2016). Climate models project that warming of the Arctic is likely to continue at more than double the global rate. Compared to the period 1995–2014, mean annual surface air temperatures in the Arctic tundra are projected to increase by 7.9°C–10°C by the end of the century in scenarios of high GHG emissions (RCP7.0 and RCP8.5). In scenarios of low GHG emissions (RCP1.9 and RCP2.6), the projected increase is 2.6°C–3.2°C (Lee et al., 2021). The Arctic is also projected to have amongst the largest increases in precipitation globally, but with high uncertainty. In contrast to climate change, LUC is projected to be very low in Arctic tundra systems (van Asselen and Verburg, 2013).

Models of vegetation response to climate project acceleration in the coming decades of observed increases in shrub dominance and boreal forest encroachment that have been driven by recent warming (Settele et al., 2014), leading to a shrinking of the area of tundra globally (medium confidence) (Mod and Luoto, 2016; Gang et al., 2017). Simulating changes in tundra vegetation is complicated by permafrost dynamics (e.g., the formation of thaw ponds and draining of existing ponds), changes in precipitation and low nutrient availability (which may promote the abundance of graminoids) (van der Kolk et al., 2016). Changes in vegetation, when combined with warming and increased precipitation effects on soil thawing and carbon cycling, are projected to modify GHG emissions and have biophysical feedbacks to regional and global climate. High uncertainty in modelled carbon cycle changes arises from differences between the vegetation models (Nishina et al., 2015; Ito et al., 2016). In addition, climate change is expected to strongly interact with other factors, such as fire, to further increase uncertainty in projections of tundra ecosystem function (Jiang et al., 2017).

Projections point to potentially large changes of canopy structure and composition within and across the terrestrial biomes in response to climate change and changes in atmospheric CO2. These changes will contribute to altered ecosystem carbon uptake and losses, biophysical climate feedbacks (Sections 2.3.2; 2.4.4; 2.5.3.2; 2.5.3.3. 2.5.3.4, 2.5.3.5, Figure 2.10, Table 2.4) and multiple other ecosystem services (Sections 2.5.3, 2.5.4) as well impacts on biodiversity (Sections 2.4.2, 2.4.3, 2.4.4, 2.4.5, 2.5.1.3, 2.5.1.4, 2.5.2, Figure Box 2.1.1, Table Box 2.1.1, Table SM2.4). Until now, most studies project changes over next decades until the end of this century.

However, there is an increasing body of literature that has found continued, longer-term responses of ecosystems to climate change, so-called ‘committed changes’, that arise from lags that exist in many systems. Many processes in ecosystems take more than a few decades to quasi-equilibrate to environmental changes. Therefore, the trends of changing vegetation cover identified in simulations of transient warming continue to show up in simulations that hold climate change at low levels of warming (medium confidence) (Boulton et al., 2017; Pugh et al., 2018; Scheiter et al., 2020). Such changes, which could tip ecosystems into an alternative state, could also be triggered by a ‘warming overshoot’ if global warming were to exceed a certain threshold, even if mean temperatures afterwards decline again (Albrich et al., 2020a).

For instance, even if warming achieved by 2100 remained constant after 2100, such committed responses continue to occur. These include: (1) continued Amazon forest loss (Boulton et al., 2017), consistent with results in Pugh et al. (2018) that found continued tropical forest cover loss across a range of models and simulation setups, and (2) across Africa, an increased shift towards woody C3 vegetation was found in equilibrium state, the overall response depending on the atmospheric CO2 concentration (Scheiter et al., 2020). In Pugh et al. (2018), the opposite was found for boreal forest cover, which showed a strong committed increase. The committed changes in vegetation composition correspond to large committed changes in terrestrial carbon uptake and losses (Boulton et al., 2017; Pugh et al., 2018; Scheiter et al., 2020), and would plausibly also appear in other ecosystem functioning and services. These studies point to the importance of having not only a multi-decadal but also a multi-century perspective when exploring the impacts of political decisions on climate change mitigation taken now. Even if climate-warming targets are met, published evidence so far suggests that fundamental changes in some ecosystems are likely as these correspond to well-understood ecosystem physiological responses that trigger long-term changes in composition.

National parks and other protected areas which, in June 2021, covered 15.7% of the global terrestrial area (UNEP-WCMC et al., 2021), conserve greater biodiversity than adjacent unprotected areas (Gray et al., 2016), and protect one-fifth of global vegetation carbon stocks and one-tenth of global soil carbon stocks (Section 2.4.4.4). This section assesses climate change specifically in protected areas. Even though it is included in a part of the chapter on projected risks, it includes both observed exposure and projected risks to gather the information on protected areas into one place.

Under continued climate change, increased temperature, aridity, drought, wildfire (Section 2.5.3.2) and insect infestations (Section 2.4.4.3.3) will tend to increase tree mortality across many parts of the world (McDowell et al., 2020). Loss of boreal and temperate forest to fire, wind and bark beetles could cause more negative than positive effects for most ecosystem services, including carbon storage to regulate climate change (Sections 2.4.4.3, 2.5.2.6, 2.5.2.7, 2.5.3.4), water supply for people (Section 2.5.3.6.1), timber production and other forest products (Chapter 5) and protection from hazards (Thom and Seidl, 2016). In addition, deforestation in tropical and temperate forests can increase local temperatures by 0.3°C–2°C (Hesslerová et al., 2018; Lejeune et al., 2018; Zeppetello et al., 2020) and this effect can extend up to 50 km (Cohn et al., 2019).

In Amazon rainforests, the relatively lower buffering capacity for plant moisture during drought increases the risk of tree mortality and, combined with increased heat from climate change and fire from deforestation, the possibility of a tipping point of extensive forest dieback and a biome shift to grassland (Oyama and Nobre, 2003; Sampaio et al., 2007; Lenton et al., 2008; Nepstad et al., 2008; Malhi et al., 2009; Salazar and Nobre, 2010; Settele et al., 2014; Lyra et al., 2016; Zemp et al., 2017b; Brando et al., 2020). This could occur at a 4°C–5°C temperature increase above that of the pre-industrial period (Salazar and Nobre, 2010). Under RCP8.5, half the Amazon tropical evergreen forest could turn into grassland through drought-induced tree mortality and wildfire, but lower emissions (RCP4.5) could limit this loss to ~5% (Lyra et al., 2016). The decline in precipitation due to reduced evapotranspiration inputs after forest loss could cause additional Amazon forest loss of one-quarter to one-third (Zemp et al., 2017a). Similarly, in Guinean tropical deciduous forest in Africa, climate change under RCP8.5 could increase mortality 700% by 2100 or 400% under lower emissions (RCP4.5; (Claeys et al., 2019). These projections indicate risks of climate change-induced tree mortality reducing tropical forest areas in Africa and South America by up to half under a 4°C increase above the pre-industrial period, but a lower projection of a 2°C increase could limit the projected increases in tree mortality (robust evidence, high agreement ).

Temperate and boreal forests possess greater diversity of physiological traits related to plant hydraulics, so they are more buffered against drought than tropical forests (Anderegg et al., 2018). Nevertheless, in temperate forests, drought-induced tree mortality under RCP8.5 could cause the loss of half the Northern Hemisphere conifer forest area by 2100 (McDowell et al., 2016). In the western USA, under RCP8.5, one-tenth of forest area is highly vulnerable to drought-induced mortality by 2050 (Buotte et al., 2019). In California, increased evapotranspiration in Sierra Nevada conifer forests increases the potential fraction of the area at risk of tree mortality by 15–20% per degree Celsius (Goulden and Bales, 2019). In Alaska, fire-induced tree mortality from climate change under RCP8.5 could reduce the extent of spruce forest (Picea sp.) by 8–44% by 2100 (Pastick et al., 2017). Under RCP8.5, tree mortality from drought, wildfire and bark beetles could reduce the timber productivity of boreal forests in Canada by 2100 below the current levels (Boucher et al., 2018; Chaste et al., 2019; Brecka et al., 2020). In Tasmania, projected increases in wildfire (Fox-Hughes et al., 2014) increase the risk of mortality of mesic vegetation (Harris et al., 2018b) and threaten the disappearance of the long-lived endemic pencil pine (Athrotaxis cupressoides) (Holz et al., 2015; Worth et al., 2016) and temperate montane rainforest (Mariani et al., 2019). These projections indicate risks of climate change-induced tree mortality reducing some temperate forest areas by half under emissions scenarios of 2.5°C–4°C above the pre-industrial period (medium evidence, high agreement ).

Globally, increasing atmospheric CO2 enhances the terrestrial sink but temperature increases constrain it, reflecting the biological process understanding highlighted in previous IPCC reports (high confidence). Analyses of atmospheric inversion model output and spatial climate data indicate a sensitivity of net ecosystem productivity to CO2 fertilisation of 3.1 ± 0.1 Gt to 8.1 ± 0.3 Gt per 100 ppm CO2 (~1°C increase) and a sensitivity to temperature of -0.5 ± 0.2 Gt to -1.1 ± 0.1 Gt per degree Celsius (Fernandez-Martinez et al., 2019). The future of the global land carbon sink (Section 2.4.4.4) nevertheless remains highly uncertain because (i) of regionally complex interactions of climate change and changes in atmospheric CO2 with vegetation, soil and aquatic processes, (ii) episodic events such as heat waves or droughts (and related impacts through mortality, wildfire or insects, pests and diseases) (Section 2.5.3.2, 2.5.3.3) are so far only incompletely captured in carbon cycle models, (iii) the legacy effects from historic LUC and environmental changes are incompletely captured but likely to decline in future and (iv) lateral carbon transport processes such as the export of inland waters and erosion are incompletely understood and modelled (Pugh et al., 2019a; Friedlingstein et al., 2020; Krause et al., 2020; Canadell et al., 2021).

Enhanced carbon losses from terrestrial systems further limit the available carbon budget for global warming staying below 1.5°C (Rogelj et al., 2018). Analyses of satellite remote sensing and ground-based observations have indicated that, between 1982 and 2015, the CO2 fertilisation effect has already declined, implying a negative climate system feedback (Wang et al., 2020c). Peatlands, permafrost regions and tropical ecosystems are particularly vulnerable due to their large carbon stocks, in combination with over-proportional warming, increases in heat waves and droughts and/or a complex interplay of climate change and increasing atmospheric CO2 (Sections 2.5.2.8, 2.5.2.9, 2.5.3.2).

Model projections suggest a reduction of permafrost extent and potentially large carbon losses for all warming scenarios (Canadell et al., 2021). Already a mean temperature increase of 2°C could reduce the total permafrost area extent by about 5–20% by 2100 (Comyn-Platt et al., 2018; Yokohata et al., 2020). Associated CO2 losses in the order of 15 Gt up to nearly 70 Gt by 2100 have been projected across a number of modelling studies (Schneider von Deimling et al., 2015; Comyn-Platt et al., 2018; Yokohata et al., 2020). Limiting the global temperature increase to 1.5°C versus 2°C could reduce projected permafrost CO2 losses by 2100 by 24.2 Gt (median, calculated for a 3-m depth) (Comyn-Platt et al., 2018). Losses are possibly underestimated in the studies that consider only the upper permafrost layers. Likewise, the actual committed carbon loss may well be larger (e.g., eventually a loss of approx. 40% of today’s permafrost area extent if climate is stabilised at 2°C above pre-industrial levels) due to the long time scale of warming in deep permafrost layers (Chadburn et al., 2017). It is not known at which level of global warming an abrupt permafrost collapse (estimated to enhance CO2 emissions by 40% in 2300 in a high-emissions scenario) compared to gradual thaw (Turetsky et al., 2020) would have to be considered an important additional risk. Large uncertainties arise also from interactions with changes in surface hydrology and/or northward migrating woody vegetation as climate warms, which could dampen or even reverse projected net carbon losses in some regions (McGuire et al., 2018a; Mekonnen et al., 2018; Pugh et al., 2018). Overall, there is low confidence on how carbon–permafrost interactions will affect future carbon cycle and climate, although net carbon losses and thus positive (amplifying) feedbacks are likely (Sections 2.5.2.10, 2.5.3.5) (Shukla et al., 2019). See also WGI AR6 (Canadell et al., 2021) for a discussion on impacts of higher-emission and warming scenarios.

Peatland carbon is estimated as about 550–1000 Gt in northern latitudes (many of these peatlands would be found in permafrost regions) (Turetsky et al., 2015; Nichols and Peteet, 2019) and >100 Gt in tropical regions (Turetsky et al., 2015; Dargie et al., 2017). For both northern mid- and high-latitude and tropical peatlands, a shift from contemporary CO2 sinks to sources were simulated in high-warming scenarios (Wang et al., 2018a; Qiu et al., 2020). Due to the lack of large-scale modelling studies, there is low confidence for climate change impacts on peat carbon uptake and emissions. The largest risk to tropical peatlands is expected to arise from drainage and conversion to forestry or agriculture, which would outpace the impacts of climate change (Page and Baird, 2016; Leifeld et al., 2019; Cooper et al., 2020). The magnitude of possible carbon losses is uncertain, however, and depends strongly on socioeconomic scenarios (Sections 2.4.3.8, 2.4.4.2; 2.4.4.4.2, 2.5.2.8).

For tropical and subtropical regions, the interplay of atmospheric CO2 with precipitation and temperature becomes of particular importance for future carbon uptake, since in warm and dry environments, elevated CO2 fosters plants with C3 photosynthesis and enhances their water-use efficiency relative to C4 species (Moncrieff et al., 2014a; Midgley and Bond, 2015; Knorr et al., 2016a). As a consequence, enhanced woody cover is expected to occur in the future, especially in mesic savannas, while in xeric savannas an increase in woody cover would occur in regions with enhanced precipitation (Criado et al., 2020). Even though semiarid regions have dominated the global trend in land CO2 uptake in recent decades (Ahlström et al., 2015), so far, most studies that investigated future climate change impacts on savanna ecosystems have concentrated on changes in the extent of land area affected (2.5.2.5) rather than on carbon cycling, with medium confidence for increasing woody cover:grass ratios (Moncrieff et al., 2014a; Midgley and Bond, 2015; Moncrieff et al., 2016; Criado et al., 2020). Increases in woody vegetation in what is now grass-dominated would possibly come with a carbon benefit, for instance, it was found that a broad range of future climate and CO2 changes would enhance vegetation carbon storage on Australian savannas (Scheiter et al., 2015). Results from a number of field experiments indicate, however, that impacts on total ecosystem carbon storage may be smaller due to a loss in below-ground carbon (Coetsee et al., 2013; Wigley et al., 2020). Nunez et al. (2021) critique existing incentives to promote the invasion of non-native trees into treeless areas as a means of carbon sequestration, raising doubts about the effects on fire, albedo, biodiversity and water yield (see Box 2.2).

Substantial climate change-driven impacts on tropical tree cover and vegetation type are projected in all studies, irrespective of whether or not the degree amounts to a forest “dieback” (Sections 2.4.3.6, 2.4.4.3, 2.5.2.6, 2.5.3.3) (Davies-Barnard et al., 2015; Wu et al., 2016a; Zemp et al., 2017a; Canadell et al., 2021) . Accordingly, models also suggest a continuation of tropical forests acting as carbon sinks (Huntingford et al., 2013; Mercado et al., 2018). A recent study combining field plot data with statistical models (Hubau et al., 2020) indicates that, in the Amazonian and possibly also in the African forest, the carbon sink in above-ground biomass already declined in the three decades up to 2015. This trend is distinct in the Amazon whereas data from Africa suggests a possible decline after 2010. The authors estimate the vegetation carbon sink in 2030–2040 to decline to zero±0.205 PgC yr -1 in the Amazon and to 0.26±0.215 PgC yr -1 in Africa (a loss of 14% compared to the present). Their results suggest that, over time, CO2 fertilisation is outweighed by the impacts of higher temperatures and drought that enhance tree mortality and diminish growth. The degree of thermal resilience of tropical forests is still uncertain, however (Sullivan et al., 2020).

The lack of simulation studies that seek to quantify all important interacting factors (CO2, drought and fire) for future carbon cycling in savannas and tropical forests and the apparent disagreement between trends projected in models compared to data-driven estimates result in low confidence regarding the direction or magnitude of carbon flux and pool-size changes. Similar to tropical peatlands, given projected human population growth and socioeconomic changes, the continued conversion of forests and savannas into agricultural or pasture systems very likely poses a significant risk of rapid carbon loss which will amplify the climate change-induced risks substantially (high confidence) (2.5.2.10, 2.5.3.5) (Aragao et al., 2014; Searchinger et al., 2015; Aleman et al., 2016; Nobre et al., 2016).

The impacts of climate-induced altered animal composition and trophic cascades on land-ecosystem carbon cycling globally are as yet unquantified (Schmitz et al., 2018), even though climate change is expected to lead to shifts in consumer–resource interactions that also contribute to losses of top predators or top herbivores (Sections 2.4.2.2, 2.5.1.3, 2.5.4; (Lurgi et al., 2012; Damien and Tougeron, 2019). Cascading trophic effects triggered by top predators or the largest herbivores propagate through food webs and reverberate through to the functioning of whole ecosystems, notably altering productivity, carbon and nutrient turnover and net carbon storage (medium confidence) (Wilmers and Schmitz, 2016; Sobral et al., 2017; Stoner et al., 2018). Across different field experiments, the ecosystem consequences of the presence or absence of herbivores and carnivores have been found to be quantitatively as large as the effects of other environmental change drivers such as warming, enhanced CO2, fire and variable nitrogen deposition (medium confidence) (Hooper et al., 2012; Smith et al., 2015). Some local and regional modelling experiments have begun to explore animal impacts on vegetation dynamics and carbon and nutrient cycling (Pachzelt et al., 2015; Dangal et al., 2017; Berzaghi et al., 2019). Turnover rate is the chief factor that determines future land-ecosystem carbon dynamics and hence carbon–climate feedbacks (Friend et al., 2014). To improve projections, it is imperative to better quantify the broader role of carnivores, grazers and browsers and the way these interact in global studies of how ecosystems respond to climate change.

The possibility of feedbacks and interactions between climate drivers and biological systems or ecological processes was identified as a significant emerging issue in AR5, and has since also been highlighted in the SRCCL and the SR1.5. It is virtually certain that land cover changes affect regional and global climate through changes to albedo, evapotranspiration and roughness (very high confidence) (Perugini et al., 2017). There is growing evidence that biosphere-related climate processes are being affected by climate change in combination with disturbance and LULCC (high confidence) (Jia et al., 2019). It is virtually certain that land surface change caused by disturbances such as forest fires, hurricanes, phenological changes, insect outbreaks and deforestation affect carbon, water and energy exchanges, thereby influencing weather and climate (very high confidence) (Table 2.4; Figure 2.10) (Bright et al., 2013; Brovkin et al., 2013; Naudts et al., 2016; Prăvălie, 2018).

Feedbacks can be positive or negative (i.e., amplify or dampen the original forcing), vary spatially and seasonally, and act over large geographic areas and long time periods (more than decades), making them difficult to observe and quantify directly (Schimel et al., 2015; Canadell et al., 2021). Due to the positive impacts of CO2 on vegetation growth and ecosystem carbon storage (high confidenc e) (Sections 2.4.4.4, 2.5.5.4) (Canadell et al., 2021), the associated climate feedback is negative (i.e., increased removal of atmospheric CO2 and dampened warming, compared to an absence of the feedback). By contrast, projected global losses of carbon in warmer climates (Canadell et al., 2021) imply a positive climate feedback. WGI (Canadell et al., 2021) assesses an overall increase in land carbon uptake through the 21st century. However, the overall strength of the carbon cycle–climate feedback remains very uncertain. One of the underlying reasons may be complex interactions with ecosystem water balance and nitrogen and phosphorous availability, which are poorly constrained by observational evidence and incompletely captured in ESMs (Section 2.5.2.10) (Huntzinger et al., 2017; Canadell et al., 2021).

Land ecosystems contribute substantially to global emissions of nitrous oxide and methane . As with CO2, these emissions respond both directly and indirectly to atmospheric CO2 concentration and climate change, and this gives rise to potential additional bio-geochemical feedbacks in the climate system. A large part of these emissions stem from land and water management, such as fertilizer application, rice production, aquaculture or animal husbandry (Jia et al., 2019). However, nearly 60% of total nitrous oxide emissions (in 2007–2016) has been estimated to stem from natural ecosystems, especially in the Tropics (Tian et al., 2019; Canadell et al., 2021), while freshwater wetlands and peatlands are estimated to contribute between 83% (top-down estimates) and 40% (bottom-up estimates) of total natural CH4 (and 31 and 20% of total methane emissions, respectively) for the period 2008–2017 (Canadell et al., 2021). Median CH4 emissions from northern-latitude wetlands in 2100 were estimated to be 12.1 and 13.5 PgC in emission scenarios leading to 1.5°C and 2°C warming, respectively (Comyn-Platt et al., 2018). Likewise, global warming has been attributed to soil N2O emission increases since the pre-industrial period of 0.8 (0.3–1.3) TgN yr -1 (Tian et al., 2020). Overall, climate feedbacks from future altered land ecosystem emissions of CH4 or N2O are uncertain, but are expected to be small (Canadell et al., 2021).

Changes in regional biodiversity are integral parts of ecosystem–climate feedback loops, including and beyond carbon cycle processes (Figure 2.10; Table 2.4). For instance, the impacts of climate-induced altered animal composition and trophic cascades on ecosystem carbon turnover (see Sections 2.4.4.4, 2.5.3.4) could be a substantive contribution to carbon–climate feedbacks (low confidence). Additional surface–atmosphere feedbacks that arise from changes in vegetation cover and subsequently altered albedo, evapotranspiration or roughness (often summarised as biophysical feedbacks) can be regionally relevant and could amplify or dampen vegetation cover changes (Jia et al., 2019).

Climate-induced shifts towards forests in what is currently tundra would be expected to reduce regional albedo especially in spring, but also during parts of winter when trees are snow-free (whereas tundra vegetation would be covered in snow), which amplifies warming regionally (high confidence) (Perugini et al., 2017; Jia et al., 2019). Trees would also enhance momentum absorption compared to low tundra vegetation, thus impacting surface–atmosphere mixing of latent and sensible heat fluxes (Jia et al., 2019). Boreal forests insulate and stabilize permafrost and reduce fluctuations of ground temperature: the amplitude of variation of ground surface temperatures was 28°C at a forested site, compared to 60°C in nearby grassland (Section 2.5.2.7) (Bonan, 1989; Stuenzi et al., 2021a; Stuenzi et al., 2021b). Likewise, a shift in moist tropical forests towards vegetation with drought-tolerant traits could possibly reduce evapotranspiration, increase albedo, alter heat transfer at the surface and lead to a negative feedback to precipitation (Section 2.5.2.6) (Jia et al., 2019). In savannas, restoration of woody vegetation has been shown to enhance cloud formation and precipitation in response to enhanced transpiration and turbulent mixing, leading to a positive feedback on woody cover (Syktus and McAlpine, 2016). While this has not yet been systematically explored, similar feedbacks might also emerge from a CO2-induced woody cover increase in savannas (low confidence) (Section 2.5.2.5).

Since biophysical feedbacks can contribute to both surface temperature warming or cooling, analyses so far suggest that, on a global scale, the net impact on climate change is small (Perugini et al., 2017; Jia et al., 2019), unless these feedbacks also accelerate vegetation mortality and lead to substantive carbon losses (Zemp et al., 2017a; Lemordant and Gentine, 2019). More than one-third of the Earth’s land surface has at least 50% of its evapotranspiration regulated by vegetation, and in some regions between 40 and >80% of the land’s evaporated water is returned to land as precipitation. Locally, both directly human-mediated and climate change-mediated changes in vegetation cover can therefore notably affect annual average freshwater availability to human societies, especially if negative feedbacks amplify the reduction of vegetation cover, evapotranspiration and precipitation (medium confidence) (Keys et al., 2016; Keys and Wang-Erlandsson, 2018).

Since AR5, freshwater ecosystems (lakes, reservoirs, rivers and ponds) have been increasingly recognised as important sources of GHG emissions (CO2, CH4 and N2O) into the atmosphere. Key mechanisms which contribute to rising GHG emissions from freshwater ecosystems are the temperature imbalance between photosynthesis and respiration (respiration increases more than photosynthesis with rising temperature), CO2 and CH4 emissions from exposed sediments during droughts, increased transport of matter from land to water, changes in water retention time in rivers and lakes and the effects of temperature on lake stratification and anoxia that favour CH4 emissions.

DelSontro et al. (2018) assembled the largest global data set to date on emission rates from lakes of CO2, CH4 and N2O and found that they co-vary with lake size and trophic state. They estimated that moderate global increases in eutrophication of lakes could translate to 5–40% increases in the GHG effect in the atmosphere. Moreover, they estimated that GHG emissions from lakes and impoundments in past decades accounted for 1.25–2.30 PgCO2 yr -1 (DelSontro et al., 2018), thus around 20% of global CO2 emissions from the burning of fossil fuels (9.4 PgCO2 yr -1) (Friedlingstein et al., 2020).

Global warming will strongly enhance freshwater CH4 emissions through a disproportionate increase in ebullition (gas flux) by 6–20% per 1°C increase in water temperature (Aben et al., 2017). It can be expected that ongoing eutrophication enhanced by climate change-related increases in the release of sediment nutrients and the loading of organic carbon and nutrients from catchments will enhance CH4 ebullition on a global scale (Aben et al., 2017; DelSontro et al., 2018; Bartosiewicz et al., 2019; Beaulieu et al., 2019; Sanches et al., 2019). The strongest increase in ebullition is expected in shallow waters where sediment temperatures are strongly related to atmospheric temperature (Aben et al., 2017). Given that small ponds and shallow lakes are the most abundant freshwater ecosystems globally, these may become hot spots of CH4 ebullition in the future (Aben et al., 2017). On average, CH4, CO2 and N2O account for 75, 23 and 2% of the total CO2-equivalent emissions, respectively, in lakes (DelSontro et al., 2018).

Furthermore, the exposure of lake and river sediments during droughts activates the decomposition of buried organic carbon. In dry river beds, mineralisation of buried organic matter is likely to increase with climate change as anoxic sediments are oxygenated downwards during drying, along with pulses of microbial activity following re-wetting of desiccated sediment. Conservative estimates indicate that adding emissions from exposed sediments of dry inland waters across diverse ecosystem types and climate zones to current global estimates of CO2 emissions could result in a 6% (~0.12 PgC yr −1) increase of total inland water CO2 emission rates covering streams and rivers (334 mmol m -2 day -1), lakes and reservoirs (320 mmol m -2 day -1) and small ponds (148 mmol m -2 day -1) (Marcé et al., 2019; Keller et al., 2020).

Overall, uncertainty as to the quantity of carbon fluxes within freshwater ecosystems and between terrestrial and freshwater systems, and subsequent emissions to the atmosphere remains very high (Raymond et al., 2013; Catalán et al., 2016; Stanley et al., 2016; Evans et al., 2017; Drake et al., 2018; Seekell et al., 2018; Sanches et al., 2019; Bodmer et al., 2020; Keller et al., 2020; Canadell et al., 2021) (see Table SM2.1.). Projections of carbon fluxes are, for example, challenged by the complex interaction between rising water temperatures, loss of ice, changes in hydrology, ecosystem productivity, increased extreme events and variation in terrestrial-matter transport. While we are still short of empirical data, particularly in the Tropics (DelSontro et al., 2018), improvements in sensor technology (Eugster et al., 2011; Gonzalez-Valencia et al., 2014; Maeck et al., 2014; Delwiche et al., 2015) and the use of statistically robust survey designs (Beaulieu et al., 2016; Wik et al., 2016) have improved the accuracy of measurements of GHG emissions in freshwater ecosystems. Global networks such as the Global Lakes Ecological Observatory Network (GLEON) increasingly allow a global view of carbon fluxes, thereby improving estimates of the contribution of freshwater ecosystems to global GHG emissions to the atmosphere.

In summary for freshwater systems, Drake et al. (2018) aggregated contemporary estimates of CO2 and CH4 emissions from freshwater ecosystems with global estimates made by Raymond et al. (2013), and arrived at an estimate of 3.9 PgC yr -1. Rivers and streams accounted for 85% and lakes and reservoirs for 15% of the emissions (Raymond et al., 2013). This trend will continue under scenarios of nutrient loading to inland waters over the next century where increased CH4 emission of inland water has an atmospheric impact of 1.7–2.6 PgC/CO2-eq y −1, which is equivalent to 18–33% of annual CO2 emissions from burning fossil fuels (medium evidence, medium agreement ) (Beaulieu et al., 2019). For comparison, annual uptake of CO2 in land ecosystems is estimated as 3.4 (± 0.9) PgC yr −1 (Friedlingstein et al., 2020). The freshwater numbers combine CO2 and CH4 and are thus not directly comparable. However, they are indicative of the importance of better accounting for freshwater systems in global carbon budgets.

Table 2.4 | Terrestrial and freshwater ecosystem feedbacks which affect the Earth’s climate system dynamics, according to (Prăvălie, 2018).

AR5 named water supply and biodiversity as freshwater ecosystem services vulnerable to climate change. We discuss the risks to these and to additional services identified by model projections based both on climate-change scenarios (Schröter et al., 2005; Boithias et al., 2014; Huang et al., 2019; Jorda-Capdevila et al., 2019) and on the Common International Classification of Ecosystem Services (high confidence) (CICES, 2018). The effects of floods, droughts, permafrost and glacier-melting on global changes in water quality, particularly with respect to contamination with pollutants, are described in Section 4.2.6.