1.3.2.1 Tropical forests

Tropical forests cover around 1.95bn hectares (including degraded portions), and are key components of the Earth system (Pan et al., 2011) (Figure 1.3.2). They are home to a disproportionate amount of Earth’s species (e.g. Slik et al., 2015; Pillay et al., 2021), store huge amounts of carbon (circa 471 ±93 GtC) in their soils and biomass, and, through evapotranspiration and their effect on cloud formation through production of aerosols and cloud condensation nuclei, have an overall cooling and moistening effect at regional scales (SPA, 2021; IPCC AR6 WG2 2021). They are also home to many Indigenous peoples and local communities, with a long history of human habitation and high biocultural diversity (Ellis et al.,2021).

As well as experiencing deforestation and degradation due to land use change across the tropics (IPBES 2019), tropical forests in South America and Asia have been undergoing unprecedented climate-driven disturbances such as increasing dry season length and intensity, more intense and frequent rainfall and temperature extremes (Lapola et al., 2023; SPA, 2021). For instance, recent extreme droughts – mainly driven by climate variability modes such as the El Niño Southern Oscillation (ENSO) in 2014-2016 and the Atlantic dipole in 2005 and 2010 (e.g. Marengo et al., 2008; Marengo et al., 2011; Jimenez-Muñoz et al., 2016; see Chapter 1.4) – have caused extensive tree mortality, even up to 36 months after peak drought (e.g. Phillips et al., 2009; Phillips et al., 2010; Berenguer et al., 2021). Given the variability of forests across the tropics, their responses to global changes are likely to differ (Allen et al., 2017). Nonetheless, even subtle changes in their structure, composition and functioning could affect the global carbon and water cycles (e.g. Esquivel-Muelbert et al., 2019; Barros et al., 2019; Hirota et al., 2021). 

Here we also consider deciduous and semi-deciduous forests (often referred to as dry forests) that coexist with evergreen forests in regions with around 1,000-2,000mm of annual rainfall, i.e. non-arid or dryland regions (Dexter et al., 2018). These dry forests may resemble (in terms of tree species composition) the dry forests in arid or dryland regions. However, because they exist in climates that can form continuous, high fuel-load flammable grass layers when canopies are opened (which is not the case in drylands), their dynamics are more comparable to neighbouring moist forests.

Evidence for tipping dynamics

Two positive/amplifying feedbacks are among the most plausible mechanisms that could lead to tipping dynamics in tropical forests, one at broader regional scales potentially causing large-scale forest collapse, and another at local scales potentially causing local forest collapse (Figure 1.3.3 and Box 1.3.2).

At regional scales, the forest-rainfall feedback is believed to be the dominant mechanism stabilising tropical forests by increasing annual rainfall levels and reducing its seasonal and interannual variability (Staal et al., 2020; Sternberg, 2001). However, under certain conditions it can instead amplify forest loss.

Accumulated deforestation or forest loss reduces forest cover, which decreases evapotranspiration and moisture flow downwind, thus reducing regional rainfall (Smith et al., 2023). This in turn may increase tree mortality in downwind forest (Phillips et al., 2009; Berenguer et al., 2021), and beyond a threshold could lead to self-sustaining forest loss in drier areas of forest (Zemp et al., 2017; Staal et al., 2020) (Figure 1.3.3).

In the Amazon, on average, around 30 per cent of the water precipitating has been evaporated within the region beforehand at least once, but with large spatial differences: in the western Amazon, almost all precipitation has previously evaporated from the basin (Zemp et al., 2014; Staal et al., 2018). In the Congo basin, almost half of all precipitation originates from the Congo forest itself (Tuinenburg et al., 2020; Te Wierik et al., 2022). For Australian and Asian forests, evidence is still lacking, but this feedback likely has less effect on forest resilience due to less dependence on precipitation stemming from land evapotranspiration due to the major importance of monsoons (Staal et al., 2020).

At the local scale, evidence from across the tropics (Cochrane et al., 1999; Staver et al., 2011; van Nes et al., 2018) suggests that a fire-vegetation feedback can maintain the ecosystem in an open vegetation state: with less tree cover, fires spread more easily due to more flammable grassy fuels and because the air is drier in an open landscape without the local moistening effect of forest canopies. The resulting enhanced fire occurrence can in turn prevent the re-establishment of trees and maintain a more open vegetation state (Martinez-Cano et al., 2022; Drüke et al., 2023). This alternative open vegetation state could be either a natural savanna with native plant species (Flores and Holmgren, 2021; Beckett et al., 2022) or a degraded open-vegetation state when invasive plants are dominant (D’Antonio and Vitousek, 1992; Veldman and Putz, 2011; Malhi et al., 2014; Barlow et al., 2018) (Figure 1.3.3).

The effects of the fire-vegetation feedback are amplified by the regional forest-rainfall feedback (Staal et al., 2020). Moreover, forest loss may increase global warming by releasing carbon to the atmosphere, which further reduces regional moisture flows, causing more forest loss (Canadell et al., 2021). Also, climate change may change wind directions and residence times of moisture in a warmer atmosphere (Gimeno et al., 2021). Tropical forest loss also may change atmospheric circulation patterns (Portmann et al., 2022) and increase regional and global warming through reductions in cloud cover and evapotranspiration.

Among tropical forests, the Amazon forest has most evidence for potential tipping points. Analysis based on early warning signals (see Chapter 1.6) indicates that over 75 per cent of the Amazon has lost resilience since the early 2000s (Boulton et al., 2022). This decline is focused mostly closer to human disturbance, as well as in the drier south and east previously identified as ‘bistable’ (i.e. with two possible alternative states) due to the forest-rainfall feedback and thus is more vulnerable to tipping (Staal et al., 2020). While the Amazon has acted as a carbon sink due to CO₂ fertilisation, in mature forest this sink peaked and started declining in the 1990s (Hubau et al., 2020) and when including degraded forest (also predominantly in the drier south and east) the Amazon as a whole is now a carbon source (Gatti et al., 2021). Recent CMIP6 models indicate that localised shifts in peripheral parts of the Amazon forest system are more likely than a large-scale tipping event (IPCC AR6 WG1 Ch5, 2021; Parry et al., 2022). However, the latter cannot be ruled out (Hirota et al., 2021) because several compounding and possibly synergistic disturbances (e.g. combining an extreme hot drought with forest fires) may play a role in reducing forest resilience, with greater resilience loss closer to human activities (Boulton et al., 2022). Such synergies are generally not considered in Earth system models (Willcock et al., 2023). 

A global warming threshold of ~3.5°C (2-6°C) has been estimated (Armstrong McKay et al., 2022), partly based on a few modelling studies that simulate some kind of nonlinear decrease in modelled properties of the Amazon forest, at least on small scales (Gerten et al., 2013; Drijfhout et al., 2015; Nobre et al., 2016; Boulton et al., 2017; Parry et al., 2022). However, most CMIP6 models do not include dynamic vegetation modules (Song et al., 2021; Canadell et al., 2021), which might make the forest artificially stable (Zemp et al., 2017). Models including deforestation, fire and dynamic vegetation have simulated widespread local-scale dieback (e.g. Cano et al., 2022; Parry et al., 2022), and also larger scale dieback in potential vegetation models (e.g. Salazar and Nobre, 2010). 

Evidence pointing against a large-scale Amazon tipping point stems from palaeoclimate reconstructions suggesting that at least some parts of the Amazon forest have been resilient to past reductions in rainfall (Wang et al., 2017; Kukla et al., 2021) and temperatures as high as projected by climate models for the rest of the century (Steinthorsdottir et al., 2020). However, these were under more stable climate conditions (and before Pleistocene with different geographic effects on climate due to tectonics; (Brierley and Fedorov, 2016), with the current rate of warming far greater than during past climate changes (Zeebe et al., 2016; Osman et al., 2021). Geographically limited data means partial dieback elsewhere cannot be ruled out for drier intervals (Wang et al., 2017; Kukla et al., 2021), particularly in the drier south, where drying is currently leading to greater resilience loss (Boulton et al., 2022). Additionally, compounding disturbances are becoming increasingly widespread across the Amazon, even in remote central parts of the system, which is leading to resilience loss (Boulton et al., 2022) and could help trigger forest dieback at larger scales (Kukla et al., 2021; Wilcock et al., 2023).

Other tropical forests have evidence for local tipping points, but are less likely to cross them. The Congo has also been suggested as a possible tipping system (Staal et al., 2020) as it may also host a large area of bistable forest with some amplification by forest-rainfall feedback (Staver et al., 2011). However, because climate models indicate wetting across large parts of the Congo, it is not considered a tipping system in response to global warming (Armstrong McKay et al., 2022). The south-east Asian rainforests lack a strong regional forest-rainfall feedback and tend to have enough rainfall from ocean proximity for forests to remain stable, thus they are not considered a tipping system in relation to global warming (Armstrong McKay et al., 2022). Other tropical forests such as the Choco in Central America or Brazilian Atlantic Forests have not been assessed in detail.

Plants can reduce moisture transpiration in response to water limitation on very short timescales (hours to days), followed by water cycle feedbacks (weeks). Deforestation has a similarly fast effect on rainfall, as loss of trees can immediately reduce evapotranspiration. Large-scale forest dieback events in response to global warming can only be expected on the timescale of decades to centuries (Armstrong McKay et al., 2022). At a local scale, empirical evidence from the Amazon and from Africa has shown that forests can shift into savannas within a few decades after repeated fires (Flores and Holmgren, 2021; Beckett et al., 2022), and on larger scales tipping may occur faster (Cooper et al., 2020).

An Amazon tipping point would have global impacts from possibly large losses of carbon to the atmosphere. The best estimates suggest that a large-scale collapse of 40 per cent of the forest before the end of this century could lead to emissions of ~30 GtC and an additional global warming of ~0.1°C (Armstrong McKay et al., 2022). The Amazon dieback would also lead to substantial rainfall reductions across the Amazon basin and in to the Southern Cone of South America (Costa et al., 2021), and may also directly influence distant parts of the Earth system via ‘teleconnections’, for example to the Tibetan Plateau (Liu et al., 2023).

Assessment and knowledge gaps

The feedbacks that could contribute to tipping behaviour are relatively well understood in principle, yet there are large uncertainties surrounding the effects of climate and land use changes on these feedbacks. For instance, CO₂-fertilisation is expected to increase forest resilience locally, but it also increases water-use efficiency, reducing forest transpiration, and may thus weaken the forest-rainfall feedback and regional forest resilience (Brienen et al., 2020; Sampaio et al., 2021; Kooperman et al., 2018; Li et al., 2023). CO₂-fertilisation of tropical forests may also be overestimated in current Earth system models (Terrer et al., 2019; Hubau et al., 2020; Wang et al., 2020). Moreover, the actual thresholds and the extent to which tipping behaviour can be expected across heterogeneous landscapes and forest communities are much less certain (Levine et al., 2016; Longo et al., 2018; Sakschewski et al., 2021). 

Considering only the Amazon as a rainforest tipping system, we have medium confidence in its potential for tipping of its bistable area (~40 per cent of the forest, predominantly in the drier south and east; Staal et al., 2020), with low confidence in the estimated tipping points and possibility of a large-scale collapse. The Congo may also be vulnerable to localised tipping (low confidence), but is unlikely to tip as a result of climate change, and localised tipping is possible but uncertain in south-east Asian rainforests.

Confidence in the tipping behaviour of tropical forests can be greatly improved through further development of models. Models can include dynamic vegetation modules and land use change to improve the representation of the forest-rainfall feedback, which would likely result in more drastic drying under high-deforestation scenarios (Parry et al., 2022). Incorporating fire dynamics in these modules would also likely result in a more bistable system (Drüke et al., 2023). In contrast, allowing for local vegetation adaptation (such as rooting depth) by including more plant types and traits in these modules would help better resolve the effect of landscape heterogeneity on tipping dynamics (Langan et al., 2017; Sakschewski et al., 2021), which may reduce the abruptness of the transition to an open degraded state (Levine et al., 2016). Efforts to increase ecological understanding of the feedback mechanisms and processes described here through observations (such as recent field studies on plant characteristics related to drought mortality throughout the Amazon basin (Tavares et al., 2023), or on the growth-survival tradeoff (Oliveira et al., 2021)) would help better understand forest dynamics and represent them in models.