Savannas & grasslands

Savannas and grasslands are characterised by the ecological dominance of grasses, sometimes with a substantial tree or shrub component (Figure 1.3.7). Savanna ecosystems are biodiverse and home to many people, but are being lost to a range of threats globally, especially because they are extensively targeted for agricultural conversion (Stevens et al., 2022, Strömberg and Staver, 2022). Even intact savannas are under threat, largely due to forest invasion or afforestation and woody encroachment, driven by grazing intensification and active fire suppression, and exacerbated by increasing atmospheric CO2 (Stevens et al., 2017) and changing rainfall regimes (Kulmatiski and Beard, 2013). Active tree planting efforts further increase the threat to savannas from afforestation and woody encroachment. 

Figure 1.3.7: Global distribution of savannas and grasslands, showing semi-arid vs. mesic distributions (centre, from: (Strömberg and Staver 2022), replotted from (Dinerstein et al. 2017). Pictured, clockwise from top left, are native grassy ecosystems in 1) Montana near Dillon, USA; 2) Alps near Mont Blanc, France; 3) Pool Department, Republic of Congo; 4) Serengeti NP, Tanzania; 5) Pench NP, India; 6) Chhaeb Wildlife Sanctuary, Cambodia; 7) Kidman Springs Ranch, Australia; 8) Gorongosa NP, Mozambique; 9) Kruger NP, South Africa; 10) Santa Cruz Province near Lago Argentino, Argentina; 11) Instituto Brasileiro de Geografia e Estatística Reserve, Brasilia, Brasil; 12) Apalachicola National Forest, Florida, USA. Photo credits: Carla Staver, Caroline Strömberg, Naomi Schwartz.

Although the converse issue receives extensive attention (e.g. Amazon rainforest collapse), the issue of savanna vulnerability to tipping points is recognised (Staver et al., 2011b) but generally neglected in literature and assessments of tipping points in the Earth system (Armstrong McKay, 2022; Wang et al., 2023). Savanna vulnerability to desertification (corresponding to a self-sustaining loss of ecosystem productivity) is sometimes cited in tipping point syntheses, but the generality of this feedback has been questioned. For example, aridification observed in western Africa’s Sahel during the 1970s and 80s has since reversed across much of the Sahel in response to a cyclic increase in rainfall (Nicholson et al., 1998; Prince et al., 2007).

Evidence for tipping dynamics

Savanna and forest are widely considered to be alternative stable ecosystem states in some climates (Staver et al., 2011a, 2011b; Hirota et al., 2011; Dantas et al., 2015; Aleman et al., 2020). In savannas and grasslands, an open tree canopy permits high grass productivity and thus the accumulation of grass fuel for frequent fires (Hennenberg et al., 2006; Lloyd et al., 2008). Fires in turn limit tree establishment (Higgins et al., 2000; Hoffmann et al., 2009), keeping the canopy open and creating a positive/amplifying feedback that potentially stabilises savannas in regions where forest is also a viable stable ecosystem state (Beckage and  Ellingwood, 2008; Staver et al., 2011a), although some apparent bistability may be the result of spatial climate variability (Good et al., 2015; Higgins et al., 2023). 

The maintenance of savannas is thus dependent on fires across large parts of their range. This has meant that widespread fire suppression (active or passive via agricultural fragmentation or grazing intensification) has triggered woody encroachment and, in extreme cases, forest invasion (Stevens et al., 2017). These feedbacks between vegetation and fire frequency and intensity have also been implicated in accelerating the invasion of alien grasses that are more flammable and also tolerate higher fire intensities than native grasses (D’Antonio and Vitousek 1992; D’Angioli et al., 2022) (Figure 1.3.8). Fire-related feedback loops may not be as significant in drier savannas where herbivores or low water availability limit the accumulation of grass and thus fuel (Archibald and Hemson, 2016; Dexter et al., 2018), and further research is needed on the tipping dynamics of arid savannas and their potential alternate states (see Drylands

Figure: 1.3.8
Figure 1.3.8: Key feedbacks that could lead to savanna tipping.

Several important thresholds are involved in this tipping point. First of all, fire spread is widely described as a percolation process (Loehle et al., 1996; Favier, 2004) – whereby a burning patch infects neighbouring or nearby flammable patches, thereby propagating fire in flammable landscapes. However, when not enough of the landscape is flammable (in this case, if trees shade grasses to prevent fuel accumulation), fires extinguish, with a clear threshold in fuel cover between ‘connected’ flammable vs. ‘unconnected’ non-flammable landscapes (Cardoso et al., 2022). In theory, this threshold can depend on the model used, but in practice, there appears to be a threshold in fuel cover of ~50-60 per cent, below which fire does not successfully spread (Archibald et al., 2009; Cardoso et al., 2022). Thus, fire suppression initiates woody encroachment or forest invasion, which can in turn decrease landscape flammability further, creating a cascade that results in the irreversible loss of open-canopy savannas.

The rate at which this happens – and ultimately the environmental space in which closed vs. open-canopy ecosystems are viable – depends also on environmental thresholds, but these are more widely disputed. A range of studies has defined the minimum required to sustain a closed forest canopy as ranging between 750 and 1,000mm mean annual rainfall (Sankaran et al., 2005; Staver et al., 2011b; Aleman et al., 2020), but more open but still fire-suppressing canopy can also form at much lower rainfall, for example in the Caatinga (Charles-Dominique et al., 2015; Dexter et al., 2018). The high-rainfall limit for savanna persistence is even less defined, as savannas can occur in areas with well over 1,600mm mean annual rainfall – for example, in the Llanos of Venezuela and Colombia (Huber et al., 2006) or the Beteke Plateau in the Republic of Congo (Nieto-Quintano et al., 2018). 

Moreover, increasing atmospheric CO2 is changing the relative photosynthetic efficiencies of ‘C4’ grasses vs. ‘C3’ trees (with C4 being the more efficient photosynthesis process) (Ehleringer and Björkman, 1977; Bond and Midgley, 2012) and is increasing plant water use efficiency across different plant types (Leakey et al., 2009; Norby and Zak, 2011). This has increased the rate of woody encroachment and forest invasion into savannas, suggesting that vulnerability of savannas to tipping points is accelerating and is not stationary with respect to climate (Higgins and Scheiter, 2012). For this reason, defining exactly how much global change might trigger savanna tipping points is not feasible (and indeed a single global tipping point may not exist). 

Several lines of evidence provide support for the irreversibility of savanna-to-forest transitions. First, palaeoecological studies have suggested that reversible increases in rainfall can result in irreversible shifts from savanna to forest, consistent with hysteresis (i.e. where reversing the driver of change does not lead to recovery; see Glossary) (Karp et al., 2023). Second, and more directly, fire experiments have demonstrated that, while fire suppression causes savannas to transition to forest-like systems, introductions of fire into forests have much smaller effects (Gold et al., 2023), likely because closed forest canopies prevent fuels from accumulating to fuel intense savanna fires. This demonstrates that managed fire reintroductions are not sufficient to reverse forest encroachment (Gold et al., 2023). 

Extreme fires can help reverse encroachment by forests when trees are fire sensitive (Silvério et al., 2013, Brando et al., 2014, Beckett et al., 2022) but extreme fires do not reverse woody encroachment (Strydom et al., 2023). In the case of savanna invasions by non-native grasses, irreversibility of transitions may be further exacerbated by resulting changes in nutrient cycling (Bustamante et al., 2012; D’Angioli et al., 2022). Together, these diverse lines of evidence suggest that savanna invasions, once initiated, may be rapid and irreversible.

The timescale of woody encroachment varies depending on environmental controls, but can happen in less than a decade, with accelerating vulnerability across savanna ecosystems due to rising CO2 (Stevens et al., 2022) and widespread enthusiasm for climate mitigation via tree planting (Bastin et al., 2019, Fagan et al., 2022). 

The climate impacts of woody encroachment and forest invasion are uncertain, however, due to substantial carbon in belowground pools in savannas (Zhou et al., 2022) and large uncertainty in how belowground carbon pools (root biomass and especially soil organic carbon) will respond to increasing woody cover (Veldman et al., 2019; Zhou et al., 2023). Hydrologically, there is evidence that an increasing tree fraction can increase rainfall interception and accelerate ecosystem water use, depleting groundwater recharge and streamflow, with implications for downstream water availability (Jackson et al., 2005, Honda and Durigan, 2016). Feedbacks with albedo (with woody vegetation being ‘darker’ than grass) have also been discussed, but little studied (Stevens et al., 2022).

Assessment and knowledge gaps

We have high confidence that Savannas are undergoing widespread degradation from woody encroachment, forest invasion, afforestation and alien grass invasion, high confidence that this is related to grazing intensification and active fire suppression and medium confidence that this is exacerbated by increasing CO2 levels. These changes are increasingly difficult to reverse with the reapplication of fires (medium confidence), although sensitivity of invading vegetation to climate extremes is variable or unknown (Zeeman et al., 2014; Case et al., 2020). Compounded by agricultural conversion and tree planting, this is rapidly eroding endemic savanna and grassland biodiversity (high confidence) (Smit and Prins, 2015, Andersen and Steidl, 2019, Wieczorkowski and Lehmann, 2022). 

Overall, savannas are likely to feature tipping dynamics at local to landscape scales (medium confidence), although large-scale synchrony may be observed if global change drivers trigger tipping points. However, Earth system feedbacks associated with savanna degradation are highly uncertain (low confidence), with particular knowledge gaps about carbon and hydrological cycle outcomes. Potential tipping points in savannas and grasslands associated with herbivory represent another major knowledge gap. 

Bezos Earth Fund University of Exeter logo
Earth Commission Systems Change Lab logo Systemiq logo
Global Tipping Points logo
Share this content