Drought, fire and grazing precursors to large-scale pine forest decline

Aim: Temperate forests are currently facing multiple stresses due to climate change, biological invasions, habitat fragmentation and fire regime change. How these stressors interact with each other influences how, when and whether ecosystems recover, or whether they adapt or transition to a different ecological state. Because forest recovery or collapse may take longer than a human lifetime, predicting the outcomes of different stressor combinations remains difficult. A clearer vision of future forest trajectories in a changing world may be gained by examining collapses of forests in the past. Here, we use long-term ecological data to conduct a post-mortem examination of the decline of maritime pine forests ( Pinus pinaster Ait.) on the SW Iberian Peninsula 7000– 6500 years ago. Location: Portugal and Spain. Methods: We compared four palaeoecological records— two with pine declines and two without— using a multiproxy approach. Bioclimatic differences between the four sites were explored. Proxies for past vegetation and disturbance (fire and grazing)


| INTRODUC TI ON
In the current era of rapid environmental change, ecosystems are under stress on various fronts.Climate change, biological invasions, habitat fragmentation and fire regime change are among the most critical stressors (Slingsby et al., 2017;Trumbore et al., 2015;Turco et al., 2018).Ecological outcomes of multiple interacting stressors are challenging to predict, as these may be greater, less or equal to the sum of the effects (Batllori et al., 2017;Côté et al., 2016;Enright et al., 2014;Foster et al., 2016).Adding to this challenge are the longterm legacies of multiple stressors-those whose ecological effects are only realized decades to millennia later (Essl et al., 2015;Kelly et al., 2011).Observations over short time series may only tell part of the story, especially concerning forest ecosystems in which longlived species play key functional roles and recovery times are slow (Gonzalez et al., 2016;Taranu et al., 2018;Trumbore et al., 2015;Willis et al., 2018).
Ecological disturbances in forests can precipitate permanent state shifts to shrublands (Enright et al., 2015;Karavani et al., 2018;Willis et al., 2010).Fire-induced deforestation is of particular concern for conifer forests, given their high flammability and their economic and cultural values as global carbon sinks and sources of timber, resins and edible fungi (Abad Viñas et al., 2016;Whitman et al., 2019).In the Mediterranean region, pine forests may permanently shift to shrublands when (a) damage to trees and seedbanks is fatal, (b) no mature trees are present nearby to initiate recovery and (c) replacement resprouting tree species are absent (Karavani et al., 2018).Fire and drought are considered the two most critical disturbance agents driving this process, both having strong selective effects on plant traits (Batllori et al., 2017;Berdugo et al., 2020;Karavani et al., 2018;Seidl et al., 2017).Interactions between fire and drought are complex.Drought tends to increase the probability of fire in high-biomass vegetation, while it reduces fire probability in low-biomass systems (Frejaville & Curt, 2015;Pausas & Paula, 2012;Pausas & Ribeiro, 2013).
Post-fire vegetation recovery is influenced by plant functional traits and the prevailing weather conditions.Drought conditions following a fire may delay recovery where the species pool is composed of obligate seeders, but even resprouters may suffer exhaustion under such conditions (Karavani et al., 2018;Parra & Moreno, 2018).
The complexity of these interactions requires multiple-stressor models to forecast ecosystem responses to environmental change.
Validation of these models is a major challenge, as observational records tend to encompass shorter time frames than the fire-and drought-frequency parameters currently being modelled (Barros et al., 2018;Batllori et al., 2017;Mouillot et al., 2002).Without empirical validation, models may be difficult to apply to real-world conservation and management problems (Côté et al., 2016).
In this paper, we analyse multiple stressors acting on the longlived tree species Pinus pinaster Ait.(maritime pine).P. pinaster is the most widespread conifer on the Iberian Peninsula and of major economic importance (Prieto-Recio et al., 2015;Torres et al., 2016).It possesses traits linked to frequent fire occurrence, including thick bark, high rates of post-fire seedling emergence and production of serotinous cones (Tapias et al., 2004;Tavşanoğlu & Pausas, 2018).
These traits are traded off in different populations, with resistance traits (thick bark) characteristic of the Western or Atlantic populations, and recovery (serotiny) and resilience (drought tolerance) traits prominent in the Eastern or Mediterranean populations (Tapias were compared with independent palaeoclimatic records.We performed functional traits analysis and used phase plots to examine the causes of pine decline.

Results:
The pine decline represents a critical transition in SW Iberia, which lies close to maritime pine's bioclimatic limits.Prolonged drought likely killed trees and suppressed the fires that normally stimulate pine germination and pinewood recovery.
Increased grazing pressure facilitated the rapid spread of resprouter shrubs.These competed with pine trees and ultimately replaced them.Our data highlight complex interactions between climate, fire, grazing and forest resilience.

Main Conclusions:
The pine decline occurred at least a century after post-fire resprouters overtook obligate seeders in the vegetation, constituting an early-warning signal of forest loss.Fire suppression, resprouter encroachment and grazing may threaten the persistence of Mediterranean forests as droughts become more frequent and extreme.

K E Y W O R D S
fire regime change, forest dynamics, functional traits analysis, palaeoecology, phase plots, tipping point et al., 2004;Zas et al., 2020).P. pinaster distribution ranges from sea level to 2,100 m along a rainfall gradient from 350 to 1,400 mm p.a. (Alía & Martín, 2003).
Here, we re-examine the causes of pine decline in the light of robust indicators for fire, grazing, drought and plant functional traits.
We hypothesize that interactions between functional traits and disturbance regimes govern long-term forest resilience and recovery.Major changes to disturbance regimes and/or community-level functional trait assemblages could lead to a loss of forest resilience and permanent state shifts from forest to heathland.To test this, we compare a new multiproxy record with three previous sequences that reflect stand-scale dynamics to reconstruct functional traits and their interactions with disturbance regimes at contrasting sites.We also predict that the ecological impacts of disturbance regime change are mitigated by local bioclimate.To assess this, we analyse pine trajectories and climatic variables to find differences between pine decline sites and areas of pine forest continuity.This study contributes to an understanding of multiple-stressor combinations that herald forest collapse and discusses how such collapses may be avoided.

| Study area
We selected four study sites for comparison-two sites with pine continuity and two with evidence of pine decline during the mid-Holocene (8,200 cal.yr BP, Figure 1a).The continuity sites are Espinosa de Cerrato (ESCE) and El Carrizal (ELCA), located in the Spanish Northern Meseta (Franco Múgica et al., 2001, Franco-Múgica et al., 2005;Morales-Molino et al., 2012, 2017).Pinus pinaster was once dominant at ELCA and present among P. nigradominated vegetation at ESCE (Morales-Molino et al., 2017).The two pine decline sites are Lagoa Travessa (LATR, Mateus, 1992) and a new site, Barbaroxa de Baixo (BXBX, 38.0790N, 8.8098W), both located on the Alentejo coast south of Lisbon (see Appendices S1.1-3).The four sites were selected with attention to factors that influence the source and fidelity of proxy data (Jacobson & Bradshaw, 1981;Whitlock & Larsen, 2001).To address the aim, the sites had to: (a) be small enough to record eco-
Pollen, an indicator of past vegetation, was extracted from the sediments using standard acetolysis-based techniques (Moore et al., 1991).Pollen was identified using regional guides (listed in Appendix S1.3).Particular attention was paid to the identification of Ericaceae and Cistaceae pollen, following morphological criteria developed by Queiroz (1999) and Mateus (1989).
Microscopic charcoal sequences are also available for three of the sites (ESCE, ELCA and BXBX), quantified following Finsinger and Tinner (2005).Charcoal records are considered robust indicators of biomass burned and fire episode frequency (Ali et al., 2012).
Methods to reconstruct other key aspects of fire regimes (intensity, severity and seasonality) from charcoal are still under development.

| Numerical analyses
To reconstruct a mid-Holocene fire history for BXBX, charcoal accumulation rates (CHAR) were calculated and normalized to z-scores following Power et al., (2008).For the macroscopic charcoal record, fire peaks related to local fire episodes were separated from the "background" using a 500-year lowess smoother in CharAnalysis (Higuera et al., 2009).Background charcoal reflects long-term changes in charcoal production and dispersal (Whitlock & Larsen, 2001) and approximates the amount of biomass that has been burned over time (Vannière et al., 2016).
Possible drivers of fire activity were examined by comparing the BXBX charcoal record with regional palaeoclimatic records (Cacho et al., 1999(Cacho et al., , 2001;;Rodrigues et al., 2009;Thatcher et al., 2020), local changes in geochemistry and diatom assemblages (Cruces, 2015, Leira et al., 2019), and lake level variations at BXBX.The latter were derived from detrended correspondence analysis of aquatic and wetland indicators, classified as deep-water limnic, shallow-water telmatic and semi-terrestrial taxa based on modern analogues (Queiroz, 1999).
Functional traits are key to understanding ecosystem resilience and responses to multiple stressors (Batllori et al., 2017;Enright et al., 2014).Integration of plant functional traits and palaeoecological data provides unique long-term insights into vegetationdisturbance interactions (Brussel et al., 2018).We assigned trait scores to pollen taxa to permit interpretation of the pollen sequences in terms of plant functional traits (Barboni et al., 2004;Brussel et al., 2018).Trait scores were derived from Iberian records in the BROT2 database (Tavşanoğlu & Pausas, 2018).Selected traits were growth form, post-fire regeneration strategy, spinescence and taxonomic class, representing a range of adaptations to fire, drought and grazing (Brussel et al., 2018;Hanley et al., 2007;Tavşanoğlu & Pausas, 2018).Taxa were included if they were: Observed traits were compared with 1,000 null models in which traits were randomly assigned to each of the pollen taxa (Brussel et al., 2018).
Resilience indicators were examined in each of the four records to understand whether the pine decline represents a critical transition.
Standard deviation is considered an appropriate resilience indicator for palaeoecological data with uneven temporal sampling (Stegner et al., 2019).The standard deviation of pine pollen sequences was analysed in R using code provided in Stegner et al., (2019) and assessed with Kendall's tau, a nonparametric correlation statistic (Dakos et al., 2010).
Phase plots illustrate interactions between a system's ecological state and environmental drivers (Davies et al., 2018;Willis et al., 2010).Here, we used phase plots to examine the causes of pine decline, comparing Pinus pollen percentages with rates of change, climate, fire and grazing proxies.

| Pinus pinaster bioclimate
Analysis of P. pinaster distribution compared with bioclimatic variables shows that the two pine decline sites, LATR and BXBX, occupy the lower limit of the species' current range in terms of dry season precipitation (Figure 1b).This contrasts with the more central position of ELCA and ESCE sites within or near the Spanish Tierra de Pinares, where pollen and macrofossil data indicate pine forest continued uninterrupted through the mid-Holocene (Franco Múgica et al., 2001;Franco-Múgica et al., 2005;Morales-Molino et al., 2012, 2017).

| Pollen (vegetation)
The most prominent feature of the pine decline sites is the pine decline itself (Figure 1c) and subsequent expansion of shrub taxa (Erica scoparia, Corema album and Juniperus-see Appendix S1.7).
At BXBX, the better dated of the two pine decline sites, pine pollen percentages decreased from 60% to 25% in the 200 years between 6,915 and 6,715 cal.BP (Figure 2).Pine pollen accumulation rates declined from 2040 to 850 grains cm −2 year −1 .A comparable decline occurs at LATR between 7,000 and 6,000 cal.yr BP, whereas no significant change in median Pinus values occurs at the other sites (Figure 1c).

| Trait scores
Reconstructed functional trait scores for the four sites appear in Figure 3. Life-form traits showed strong non-random selection at the two pine decline sites, with shrub abundance exceeding 95% confidence intervals (CIs) of the null models after the pine decline.
No evidence of non-random selection is apparent at the pine continuity sites.Pine declines occurred after resprouters became more abundant than obligate seeders at BXBX and Lagoa Travessa (Figure 3).Spinescence also exhibited non-random selection at LATR and BXBX, particularly after the pine decline.
No significant change is recorded at the pine continuity sites.
Pine decline sites experienced rapid rates of change during the decline compared with continuity sites (Figure 4a).Precursors of the pine decline include a temporary reduction in regional precipitation, a decrease in local fire activity and an increase in grazing indicators (Figure 4b-d).
F I G U R E 3 Traits analysis of the palaeoecological records with 90% and 95% confidence intervals derived from bootstrapping (dotted lines; see Brussel et al., 2018).Upper panels: life-form (plant height).Lower panel: post-fire regeneration strategy (ability to resprout or reemerge as seedlings after fire) and spinescence.PD: pine decline

| Causes of the pine decline
Pine decline represents a critical transition in the ecological history of the Western Mediterranean (Table 1).A combination of drought, fire suppression, interspecific competition and the expansion of grazing appears to have driven pine decline in SW Iberia during the mid-Holocene.These drivers had greater effect in SW Iberia as this location lies close to the bioclimatic limits for Pinus pinaster compared with the Spanish Northern Meseta (Figure 1).It is probable that the SW Iberian populations were also more drought sensitive than the Spanish populations (Zas et al., 2020).

| Drought
Drought stress is regarded as a key predictor of mortality in Iberian P. pinaster stands in the present day (Navarro-Cerrillo et al., 2018;Prieto-Recio et al., 2015).Droughts cause stem contraction in P. pinaster, and the trees enter a quiescent state (Vieira et al., 2013).In this state, prolonged and/or severe droughts lead to carbon starvation, increased susceptibility to pathogen attack, and eventually hydraulic failure and tree mortality (McDowell et al., 2008).In SW Iberia during the mid-Holocene, a multidecadal drought is clearly registered in marine and terrestrial isotopic records between 7,000 and 6,500 cal.

| Fire regime change
The failure of SW Iberian P. pinaster populations to recover after the end of the drought suggests that fire and competition factors came into play.Ecological models conceptualize fire as a key driver of deforestation in the Mediterranean region (Baeza et al., 2007;Batllori et al., 2017;Karavani et al., 2018;Mouillot et al., 2002) and in Mediterranean-type ecosystems globally (Bowman et al., 2013;Enright et al., 2015).Fire-induced deforestation is also implicated in the creation and persistence of many European heathlands (van der Knaap & van Leeuwen, 1995;Loidi et al., 2010;López-Merino et al., 2012;Odgaard, 1992;Odgaard & Rasmussen, 2000).Our data provide another possibility-that fire deprivation or suppression can lead, under conditions of environmental stress for pines, to a state shift from pine forest to heathland.
The regular occurrence of fires prior to the pine decline (Figure 2) shows the pine forests were resilient to fire return intervals as low as 30 years (cf.Garcia-Gonzalo et al., 2011, Leys et al., 2014, Mouillot et al., 2002).Pine decline occurred during two centuries of local fire absence, the longest fire-free interval of the mid-Holocene at BXBX.Pine decline in SE Spain also occurred during a period of low fire activity (Carrión et al., 2001).Pinus pinaster tolerates a variable or mixed fire regime (Fernandes & Rigolot, 2007), yet pine seedling emergence rates decrease rapidly with time since fire (Pausas et al., 2008).This is perhaps due to the short life span of P. pinaster seeds once released from the cones (Ferrandis et al., 1996).A 200year absence of fires is likely to have had detrimental effects on pine regeneration depending on levels of serotiny in the population.
However, fire's absence cannot explain the inability of pine populations to recover after the pine decline, given that regular fires returned after the drought phase (Figure 2) and pines in Western Iberia typically produce no or few serotinous cones (Tapias et al., 2004).
Competition and grazing thus emerge as probable explanations for poor pine recovery.

| Interspecific competition
Interspecific competition at the pine decline sites is indicated by the switch in dominance from seeders to resprouters in Figure 3.This dominance shift occurs before the pine decline and may represent an early-warning signal for pine decline.Ecological studies indicate that shrubby resprouters are quicker to recover from drought than obligate seeders (Parra & Moreno, 2018;Zeppel et al., 2016) and often outcompete pine trees in post-fire recovery (Calvo et al., 2008;Nuñez et al., 2003;Taboada et al., 2017).Pinus pinaster seedlings are shade-intolerant and often outcompeted by resprouters (Batllori et al., 2017;Calvo et al., 2008;Torres et al., 2016), which also compete with Mediterranean pine forests for moisture, promoting drought stress (Karavani et al., 2018).Our data suggest that pine decline was the result of both abiotic stress (multidecadal drought) and biotic interactions (competition) in the context of fire regime change (Carrión et al., 2001(Carrión et al., , 2003(Carrión et al., , 2010)).
Our observations suggest that obligate seeder trees may be replaced by shrublands in the absence of fire (Figure 4).Ecological models that simulate the effects of drought-fire interactions in Mediterranean vegetation provide little indication of this potential outcome (Batllori et al., 2017;Mouillot et al., 2002).This may be because the duration of mid-Holocene drought in SW Iberia was longer than those simulated in models (e.g. 15 years in Batllori et al., 2017).
It may also reflect the additional effect of grazing pressure, which has received less attention in regional modelling simulations, but is regarded as a critical top-down control on woody plant populations (Archibald & Hempson, 2016;Bond, 2005;Midgley et al., 2010).

| Grazing
Grazing is a disturbance that persisted long after the drought ended in SW Iberia (Figure 4).Erica scoparia, the main resprouter that replaced pine at LATR and BXBX, is promoted by grazing (Bartolomé et al., 2005).The increase in grazing indicators (Figures 2 and 4) prior to pine decline indicates that herbivory may have played a role in triggering and maintaining the state shift.
Grazing reduces biomass, impacting on fire regimes (Archibald & Hempson, 2016).In the south-western USA, when European farmers introduced livestock to pine forests in the 19th century, grazing prevented surface fires that had previously been a regular occurrence (Swetnam et al., 1999).This grazing-induced change in fire regime allowed pine competitors to proliferate (Fulé et al., 1997;Savage & Swetnam, 1990;Swetnam et al., 1999).A review by Richardson et al., (2007) describes numerous examples of grazing impacts on pine forests worldwide.
As SW Iberia was colonized by Neolithic farmers, we suggest that the domestic livestock suppressed fire and altered pine recruitment, accelerating pine decline.Native herbivores have been shown to have significantly different effects on seedling establishment, plant diversity and vegetation structure compared with introduced livestock (Perea et al., 2016).(Carvalho, 2002;Soares, 1995Soares, , 1996)).
Intense grazing in the absence of fire strongly favours Erica scoparia (Bartolomé et al., 2005), the taxon that expands most rapidly before and during the pine decline.Bartolomé et al., (2005) suggest that fire puts downward pressure on E. scoparia recruitment by depleting seedbanks, whereas grazing facilitates rapid spread.The decline in spinescent shrubs (Figure 3) may also relate to overgrazing.Spinescence is a trait regarded as an early evolutionary defence against vertebrate herbivory (Charles-Dominque et al., 2016;Hanley et al., 2007) and is often ineffective against grazers in Mediterranean shrublands during drought (Kohl et al., 2014;Papachristou et al., 2005;Rogosic et al., 2006).

| Integrating drought, fire and grazing stressors
The relative importance of herbivory and fire in biomass consumption is strongly dependent on rainfall gradients.Fire activity in Mediterranean zones can be "switched on" when climate-controlled thresholds are crossed and fuel loads/connectivity increase (Pausas & Paula, 2012).In the African context, herbivory is the primary consumer in low rainfall zones, but fire takes over as rainfall increases (Archibald & Hempson, 2016).This threshold is lower on nutrientpoor soils.Hence, biomass consumption in drought phases (such as the pine decline) may be dominated by herbivory (Figure 2).Grazing's potential impact on fire spread is integrated into Pausas and Paula's (2012) aridity/productivity gradient in Figure 5, along with a conceptual model of ecological succession in maritime pine forests that compares grazing and drought conditions to fire-led succession.
Given the prediction of more severe droughts in the future (Batllori et al., 2013;Turco et al., 2018), there are suggestions that resilience of Mediterranean forests could be improved by artificially introducing resprouters into the understorey (see Gavinet et al., 2016;Karavani et al., 2018;Moreno-Fernández et al., 2018).
Our data suggest that such a strategy could push maritime pine forests closer to a tipping point where forest loss could be expected under drought conditions, grazing pressure or as a result of interspecific competition (Calvo et al., 2008;Prieto-Recio et al., 2015).
Addition of resprouters may increase moisture stress (Karavani et al., 2018) and fuel connectivity in a way that encourages severe, mortality-inducing canopy fires (Botequim et al., 2017).Greater impacts might be expected among pine populations with few recovery and resilience traits (Zas et al., 2020).As grazing also favours shrubby resprouters, management cannot rely on vertebrate herbivores to replace fire's ecological functions (Bartolomé et al., 2005;Hean & Ward, 2012).

| CON CLUS IONS
The aim of this paper was to analyse a critical transition from Mediterranean forest to shrubland.The palaeorecord provides strong evidence of threshold responses in vegetation and fire regime in response to drought and grazing pressures at the bioclimatic limit of Pinus pinaster's range.Our data support the theory that relatively frequent fire (30-130-year return interval) gave mid-Holocene P. pinaster an advantage over competitors.These competitors showed early-warning signals of overtaking pine, manifested as a dominance switch from obligate seeders to post-fire resprouters at least a century before pine decline.We find a good agreement between the trajectory of the pine decline and ecological models involving multiple stressors, which provide more convincing mechanisms for the pine decline than an aridification trend or human impact.However, current models require improvements to predict forest-to-shrubland transitions where drought and grazing interact to suppress fire and thus limit recruitment in obligate seeders.
The most important implication of our research is that manage-

F
I G U R E 1 (a) Map of Pinus pinaster's distribution (shaded) and the study sites mentioned in the text (ESCE-Espinosa de Cerrato; ELCA-El Carrizal; LATR-Lagoa Travessa; BXBX-Barbaroxa de Baixo); (b) bioclimate of P. pinaster on the Iberian Peninsula (Abad Viñas et al., 2016; Appendix S1.1), showing the position of the four study sites in relation to dry season precipitation and warm season temperatures; (c) Pinus pollen percentages in the four palaeoecological records before and after the greatest decrease in pine pollen during the mid-Holocene system dynamics at a local to extra-local (stand) scale; (b) have minimal stream inflow and slope run-off that could introduce material from beyond the local area; (c) contain sediments that have accumulated continuously and rapidly to ensure an unbroken record with complete fossilization.
(a) easily determined to species or genus level and attributable to one/few plant species; (b) sufficiently abundant in the pollen record to represent population dynamics; (c) representative of plants in the site's vicinity (in the case of well-dispersed Pinus and Juniperus pollen, conifer stomata or macrofossils are reliable indicators of local presence: Ammann et al., 2014); and (d) unambiguously associated with terrestrial vegetation.

F
Temporal trends in principal proxies of the BXBX palaeoecological record: (a) wetland taxa ordination scores (DCA axis 1); (b) macroscopic charcoal abundance expressed as z-scores (line) and significant charcoal peaks (triangles, see Appendix S1.10); (c) combined abundance of grazing indicators(Apiosordaria,  Cercophora, Coniochaeta, Podospora, Sordaria, Sporormiella)  shown as fungal spore accumulation rates (spores/cm 2 /year) for raw (dashed line) and lowess-smoothed data (solid line); (d) pine pollen percentages with 95% confidence intervals.See Appendices S1.7-9 for complete pollen, spore and charcoal records proxy, macroscopic charcoal, has an average sampling resolution of 10 years and is the most abundant during two phases: 7180-6825 and 6615-5825 cal.BP (Figure2).Peak analysis of macroscopic charcoal yielded an average fire return interval of approximately 80 years in the early part of the record, with a decreasing frequency after 6,300 cal.BP.The longest fire-free interval was approx.210 years (6825-6615 cal.BP) and the shortest 30 years (6925-6895 cal.BP; Figure2).Extended fire-free intervals after 6,000 cal.BP are accompanied by the highest representation of fire-sensitive Juniperus (Appendix S1.7).Longer fire-free intervals in the late Holocene are not considered here (see Appendix S1.7).Detrended correspondence analysis (DCA) of BXBX wetland indicators (Appendix S1.8) weighted limnic taxa positively on axis 1 (e.g.Pediastrum algae and Nymphaea alba pollen) and semi-aquatic and telmatic taxa negatively (e.g.Hydrocotyle, Cyperaceae, Hypericum elodes).The DCA trend shows strong affinities with regional precipitation changes in isotopic records, local hydrological changes and fire history (Figure2, Appendix S1.11).The latter is highly correlated with the DCA result (Spearman's rho: 0.73, p < .001).
Testing for a critical transition in each of the palaeorecords around 6,750 cal.yr BP.Sites listed in N-S order

F
Phase plots showing (a) pine rate of change (ROC) with the rapid decline of Pinus pollen at pine decline sites (solid lines) compared with pine continuity sites (dashed lines); (b-d) driver-response relationships in the BXBX record (smoothed curves).Sea-surface temperatures fromRodrigues et al., (2009) ment decisions made now (in relation to grazing, fire and resprouter establishment) will determine how future ecosystems respond to climate change.F I G U R E 5 A conceptual model for fire-versus drought/grazingmodulated vegetation dynamics, inferred from the palaeorecord: (a) schematic showing the main directions of vegetation change (solid arrows) and subsidiary pathways (dotted arrows) between mature pines, juvenile pines and resprouter shrubs in relation to fire and drought/grazing; (b) the relative effects of grazing and fire in consuming biomass along a gradient of available moisture (adapted from Archibald & Hempson, 2016); and (c) the main drivers of fire in relation to available moisture, showing the hypothesized effect of grazing in reducing the probability of fire spread under fuel-limited conditions (adapted from Pausas & Paula, 2012).Plant images courtesy of Gobius Comunicação e Ciência Common pre-Neolithic herbivores in SW Iberia included red deer (Cervus elaphus), wild boar (Sus scrofa),