Journal list menu

Volume 34, Issue 4 e2971
Open Access

Ungulates mitigate the effects of drought and shrub encroachment on the fire hazard of Mediterranean oak woodlands

Xavier Lecomte

Xavier Lecomte

Forest Research Center, Associate Laboratory TERRA, School of Agriculture, University of Lisbon, Lisbon, Portugal

Center for Applied Ecology “Prof. Baeta Neves” (CEABN-InBIO), School of Agriculture, University of Lisbon, Lisbon, Portugal

Search for more papers by this author
Miguel N. Bugalho

Corresponding Author

Miguel N. Bugalho

Center for Applied Ecology “Prof. Baeta Neves” (CEABN-InBIO), School of Agriculture, University of Lisbon, Lisbon, Portugal


Miguel N. Bugalho

Email: [email protected]

Search for more papers by this author
Filipe X. Catry

Filipe X. Catry

Center for Applied Ecology “Prof. Baeta Neves” (CEABN-InBIO), School of Agriculture, University of Lisbon, Lisbon, Portugal

Search for more papers by this author
Paulo M. Fernandes

Paulo M. Fernandes

Center for the Research and Technology of Agro-Environmental and Biological Sciences (CITAB), University of Trás-os-Montes and Alto Douro (UTAD), Vila Real, Portugal

ForestWISE—Collaborative Laboratory for Integrated Forest and Fire Management, Vila Real, Portugal

Search for more papers by this author
Andreu Cera

Andreu Cera

Center for Applied Ecology “Prof. Baeta Neves” (CEABN-InBIO), School of Agriculture, University of Lisbon, Lisbon, Portugal

Search for more papers by this author
Maria C. Caldeira

Maria C. Caldeira

Forest Research Center, Associate Laboratory TERRA, School of Agriculture, University of Lisbon, Lisbon, Portugal

Search for more papers by this author
First published: 05 April 2024

Handling Editor: Juan C. Corley


Climate change is increasing the frequency of droughts and the risk of severe wildfires, which can interact with shrub encroachment and browsing by wild ungulates. Wild ungulate populations are expanding due, among other factors, to favorable habitat changes resulting from land abandonment or land-use changes. Understanding how ungulate browsing interacts with drought to affect woody plant mortality, plant flammability, and fire hazard is especially relevant in the context of climate change and increasing frequency of wildfires. The aim of this study is to explore the combined effects of cumulative drought, shrub encroachment, and ungulate browsing on the fire hazard of Mediterranean oak woodlands in Portugal. In a long-term (18 years) ungulate fencing exclusion experiment that simulated land abandonment and management neglect, we investigated the population dynamics of the native shrub Cistus ladanifer, which naturally dominates the understory of woodlands and is browsed by ungulates, comparing areas with (no fencing) and without (fencing) wild ungulate browsing. We also modeled fire behavior in browsed and unbrowsed plots considering drought and nondrought scenarios. Specifically, we estimated C. ladanifer population density, biomass, and fuel load characteristics, which were used to model fire behavior in drought and nondrought scenarios. Overall, drought increased the proportion of dead C. ladanifer shrub individuals, which was higher in the browsed plots. Drought decreased the ratio of live to dead shrub plant material, increased total fuel loading, shrub stand flammability, and the modeled fire parameters, that is, rate of surface fire spread, fireline intensity, and flame length. However, total fuel load and fire hazard were lower in browsed than unbrowsed plots, both in drought and nondrought scenarios. Browsing also decreased the population density of living shrubs, halting shrub encroachment. Our study provides long-term experimental evidence showing the role of wild ungulates in mitigating drought effects on fire hazard in shrub-encroached Mediterranean oak woodlands. Our results also emphasize that the long-term effects of land abandonment can interact with climate change drivers, affecting wildfire hazard. This is particularly relevant given the increasing incidence of land abandonment.


The increased frequency of extreme events, such as droughts and wildfires, may interact with other drivers of ecosystem changes. Shrub encroachment (Archer et al., 2017) and the expansion of wild ungulates, partially resulting from abandonment of land and outdoor livestock production (Carpio et al., 2020), might critically contribute to these changes. The interaction of these drivers has the potential to affect ecosystem functioning, such as fire resistance (Ali et al., 2022), particularly in hotspot regions for climate change and land abandonment. These are areas where the climate responds rapidly to global warming and where fast land abandonment is occurring (Giorgi, 2006; Kuemmerle et al., 2016), such as Mediterranean regions.

Mediterranean regions are anticipated to face more severe and frequent droughts (Ali et al., 2022). The main effects of drought on plant communities and ecosystems result from decreased soil water availability and higher water evapotranspiration demand (Caldeira et al., 2015). These conditions trigger physiological disorders in plants (Menezes-Silva et al., 2019) and alter soil nutrient cycling (Matías et al., 2011). Drought also decreases net primary productivity (Gao et al., 2019) and may modify plant species composition and community structure (Martínez-Vilalta & Lloret, 2016). Drought legacies, structural damage to plants (Menezes-Silva et al., 2019), or alteration in plant–soil feedbacks may affect plant and ecosystem functioning for years to come (Kannenberg et al., 2020), decreasing plant resistance to subsequent drought events (Peltier & Ogle, 2019) and increasing plant mortality (Anderegg et al., 2019). Shrub encroachment (i.e., the increase in shrub density and biomass) is a global phenomenon, which affects grasslands and savanna-type ecosystems, such as Mediterranean oak woodlands (Sala & Maestre, 2014; Stevens et al., 2017). Mediterranean oak woodlands are traditional cultural human-shaped ecosystems typical of the western Mediterranean Basin, which are denominated “montados” in Portugal and “dehesas” in Spain (Bugalho, Caldeira, et al., 2011). These ecosystems are mainly used for livestock production (grassland and acorns), recreational hunting, and, where cork oak (Quercus suber L.) occurs, cork harvesting. Cork is a nontimber forest product used mainly for wine bottle stoppers and is harvested every 9–12 years, without the need to fell trees—a practice that has increased since the 19th century. Rotational shrub clearing (5–7 years) is a common management practice for improving livestock forage, facilitating cork harvest and reduce shrub cover and fire hazard. However, these ecosystems face challenges, such as rural depopulation, land abandonment, and management neglect, that are causing the encroachment of shrubs (Acácio et al., 2017; Bugalho, Caldeira, et al., 2011).

Shrub encroachment increases fire hazard through fuel load accumulation (Anderson et al., 2015; Lecomte et al., 2019) and by modifying the structural properties of the fuel complex that enhances heat transfer (Fernandes, 2009). Dense and continuous shrub communities, particularly those of flammable species, favor fire propagation, increasing both fire intensity and flame length (Fernandes et al., 2000). Shrub-encroached ecosystems, indeed, are among the most fire-prone ecosystems (Syphard et al., 2007), particularly in Mediterranean-type climate regions (Rego & Silva, 2014). Additionally, drought conditions may increase the understory shrub flammability by directly reducing fuel moisture (Dennison & Moritz, 2009) and indirectly by converting live vegetation into dead fuel (Nolan et al., 2020). For example, in Portugal, where drought conditions have progressively intensified over the last few decades (Páscoa et al., 2021), shrub-encroached oak woodlands are exhibiting a higher frequency of burning (Acácio et al., 2017; Guiomar et al., 2015).

Wild ungulate populations may expand across their geographic distribution areas, due to factors such as land abandonment (e.g., increase in woody areas), land-use changes (e.g., reforestation), and the lack of predators (Carpio et al., 2020); in the Iberian Peninsula, ungulates were also introduced for hunting purposes, namely, in oak woodlands (Bugalho et al., 2001). The expansion of wild ungulates may have negative effects on the density, recruitment, and height of woody species through the consumption of plant biomass (Bugalho et al., 2013; Horsley et al., 2003). This may ultimately affect the tree and shrub species composition of the understory over the long term (e.g., Boulanger et al., 2018). The negative effects of ungulates on woody plants can be further aggravated by drought conditions (Vuorinen et al., 2020), causing woody plant mortality and, thus, increasing fuel load and fire hazard (Lecomte et al., 2019; Lovreglio et al., 2014).

The combined effects of drought and browsing on fire hazard are seldom addressed and understood, despite their critical importance for the conservation and management of ecosystems (Doblas-Miranda et al., 2017). In this study, we investigate how drought, shrub encroachment, and ungulate browsing jointly affect fire hazard in Mediterranean oak woodlands. This is crucial due to the potential long-term implications for the structure and functioning of these ecosystems. Increased understory fire hazard may lead to canopy fire, resulting in high adult oak mortality (Lecomte et al., 2019).

We use data from an 18-year-old ungulate fencing exclusion experiment conducted in a Mediterranean oak woodland, where outdoor livestock production was abandoned and wild ungulates occur. Our research plots were encroached by Cistus ladanifer L., a dominant native shrub in our study area (Bugalho, Lecomte, et al., 2011; Lecomte et al., 2016, 2017, 2019). The fencing exclusion experiment simulated long-term land abandonment and management neglect, comparing conditions with (no fencing) and without (fencing) wild ungulate browsing. In particular, within ungulate browsing excluded areas (fenced), C. ladanifer shrubs have the potential to expand into the oak woodland understory, from the large and persistent soil seed banks (Lecomte et al., 2016), due to lack of shrub browsing and shrub clearing. In the present study, we aim to understand how cumulative drought interacts with shrub encroachment and ungulate browsing, ultimately affecting fire hazard and ecosystem resilience. We compare (1) shrub population density, biomass, and fuel load and (2) modeled fire behavior (rate of fire spread, fire intensity, and flame length) in plots browsed and protected from wild ungulates, considering both drought and nondrought scenarios.

We hypothesize that (1) overall, drought increases the flammability and fire hazard of shrub understory through higher accumulation of dead shrub biomass and by lowering the shrub live-to-dead ratio, with a more pronounced effect inside exclosures; (2) ungulate browsing combined with drought further increases dead shrub biomass and, therefore, fire hazard outside the exclosures; however, (3) over the long term (18 years), cumulative ungulate browsing reduces shrub biomass and fuel load, mitigating the increased effects of drought on fire hazard (modeled in the drought scenario outside the exclosure) (Figure 1).

Details are in the caption following the image
(A) Illustrative graphic: Persistent ungulate browsing pressure over the long term can reduce shrub encroachment and total fuel load and, therefore, mitigate fire hazard in shrub-encroached ecosystems (Lecomte et al., 2019). Browsing under drought conditions increases shrub mortality and flammability by lowering the live-to-dead ratio of plant biomass, increasing fire hazard. (B) Unbrowsed plot in 2013, after cumulative drought with 17% shrub mortality. (C) Cistus ladanifer individual browsed by deer species. (D) Browsed plot in 2013 with nearly 85% shrub mortality. The paired unbrowsed plot is visible in the background. Content creator (A) and photo credit (B–D): Xavier Lecomte.


Study site

The study site is a 900-ha Mediterranean oak woodland located in Vila Viçosa, southeast Portugal (38°49′ N, 07°24′ W). This is a privately owned enclosed estate that has been traditionally managed, involving activities such as shrub clearing, outdoor livestock production, and cork harvesting, which gained importance starting in the 19th century. In 1994, red (Cervus elaphus L.) and fallow (Dama dama L.) deer were introduced in the estate for hunting purposes, replacing outdoor livestock production, while cork harvesting and shrub clearing were maintained. Red deer are native to the region, while fallow deer were probably introduced during Roman times (ca. first to fifth centuries ad) (Davis & MacKinnon, 2009). During the early 20th century, both red and fallow deer were extirpated from the region, mainly due to overhunting, with small restricted populations remaining in different parts of the country. However, in recent times, free-ranging wild ungulates, such as red and fallow deer, are expanding in Europe, including the Iberian Peninsula, and becoming overabundant. This expansion is attributed to favorable habitat changes related to land abandonment and land-use changes, such as afforestation. Additionally, introduction in some estates for hunting purposes contributes to the increasing abundance of these ungulates (Carpio et al., 2020). Despite the absence of external deer migration into the enclosed estate, deer population densities have been on the rise. This increase is because of a limiting deer-culling policy defined and implemented by the land manager for hunting purposes. In 2001, at the start of the study, there were 0.35 individuals of red deer and 0.1 individuals of fallow deer per hectare (Bugalho, Lecomte, et al., 2011). More recently, in 2015, overall deer densities surpassed one deer per hectare (M. N. Bugalho & X. Lecomte, unpublished data). Such deer densities are not uncommon in the Iberian Peninsula (Camargo et al., 2021).

The climate is of a Mediterranean type, characterized by mild and wet winters and hot and dry summers. The long-term (1981–2010) mean annual precipitation in the study site is 585.3 mm, with a coefficient of variation of 26.7%. Precipitation mainly occurs between October and May. The long-term mean maximum temperatures reach 31.1°C in July, the hottest month, while mean minimum temperatures reach 5.8°C in January, the coldest month, resulting in mean annual temperatures of 15.9°C (Évora meteorological station, 1981–2010, accessed in June 2017, htpp:// During the study period (2001–2019), precipitation had a coefficient of variation of 27.3% and a SD of 144.0 mm.

We define a drought event as a negative precipitation deviation from the long-term average (1981–2010), during a hydrological year, which spans from October to September (e.g., Wilhite & Glantz, 1985). Additionally, we defined the intensity of drought years according to the Standardized Precipitation Index (SPI) (McKee et al., 1993). During the study period, seven major drought years occurred, ranging in severity from moderate to extreme, with four of them classified as extreme drought events (SPI ranging from −1.50 to −1.99). These extreme events characterized by a precipitation deficit above 200 mm took place in 2004–2005 (total precipitation of 346.1 mm), 2007–2008 (359.2 mm), 2011–2012 (320.9 mm), and 2018–2019 (326.8 mm) (Figure 2). The cumulative precipitation deficit was only 11.3 mm between 2001 and 2007 but increased to 454.1 mm in 2013, 612.1 mm in 2015, and 1054.2 mm in 2019. From June to September (2002–2019), the critical fire period in Portugal, the mean, maximum, and minimum wind speed were 9.60, 24.48, and 1.08 km/h, respectively.

Details are in the caption following the image
Precipitation deviations from long-term climatic normal average (585.3 mm for 1981–2010 period) during hydrological years (1 October–30 September) from 2001 to 2019 in southern Portugal (38°31′ N, 8°01′ W; data from Drought events are defined as periods of below-normal precipitation over the hydrological period of October to September. The light gray bars represent seven major dry years that occurred during the period of study, with four extreme years (2004–2005, 2007–2008, 2011–2012, and 2018–2019) characterized by a precipitation deficit exceeding 200 mm. Cumulative precipitation deficit is represented at the top of the figure for each sampling year.

The study site is characterized by poorly developed haplic leptosols, with a dominant bedrock of schist. The topography is smooth, with an average elevation of 395 m. Tree cover is composed of the evergreen holm oak (Quercus rotundifolia Lam.) and cork oak (Q. suber L.), while the shrub understory is dominated by the evergreen shrub gum rockrose cistus (C. ladanifer L.). The herb understory mainly consists of annual grass species (e.g., Bromus madritensis L., Gaudinia fragilis (L.) P.Beauv.), forbs (e.g., Andryala integrifolia L., Leontodon taraxacoides (Vill.) Mérat), and legumes (e.g., Vicia disperma DC., Ornithopus compressus L.). This vegetation cover within the estate is typical of oak woodlands in the region.

Cistus ladanifer is a malacophyllous and drought-tolerant shrub species native to the western Mediterranean area. It has a relatively long life span of up to 30 years, distinguishing it from other Cistus species (de Vega et al., 2008). This species is highly competitive and well adapted to stressful conditions, such as low soil water and nutrient contents, and high exposure to solar radiation and temperature. C. ladanifer is able to quickly recover from drought by efficiently using pulses of rainfall (Caldeira et al., 2015; Ramírez et al., 2012). The leaves of C. ladanifer are impregnated by terpene-derived resin (Haberstroh et al., 2018), increasing its flammability, delaying litter decomposition, and promoting fine fuel accumulation (Horner et al., 1988), thereby enhancing fire spread. In summer, C. ladanifer maintains a low tissue moisture content, with values that may reach 80% in midsummer and less than 50% in dry years, contributing to its high flammability (Chuvieco et al., 2009). Additionally, C. ladanifer is an obligate seeder, recruiting only from the soil seed bank (not resprouting species), and producing seeds that can remain viable for 6–7 years (Clemente et al., 2007). C. ladanifer is also a fire-adapted species (Calvo et al., 2005). Although not a preferred forage, ruminant herbivores, such as deer, consume C. ladanifer, especially during summer when most of the annual herbs are senescent and of low nutritional value (Bugalho et al., 2001; Bugalho & Milne, 2003).

Experimental design

The study spanned 18 years, from the establishment of the plots in 2001 to 2019. In July 2001, we randomly installed five blocks of paired, fenced (unbrowsed), and unfenced (browsed) plots, each measuring 25 m × 25 m in the study area. The 2.20-m-high fence with a mesh size of 30 × 15 cm, excluded deer but not small herbivores (e.g., rabbits, hares, voles). These plots were located in areas with no shrubs at the time of their establishment, having been plowed in previous years for shrub clearing. At the beginning of the experiment, the mean tree density was 98 ± 6 trees/ha, with a mean diameter at breast height of 42.4 ± 1.4 cm and a mean tree canopy cover of 59.9 ± 3.3%. The trees were approximately 80 years old. We divided each plot into 40 subplots of 2 × 4 m, interspersed with paths 1 m in width, to conduct other studies within this long-term ungulate exclusion experiment. However, we only used data from four blocks (paired fenced and unfenced plots) in our analysis because, in July 2004 (i.e., 3 years after the beginning of the experiment), one unfenced plot was inadvertently plowed by the land manager and lost.

Vegetation measurements

To estimate the population density of C. ladanifer, we randomly chose 18 subplots measuring 2 × 4 m each within each browsed and unbrowsed plot. In March 2007, 2013, and 2015 and December 2019, we counted all standing live and dead C. ladanifer individuals in these subplots. Previous research (Lecomte et al., 2019) showed that sampling 18 subplots provided a reliable estimate of the shrub population density within the experimental plots. We categorized C. ladanifer individuals into two groups: adults, characterized by basal trunk diameters equal to or larger than 10 mm, and juveniles, with basal trunk diameters smaller than 10 mm, including seedlings and saplings. To estimate the aboveground biomass of C. ladanifer in 2015, for fire behavior modeling purposes, we adopted the following procedure. Thirty shrubs were randomly selected and cut to ground level in both browsed and unbrowsed plots. This involved selecting between six and 10 individuals per plot, ensuring they were distinct from the 18 subplots used for assessing shrub density.

Following the harvest, shrubs were separated into dead and live components and categorized by diameter class, an essential step for subsequent fire simulation, as detailed below. Subsequently, the separated components were oven-dried at 60°C until reaching constant mass, after which they were weighed. We estimated the aboveground biomass of C. ladanifer within each subplot using allometric equations. These equations identified trunk diameter as the best predictor of biomass as established in prior research (Lecomte et al., 2019).

In 2015, we also determined herb biomass by clipping all aboveground herb plants within four 50 × 25-cm quadrats randomly placed in each plot. A total of 16 quadrats per treatment were clipped. We determined herb biomass in early June 2015, coinciding with the peak of herb production in the study area. Simultaneously, we determined litter biomass, including leaves and branches from trees and shrubs, as well as dead herbs accumulated at soil surface from previous years. This was achieved by collecting all dead plant material within eight randomly located 50 × 25-cm quadrats in each plot, totaling 32 quadrats per treatment. The collected litter and herb samples were oven-dried at 60°C until reaching constant mass, after which they were weighed.

Fire behavior modeling

In 2015, we used the BehavePlus 5.0.5 fire modeling software to simulate potential fire behavior. This software incorporates wind speed, fuel characteristics, and slope as input variables (Andrews, 2014). Specifically, we used the SURFACE, CROWN, and SCORCH models. For the input data, the biomass of C. ladanifer (as previously described) was separated into dead and live components and categorized by diameter class, as required by BehavePlus. This categorization included live woody material (foliage and stems with diameter <0.64 cm) and the 1-h (diameter <0.64 cm), 10-h (diameter 0.64–2.54 cm), and 100-h (diameter >2.54 cm) time lag dead fuel classes (Andrews, 2009). These time-lag classes play a crucial role in estimating the expected rate of fuel moisture change, which significantly influences fire behavior. For example, the live woody class (diameter <0.64 cm) includes small twigs and leaves that can quickly dry out, becoming more flammable. Litter biomass was also categorized by the same diameter classes (i.e., <0.64 cm, 0.64–2.54 cm, >2.54 cm) as the dead and live plant biomass. Given that herbs are fully cured in summer, we added herb biomass to the 1-h class. Regarding the tree parameters in the CROWN model, mean canopy height, mean canopy base height, and mean canopy bulk density were 8.0 ± 0.2 m, 2.30 ± 0.09 m, and 0.11 ± 0.02 kg m−3, respectively (Lecomte et al., 2019).

The latter data were employed to simulate potential fire behavior using BehavePlus. To ensure realistic estimates of rate of fire spread, we corrected the fuel bed (average height of surface fuels) using the empirical and robust model developed by Anderson et al. (2015) specifically for shrublands. This model uses field vegetation height as input, alongside fuel moisture and wind conditions consistent with the BehavePlus simulation (detailed below). We then adjusted the fuel depth of the custom fuel models to align BehavePlus predictions of rate of spread with those derived from the Anderson et al. (2015) equation. This adjustment ensures a harmonized and reliable representation of fire behavior potential in our study area.

The fire behavior simulations assumed a 1-h fuel moisture content typical of extreme fire weather under Mediterranean summer conditions (e.g., Fernandes, 2009). To establish this condition, we utilized data from automatic weather stations in the region spanning 2001–2018. Employing the 97th percentile of the Fine Fuel Moisture Code (FFMC) of the Canadian Forest Fire Weather Index (FWI) system as a threshold, we obtained a 1-h moisture content of 4%. As commonly assumed (e.g., Andrews, 2009), we then estimated 10- and 100-h fuel moisture as 10-h = 1-h + 1 and 100-h = 1-h + 2, respectively. For live foliage, we assumed moisture contents of 80% (Chuvieco et al., 2009; Fernandes, 2009), representing a nondrought scenario. The simulations were conducted with a terrain slope of 5° (plot range slope varied between 1.4° and 7.5°, averaging 4.6° ± 0.5°) and an ambient temperature of 31°C. Midflame wind speed ranged from 0 to 30 km/h in 5-km/h increments, with the upper limit of 30 km/h representing extreme conditions. Output variables for analysis were fire surface rate of spread (in meters per minute), fireline intensity (in megawatts per meter [MW/m]), flame length (in meters), and occurrence of crown fire. Fireline intensity represents the quantity of energy released by the frontline of the wildfire. Crown fire occurrence is a function of surface fire-line intensity and is influenced by specific combinations of canopy base height and foliar moisture (Van Wagner, 1977).

Fire hazard in drought and nondrought scenarios

To quantify fire hazard under both drought and nondrought scenarios, we employed BehavePlus, as described earlier. For the drought scenario, we used the data collected during the drought year of 2015, which resulted from a cumulative precipitation deficit of 612.1 mm since the beginning of the experiment in 2001. Considering the relatively long life span and ecological resilience of C. ladanifer, particularly its ability to withstand drought conditions (e.g., Caldeira et al., 2015), we assumed for the nondrought scenario that all individuals recorded in 2015 remained alive. In modeling the drought scenario, we set the moisture content of the live foliage at 55%, compared to the 80% moisture assumed for the nondrought scenario (Chuvieco et al., 2009; Fernandes, 2009).

Statistical analyses

We conducted analyses on C. ladanifer population density, considering total and live number of individuals and the proportion of dead shrubs. Generalized linear mixed models (GLMMs) were employed with a negative binomial error distribution to avoid potential overdispersion by zero inflation. The response variable was the count of individuals within each subplot. Ungulate browsing, year (2007, 2013, 2015, and 2019), and their interaction were considered as fixed effects. Moreover, two random factors were included: block, to keep paired plots, and subplot nested within the plot, to account for the hierarchical structure of the experimental design. We did not add year as a random factor, as the among-year variation in population density was relevant for the tested hypotheses. When the interaction was statistically significant, we assessed multiple comparisons with the emmeans package (Lenth, 2018) with Bonferroni correction. For the analyses of dead shrubs and juveniles, we excluded data from the year 2007 as no records of dead shrubs and juveniles were observed in that year. We extended our analyses to examine the effects of browsing on C. ladanifer biomass (both overall biomass and live to dead ratios), as well as litter and herbs (only sampled in 2015). GLMMs were used with a negative binomial distribution error. In this case, the data were the mean of the observations within each plot, derived from the subplots. Ungulate browsing was considered a fixed effect, and block of paired plots was treated as a random effect. Furthermore, we assessed the effect of drought scenarios on fuel load classes for each browsing treatment while examining the interaction between drought scenarios and browsing treatment. When the interaction was statistically significant, we used multiple comparisons using the emmeans package, as previously described. GLMMs were fitted using the “glmmTMB” function within the GLMMTMB package (Brooks et al., 2017). Statistical significance of fixed factors was tested with Type II Wald chi-squared (ꭓ2) test. All statistical analyses were conducted using R version 4.0.2. (R Core Team, 2021).


Population density of C. ladanifer

Throughout the study period, browsing had a strong negative effect on the total adult population density of C. ladanifer, including both live plus dead individuals. The density was significantly lower in the browsed plots compared to the unbrowsed plots (ꭓ2 = 45.62, df = 1, p < 0.001) (Figure 3). Furthermore, there was a significant interaction between browsing and year (ꭓ2 = 21.98, df = 3, p < 0.001). Specifically, the population density of living C. ladanifer in the unbrowsed plots increased between 2007 and 2013, followed by a subsequent decrease. In contrast, within the browsed plots, the population density continuously decreased from 2007 to 2019.

Details are in the caption following the image
Density of live, dead, and total adult Cistus ladanifer shrubs in unbrowsed and browsed plots in 2007, 2013, 2015, and 2019 (mean ± SEM). *** indicates significant differences between browsing treatments (p < 0.001) for total shrub density.

Year had a significant and strong effect on the proportion of dead C. ladanifer shrubs (ꭓ2 = 14.82, df = 2, p < 0.001) in both browsed and unbrowsed plots. There was also a significant interaction between year and browsing on the proportion of dead shrubs (ꭓ2 = 59.35, df = 2, p < 0.001). This interaction suggests that drought conditions increased the number of dead shrubs in the browsed plots relative to the unbrowsed plots (ꭓ2 = 2.96, df = 1, p = 0.085). In 2007, with a precipitation deficit of only 11.3 mm, there were no dead shrubs in either browsed or unbrowsed plots. However, in 2013, with a cumulative precipitation deficit of 454.1 mm, the proportion of dead shrubs increased to 17% in unbrowsed plots and 73% in browsed plots. In 2015, with a cumulative precipitation deficit of 612.1 mm, the proportion of dead shrubs further rose to 57% in unbrowsed plots and 85%, in browsed plots. Finally, in 2019, with a cumulative precipitation deficit of 1054.2 mm, the proportion of dead shrubs approached 80% in unbrowsed plots and reached 100% in browsed plots (Figure 3).

The number of C. ladanifer juveniles varied significantly among years (Table 1) (ꭓ2 = 14.13, df = 2, p < 0.001). Browsing significantly reduced the number of C. ladanifer juveniles over time (ꭓ2 = 61.94, df = 1, p < 0.001). Specifically, the number of live juveniles was approximately 98% and 99% lower in 2013 and in 2015, respectively, in browsed plots compared to unbrowsed plots. In browsed plots, we found no standing dead juveniles in 2013, 2015, or 2019 or live juvenile shrubs in 2019. However, in unbrowsed plots, there were 67% dead juveniles in 2015 and 34% in 2019.

TABLE 1. Live and dead Cistus ladanifer juveniles (number of individuals per hectare) observed in 2013, 2015, and 2019 in unbrowsed and browsed plots (location 38°49′ N, 07°24′ W) (mean ± SEM).
Plot treatment 2013 2015 2019
Live Dead Live Dead Live Dead
Unbrowsed 10,625 ± 4540 0 2986 ± 659 6059 ± 3854 16,649 ± 5648 8559 ± 4247
Browsed 191 ± 77 0 87 ± 45 0 0 0

Aboveground biomass of C. ladanifer, herbs, and litter

In 2015, the total biomass of C. ladanifer was 24.56 ± 2.73 t/ha in the unbrowsed plots and 0.67 ± 0.13 t/ha in the browsed plots. Both dead and live biomass were significantly higher in the unbrowsed plots (dead: ꭓ2 = 19.23, df = 1, p < 0.001 and live: ꭓ2 = 10.20, df = 1, p < 0.001) (Figure 4).

Details are in the caption following the image
Cistus ladanifer dead and live plant biomass in 2015 in unbrowsed and browsed plots (mean ± SEM). Different letters indicate significant differences between treatments (p < 0.001).

In 2015, the live-to-dead biomass ratio of C. ladanifer also differed between browsed and unbrowsed plots (ꭓ2 = 4.28, df = 1, p < 0.05). In the unbrowsed plots, dead shrub biomass accounted for 46% of total biomass, resulting in a live-to-dead ratio of 1.17. However, in the browsed plots, dead biomass reached 83% of total biomass, leading to a live-to-dead ratio of 0.20. In 2015, there were no significant differences in herb biomass between browsed and unbrowsed plots (0.30 ± 0.04 t/ha vs. 0.40 ± 0.10 t/ha, ꭓ2 = 0.05, df = 1, p = 0.8). However, the biomass of litter was 50% higher in the unbrowsed plots compared to the browsed plots (5.67 ± 0.82 t/ha vs. 2.85 ± 0.53 t/ha, ꭓ2 = 3.72, df = 1, p = 0.052).

Fuel load characteristics

In 2015, C. ladanifer total fuel load was significantly higher in the unbrowsed plots than in the browsed plots, regardless of the drought scenario (ꭓ2 = 43.97, df = 1, p < 0.001) (Figure 5). Furthermore, the fuel load in the drought scenario was significantly higher than in the nondrought scenario (ꭓ2 = 5.54, df = 1, p < 0.05). Drought increased total fuel load by 58% in unbrowsed plots (10.75 ± 1.19 t/ha vs. 16.97 ± 2.06 t/ha, nondrought vs. drought scenarios, respectively). In contrast, the increase in total fuel was 34% in browsed plots (0.47 ± 0.09 t/ha vs. 0.63 ± 0.11 t/ha). Drought strongly modified C. ladanifer fuel load components. In both unbrowsed and browsed plots, drought significantly reduced the live woody fuel load component (ꭓ2 = 5.8, df = 1, p < 0.05), increased 1-h dead fuel load (ꭓ2 = 7.42, df = 1, p < 0.01) and 10-h dead fuel load (ꭓ2 = 10.69, df = 1, p < 0.001), and generated a previously nonexisting 100-h dead fuel load (ꭓ2 = 4398, df = 1, p < 0.001) (Figure 5).

Details are in the caption following the image
Fuel load of Cistus ladanifer categorized by size classes in unbrowsed and browsed plots, for both drought and nondrought scenarios in 2015: (a) and (c) fuel load without drought and (b) and (d) fuel load with drought (mean ± SEM). There were significant differences (p < 0.001) for each fuel component between the drought and nondrought scenarios in both unbrowsed and browsed plots. Drought significantly affected fuel load classes, generating 100-h dead fuel load in both treatments (1.665 ± 0.268 t/ha vs. 0.012 ± 0.002 t/ha in unbrowsed and browsed plots, respectively).

Fire behavior

The previously estimated fuel load characteristics strongly affected modeled fire behavior (Figure 6). In the unbrowsed plots, drought increased modeled fire surface rate of spread, fireline intensity, and fire flame length by 46.1%, 85.9%, and 33.5% (averaged across 15 and 30 km/h wind speed simulations), respectively (Figure 6). For unbrowsed plots, flame length exceeded the critical threshold of 1.3 m, even in no-wind conditions, for both drought scenarios. This threshold implies a transition from a surface fire to a severe tree crown fire that often kills adult oaks (Catry et al., 2012). In the browsed plots, drought increased fire surface rate of spread, fireline intensity, and fire flame length by 38.3%, 31.7%, and 17.1%, respectively (Figure 6). In the browsed plots, a wind speed just above 25 km/h would induce surface-to-crown fire transition in the nondrought scenario and a wind speed above 20 km/h in the drought scenario. On average, ungulate browsing decreased the fire surface rate of spread, fireline intensity, and fire flame length by 51.8%, 89.4%, and 64.2%, respectively, independently of the drought scenarios (Figure 6).

Details are in the caption following the image
Modeled (A) fire surface rate of spread (in meters per minute), (B) fireline intensity (in megawatts per meter [MW/m]), and (C) fire flame length (in meters) in drought and nondrought scenarios in unbrowsed and browsed plots. Modeling done for wind speed conditions of 15 and 30 km/h and terrain constant slope of 5°. The dotted line in (C) is the critical threshold of 1.3 m flame length, implying surface-to-tree crown fire transition.


Drought, shrub encroachment, and associated fire, along with rising ungulate populations, may interact to affect ecosystem structure and functioning. Browsing under drought conditions may increase shrub mortality, decrease the ratio of live to dead plant material, and increase plant flammability and fire hazard. Several studies have investigated the effects of herbivores, drought, and fire on the ecological dynamics of shrub encroachment in several ecosystems, namely, grasslands and savannas (e.g., Roques et al., 2001; Twidwell et al., 2014; Weber-Grullon et al., 2022). However, to our knowledge, no studies to date have assessed the interactive effects of herbivory, drought, and shrub encroachment on ecosystem wildfire hazard (e.g., Doblas-Miranda et al., 2017). In this study, we show that, in drought scenarios, ungulate browsing leads to higher shrub mortality and flammability, which, therefore, result in an increased ecosystem fire hazard. Our results further suggest that ungulate browsing may reduce shrub biomass and fuel load, offsetting the amplified effects of drought on flammability and fire hazard.

We show that the proportion of dead shrubs increases following drought, and browsing magnifies this effect. Several studies demonstrated high woody plant mortality following drought (Anderegg et al., 2019). For example, in a Juniperus-Quercus woodland in Texas, USA, 22% of the woody plants died after the cumulative effects of 11 years of flash droughts (i.e., rapidly intensifying drought characterized by moisture deficits and high temperatures) (Twidwell et al., 2014). In a ponderosa pine (Pinus ponderosa) and pinyon–juniper mixed woodland in northern Arizona, USA, one study recorded a 10% mortality for woody plants following a drought (Koepke et al., 2010). In our study site, within the unbrowsed plots, 10% of the C. ladanifer population was dead following the 2011/2012 drought, one of the severest droughts in the Iberian Peninsula in the last 50 years (Trigo et al., 2013). This drought exposed the shrubs to a cumulative precipitation deficit of 454.1 mm since 2001. By 2019, only 20% of the C. ladanifer population remained in the unbrowsed plots, reflecting the impact of seven recurrent droughts that increased the precipitation deficit to 1054.2 mm. While biotic factors, such as pests and diseases, could contribute to shrub mortality following drought (Stephenson et al., 2019), our observations revealed no signs of pests or diseases during the period of study. Moreover, C. ladanifer's insecticidal properties (Sosa et al., 2004) might have played a protective role against such threats. Drought years can significantly diminish the population density of woody plants, as shown in our site and by other authors elsewhere. For example, in the Santa Monica Mountains in California, USA, drought reduced shrub stand density by 63.4% within a mature mixed chaparral community (Venturas et al., 2016). This decline was probably induced by intraspecific competition for water, exacerbated by the high population density of the shrub stands (Roques et al., 2001). C. ladanifer is highly tolerant to drought and opportunistic in using short pulses of rainfall (e.g., shallow root system) (Ramírez et al., 2012). However, other studies (Caldeira et al., 2015; Haberstroh et al., 2021) showed, in the same region, that C. ladanifer transpiration was critically reduced after a severe drought and that shrubs were unable to recover immediately due to drought legacies, such as decreased leaf area. In our study, drought legacy effects, such as increased embolism or decreased leaf area (Peltier & Ogle, 2019), may have reduced the drought tolerance of C. ladanifer, contributing to the high proportion of dead shrubs observed following drought years.

Additionally, because drought decreased plant live-to-dead ratios and increased fuel loads and plant flammability, a higher fire hazard could be expected after drought events (Ruthrof et al., 2016; Stephens et al., 2018). Browsing, in addition to drought, further increased the proportion of dead shrubs in our site, as evidenced by other studies. For example, browsing by sheep, coupled with drought, had an additive effect on the mortality of the Eremophila matilandii shrub in Western Australia (Watson et al., 1997). Similarly, browsing by cattle resulted in only a 2% survival of shrubs (Atriplex vesicaria and Maireana astrotricha) after a drought event in Southern Australia (Read, 2004). As a consequence of the combined effects of browsing and drought, higher fuel load and fire hazard could be expected in our site. However, our long-term observations under high ungulate population density revealed a significant reduction in overall shrub biomass and fuel loads, countering the increased fire hazard induced by drought alone. Browsing pressure can modulate shrub cover and biomass (Archer et al., 2017), allowing for diverse management objectives. For example, in areas where management aims for high ungulate population densities for hunting purposes, as in our study area, this approach may contribute to reducing shrub biomass and fuel loads (e.g., Lecomte et al., 2019). Conversely, conservation and restoration efforts targeting the maintenance of higher shrub cover may require lower browsing pressure (e.g., Xu et al., 2023). Given the increased frequency of droughts, it becomes imperative to investigate the impacts of ungulates and drought on shrub mortality, particularly in areas where ungulate browsing may not significantly alter shrub biomass but can decrease live-to-dead plant ratios, increasing fuel flammability and fire hazard. To establish this, it is crucial to distinguish between the effects of total biomass and the effects of dead biomass on fire hazard. Thresholds in browsing pressure may occur, for which increased shrub mortality and flammability, resulting from the joint effects of drought and ungulates, combined with high shrub biomass, ultimately increasing fire hazard. These aspects warrant additional research, especially given the current context of global change.

Drought and ungulate browsing also affected the population density of shrub juveniles (saplings and seedlings), which can alter the ecosystem susceptibility to fire. The mortality of juveniles has the potential to increase fine fuel load, thereby impacting fire intensity and rate of spread (Anderson et al., 2015; Fernandes, 2001). However, in our study site, there was a limited contribution of juveniles to the total fuel load, accounting for less than 3% of the total biomass in the unbrowsed plots. In these plots, larger seed banks may have contributed to the higher C. ladanifer juvenile density, in contrast to the browsed plots, where flower and seed predation is prevalent (Lecomte et al., 2016, 2017). Live juveniles were scarce in the browsed plots, probably because of ungulate consumption (Bugalho & Milne, 2003), which also prevents shrubs from reaching the reproductive stage (Augustine et al., 2020; Scogings & Gowda, 2020). Long-term persistent browsing may have the potential to drive C. ladanifer to local extinction, potentially affecting tree regeneration and animal biodiversity (Bugalho, Lecomte, et al., 2011; Katona & Coetsee, 2019). Even in the unbrowsed plots, where there was a high density of juveniles of C. ladanifer, a significant proportion of juveniles were dead (67% and 34% in 2015 and 2019, respectively). The reduction of soil moisture during drought, coupled to a higher susceptibility of juveniles to drought (Caldeira et al., 2014; Lloret et al., 2009) exacerbated by density-dependent mortality (Lambers et al., 2002) may contribute to explaining the observed results. Additionally, herbivory by small mammals such as rabbits, hares, and voles not prevented by the fence mesh size from entering the plot could have contributed to the observed high mortality.

Our findings suggest that ungulate browsing can serve as a strategic management tool to reduce fuel load counteracting the potential effects of drought on shrub flammability and mitigating fire hazard. The ongoing global changes include interaction of extreme climate events with land-use changes and factors as land abandonment. The projected rise in the frequency of extreme droughts, which are occurring globally, and particularly in Mediterranean regions, alongside the expanding population of wild ungulates, coupled with shrub encroachment, will affect the severity of fire occurrence, ecosystem resilience, and the services they generate. Research on the combined effects of drought, shrub encroachment, and ungulate browsing on fire hazard is strongly needed to enhance our ability to predict and comprehend ecosystem responses to the complex dynamics of global change.


We are thankful to Fundação da Casa de Bragança for granting access to its estate Tapada Real de Vila Viçosa and to Alicia Horta, Joaquim Mendes, Lurdes Marçal, and Madalena Barreira for support during field and lab work. We are also very thankful to Elisabeth Huber-Sannwald and the anonymous referees who substantially improved the manuscript. We acknowledge funding through projects CERTFOR (PTDC/ASP-SIL/31253/2017) and DRESIL (2022.09115.PTDC). We also thank the Portuguese Science Foundation (FCT) for funding the research units CEABN-InBIO (Project UID/BIA/50027/2020 and POCI-01-0145-FEDER-006821), CEF (UIDB/00239/2020), and CITAB (UIDB/04033/2020), as well as financial support to XL (SFRH/BD/90753/2012), MNB (contract DL 57/2016/CP1382/CT0030), and FXC (SFRH/BPD/93373/2013).


    The authors declare no conflicts of interest.


    Data (Lecomte et al., 2022) are available in Zenodo at