Fire frequencies are changing in Neotropical savannas and forests as a result of forest fragmentation and increasing drought. Such changes in fire regime and climate are hypothesized to destabilize tropical carbon storage, but there has been little consideration of the widespread variability in tree fire tolerance strategies. To test how aboveground carbon stocks change with fire frequency and composition of plants with different fire tolerance strategies, we update the Ecosystem Demography model 2 (ED2) with (i) a fire survivorship module based on tree bark thickness (a key fire-tolerance trait across woody plants in savannas and forests), and (ii) plant functional types representative of trees in the region. With these updates, the model is better able to predict how fire frequency affects population demography and aboveground woody carbon. Simulations illustrate that the high survival rate of thick-barked, large trees reduces carbon losses with increasing fire frequency, with high investment in bark being particularly important in reducing losses in the wettest sites. Additionally, in landscapes that frequently burn, bark investment can broaden the range of climate and fire conditions under which savannas occur by reducing the range of conditions leading to either complete tree loss or complete grass loss. These results highlight that tropical vegetation dynamics depend not only on rainfall and changing fire frequencies but also on tree fire survival strategy. Further, our results indicate that fire survival strategy is fundamentally important in regulating tree size demography in ecosystems exposed to fire, which increases the preservation of aboveground carbon stocks and the coexistence of different plant functional groups.
Tropical savannas and forests are important components of the land carbon
sink (Pan et al., 2011; Liu et al., 2015; Ahlström et
al., 2015). However, their ability to continue sequestering carbon is
uncertain (Malhi et al., 2008), due in part to the impact of
projected increases in drought frequency and changes in fire regime on woody
carbon stocks (Brando et al., 2014). Globally, tropical forests,
savannas, and grasslands comprise
Fire is critical in defining the vegetation structure and distribution of tropical savannas and forests (Bond et al., 2005; Hoffmann et al., 2012a; Staver et al., 2011b). A positive feedback between flammable grass presence and tree fire mortality in open canopies has been identified as an important mechanism in maintaining savanna regions (Archibald et al., 2009; Hoffmann et al., 2012b; Staver et al., 2011a). Some trees can persist, despite high fire frequencies in many savannas, because they invest heavily in building thick bark (Pellegrini et al., 2017a). Thick bark insulates the xylem and phloem from fire damage, increasing the probability of tree fire survival (Brando et al., 2012; Hoffmann et al., 2012a), and potentially decreasing ecosystem carbon vulnerability to increasing fire frequency with global climate change (Pellegrini et al., 2016b). However, this increase in bark investment may come at a growth cost, as thicker-barked savanna species grow more slowly than thinner-barked forest species under similar growing conditions (Hoffmann et al., 2012a; Rossatto et al., 2009). Though many savanna trees have growth strategies that employ thick bark as a fire survival mechanism, species vary greatly in their investment in bark (Pausas, 2015; Pellegrini et al., 2017a; Rosell, 2016).
Climate, particularly precipitation, has the potential to interact with fire frequency and tree growth strategy (Brando et al., 2014). In locations with low precipitation, tree growth rates are much slower than in locations with high precipitation (Baker et al., 2003). Slower growth rates result in a population of smaller trees with relatively thinner bark than their larger counterparts, making their stems more vulnerable to fire. In addition to climate, physiology can play a role because some trees grow more rapidly than others due to differences in maximum photosynthetic capacity or specific leaf area (Rossatto et al., 2009). Tree growth rate is critical for determining the ability of trees to recover from fire, and consequently the fire frequency necessary to maintain savanna.
Simulating savanna vegetation dynamics is notoriously difficult. However, progress has recently been made in identifying some key mechanisms needed to stabilize savannas in dynamic global vegetation models (DGVMs) (Scheiter and Higgins, 2009; Lehsten et al., 2016; Baudena et al., 2015; Higgins et al., 2000; Haverd et al., 2013; Lasslop et al., 2016). Despite recent progress, DVGMs are still unable to fully capture global savanna extent as emergent features, generally overpredicting the extent of either grasslands or tropical forests (Lasslop et al., 2016). This makes it difficult to quantify the impacts of projected climate change on carbon storage in the tropics because both carbon storage capacity and the resistance of ecosystem carbon to changes in precipitation and fire regime vary across tropical biomes. Additional mechanisms observed to be important for maintaining tree–grass coexistence in empirical studies, such as variability in tree fire survival strategy (Hoffmann et al., 2012a), are underrepresented in models (Haverd et al., 2013; Lehsten et al., 2016). As a result, the implications of including variability in tree fire survival strategy are relatively untested, providing one useful avenue to improve modeling of savanna–forest dynamics.
To better understand the sensitivity of tropical carbon storage to changes in rainfall regime and fire frequency, we updated the Ecosystem Demography model 2 (ED2) to include distinct tropical savanna and forest plant functional types (PFTs), each with a different bark investment strategy. We evaluated our model's predictions against observations of savanna and forest tree growth rates, tree inventories, and total aboveground woody carbon (AGB) for different fire frequencies using field data from savannas and forests in the Cerrado region of Brazil. We then used the model to test the following two hypotheses: (1) including bark investment as a tree fire survival strategy decreases simulated carbon losses to increasing fire frequency, regardless of precipitation regime, due to the higher probability of survival for thicker-barked, larger trees, and (2) including bark thickness as a fire survival strategy expands the environmental conditions under which trees and grasses coexist by allowing for increased tree survival in frequently burned savannas, but also comes at a growth-rate cost that limits trees' ability to shade out grasses. Finally, we considered the effects of changing fire frequency and fire survival strategy on ecosystem composition and aboveground carbon vulnerability along a rainfall gradient in the Neotropics.
Our model simulations were carried out in a cohort-based terrestrial biosphere model, ED2, using parameterizations from Xu et al. (2016) unless otherwise specified. ED2 explicitly scales up tree-level competition for light, water, and nutrients to the ecosystem level (Medvigy et al., 2009; Medvigy and Moorcroft, 2012). The effects of water limitation on photosynthesis have been previously identified to be important for simulating savanna–grass dynamics (Baudena et al., 2015; Lasslop et al., 2016). Correspondingly, a novel aspect of the version of ED2 used in this study is the mechanistic representation of water-limited photosynthesis, whereby leaf and stem water potential are tracked and used to solve for root zone water uptake, transport of water vertically through the sapwood, and transpiration of water into the atmosphere. Variability in hydraulic traits such as turgor loss point, xylem water conductivity, and marginal water use efficiency determine PFT-specific responses to changes in leaf and stem water potential. Importantly, this mechanistic water limitation scheme has been demonstrated to better resolve vegetation dynamics in water limited tropical ecosystems (Xu et al., 2016), such as tropical savanna and forest regions.
We have incorporated in ED2 the following new processes important to
ecosystem fire resistance in the Neotropics: (1) a PFT-specific bark
investment strategy in two updated tropical PFTs; (2) a carbon tradeoff
between bark production and tree height, canopy area, sapwood area, rooting
depth, and leaf carbon; (3) a fire survivorship function dependent on
individual tree bark thickness; and (4) a dynamic feedback between tree
size, survivorship probability, and grass biomass availability. Updated
model codes are included as Supplement (ED2Model_Supplement_S1.tar.gz). The two new
PFTs represent a generic tropical forest tree PFT and tropical savanna tree
PFT. Both are based on the tropical brevideciduous PFT from
Xu et al. (2016). Previous PFT versions do not
include a bark thickness trait. The brevideciduous PFT was chosen because
its intermediate wood density and specific leaf area represent a drought
survival strategy incorporating both drought avoidance and resistance. This
intermediate strategy is utilized by a broad array of tropical tree species
(Xu et al., 2016). The savanna and forest PFTs
differ only in their bark investment strategy and the associated tradeoffs.
In the model, individual tree bark thickness is calculated according to the
following equation (Thonicke et al., 2010):
The carbon cost of a tree investing in bark is difficult to quantify. Here,
we incorporated a cost through the tree allometric relations. Bark turnover
rate in these systems is assumed to be negligible. Ordinarily in cohort- or
individual-based models, dbh is allometrically related to woody biomass,
leaf biomass, crown height, crown area, rooting depth, and sapwood area. The
model also uses a standard allometric relationship between dbh and woody
biomass. However, the new model relates the other derived properties to dbh
with bark excluded (denoted dbh'). Thus, for a given dbh, a
PFT with a large
The advantage of having thicker bark is incorporated through our fire
survivorship function. This function prescribes that trees with thicker bark
are more likely to survive a fire event than trees with thinner bark. Thus,
large trees and trees with a large
We conducted two classes of single-grid cell simulations. We first evaluated
our model performance against observed datasets within the Cerrado region of
Brazil at different fire frequencies. We then performed model experiments to
assess the influence of bark investment strategy on vegetation carbon and
tree–grass coexistence across a rainfall and fire gradient in the Cerrado.
Our model experiments included two sets of simulations. First, we ran a
control simulation that included C
Model evaluations at IBGE Ecological Reserve.
In our simulations, ecosystems were spun up from tree seedlings initialized
at a density of 1.0 seedling m
We evaluated performance of the updated model using ecosystem measurements
from study sites located in IBGE and the adjacent JBB Ecological Reserves
within the Cerrado region of Brazil at approximately 15.95
At the tree level, we compared annual diameter increments in the simulation
with bark and without fire disturbance to annual diameter increments of
twelve paired savanna and forest species measured over the years 2006–2007
(Rossatto et al., 2009). We performed a 35-year model spin-up from
seedlings in accordance with the disturbance history and age structure of
the site (Rossatto et al., 2009) and then examined the range of
average annual diameter increments of our savanna and forest PFTs over a
20-year period to the range of observed annual diameter increments. In our
simulations, the dbh size classes included in the calculation for the
savanna PFT ranged from 5 to 9 cm with a mean of
Next, we assessed the ability of the model to predict observed tree size
class distributions and measurements of AGB at sites with different fire
frequencies to evaluate if the updated model with bark could more accurately
capture ecosystem-level carbon dynamics and tree size abundance in response
to fire compared to the control model without bark. We compared model
simulated tree size distributions for trees with dbh
Schematic of model experiments along a rainfall gradient. Varied model inputs include fire frequency and precipitation regime (associated with a particular study site locations in the Cerrado region of South America). Different model versions were used to understand the effect of including bark as a fire survival strategy on model outputs of various ecosystem metrics including aboveground woody carbon, tree size class distribution, and tree crown area fraction.
To assess the outcome of including bark investment as a fire survival
strategy on carbon vulnerability and tree–grass coexistence, we conducted
experiments along sites with a rainfall gradient within the Cerrado. We
included locations with MAP within the first (MAP
Simulated growth rates for both the savanna and forest PFTs fell well within
the observed range for Cerrado species when the allometric growth tradeoff
with bark thickness was included (Fig. S1). The median growth rate for the
savanna PFT was within
Including a bark investment strategy in the model resulted in trees becoming more fire resistant with increasing size. As a result, for higher fire frequencies, simulations where bark was used to determine fire-driven mortality and growth tradeoffs predict a larger maximum tree size, which more closely reflects observed size class distribution and skew compared to the model without bark (Fig. S2). Further, at high fire frequencies, simulations with bark had the smallest percent difference in predictions of the median size class (23.0 % compared to 25.9 % for the simulation without bark) (Fig. S2a). At intermediate fire frequencies, simulations with bark also had a smaller percent difference in predictions of the median size class (27.0 % compared to 51.7 % for the simulation without bark) and better predict the skew in number count towards larger tree size classes (Fig. S2b). For low fire frequencies, including a bark investment strategy did not improve predictions of maximum tree size; however a better prediction for median tree size and skew towards larger trees was achieved (38.6 % compared to 44.1 % for the simulation without bark) (Fig. S2c).
Tree fire survival strategy impacts aboveground tree
biomass with fire disturbance. Observed and model-simulated aboveground
woody biomass (AGB) under
Size-specific survivorship affected predictions of ecosystem AGB under
different fire frequencies. At high fire frequencies, simulations with a
bark investment strategy captured the observed AGB within its predicted
range and had the lower percent error between simulated and observed mean
AGBs (an overestimation by 36.2 % compared to an underestimation by
38.8 % for the simulation without bark) (Fig. 2a). Under intermediate fire
frequencies, simulations with a bark investment strategy overestimated mean
AGB by 0.8 %, and the observed AGB was fully within the simulated
interquartile range, whereas the simulation without bark underestimated mean
AGB by 20.2 % and did not capture the observed AGB within the range of
predicted values (Fig. 2b). At low-frequency fire, the simulation without
bark predicts the observed AGB marginally more accurately than the model with
a bark investment strategy (percent errors of
Bark fire survival strategy increases the fraction of
aboveground woody biomass in large trees, particularly with frequent fire.
Model-simulated fraction of aboveground woody biomass (AGB) present in
different tree diameter at breast height (dbh in cm) size classes for
2
Bark fire survival strategy buffers aboveground woody
biomass loss with frequent fire, particularly at high MAP. Model-simulated
total aboveground woody carbon (AGB) at different MAP and forced fire
regimes for the model with bark
We found substantial differences in the fraction of AGB present in different tree size classes between the original model without bark and the updated model with a bark investment strategy, but these differences depended on fire frequency and precipitation. The impact of including bark thickness increased with MAP and fire frequency because the higher growth rates allowed for trees to grow larger faster, but the increased fire frequency restricted the growth of trees that did not invest in thick bark (Figs. 3, S3). When fire was frequent, the model without the bark investment strategy predicts a smaller absolute AGB and a large contribution of small- and intermediate-sized trees to total AGB, whereas the model version with bark investment strategy predicts a much higher absolute AGB with virtually all biomass allocated in the larger tree size classes (Figs. 3–4). These results were robust regardless of the cost of bark investment to tree growth (see Sect. 2.1) (Fig. S4). The greater proportion of biomass in large trees was due to the low probability of mortality during a fire because of the insulating capacity of bark and the relationship between tree size and bark thickness (Eq. 1). However, it should be noted that the biomass fraction of small trees is not illustrative of the frequency of small trees, as small tree biomass comprises a very small fraction of AGB when large trees are present. Thus, small trees grow between fires even in the simulations with a bark investment strategy, though they comprise very little of the absolute biomass. Indeed, the model with bark investment strategy predicts cumulative AGB mortality (resulting from fire, windthrow disturbance, age-related mortality, and carbon starvation) to occur almost exclusively in the small and large tree size classes, excluding midsized trees (Fig. S5). This absence of midsized trees mortality is indicative of an absence of midsized trees due to the high fire-driven mortality rate of small, low biomass trees. AGB mortality in large trees results from a low but persistent age-related mortality and windthrow disturbance. In contrast, trees of all size classes are present in the no bark simulation, and correspondingly the model predicts AGB mortality across all size classes, consistent with a large number of smaller trees, fewer intermediate- and large-sized trees, and an overall lower total AGB (Fig. 4, S5). Trends in AGB growth reflect almost identical biomass patterns. As would be expected, the difference in tree size distributions between the model without bark and the model with a bark investment strategy decreased substantially when fires were eliminated (Figs. 3b, d, f, S5b, d, f).
We also found that tree size distributions were largely unaffected by MAP in the presence of frequent fire in simulations with a bark investment strategy (Figs. 3a, c, e, S4); only minor impacts were found at intermediate fire frequency and high MAP (Fig. S3c). In contrast, the model without a bark investment strategy predicts that small size classes comprised a substantial fraction of AGB for high fire frequency simulations, particularly at low MAP (Fig. 3a, c, e).
Incorporating bark thickness decreased predicted carbon losses with
increasing fire frequency (Fig. 4) because larger, thick-barked trees made
up the majority of AGB (Fig. 3) and had a very low probability of mortality
during a fire. When a bark investment strategy was included, fire caused
almost no reduction in biomass at the wettest site (1660 mm yr
There was a strong interaction between precipitation, fire, and bark investment strategy. When no bark investment strategy was included, both fire frequency and precipitation exerted an equivalently strong control on total AGB, and the range in AGB after 100 years of growth increased substantially with increasing precipitation, but strongly depended on fire frequency (Fig. 4b). However, when bark investment was included as a fire survival strategy, MAP exerted a much stronger control than fire on the total AGB (Figs. 4a, S6a), indicating the important role of water availability regulating growth when species become fire resistant.
Bark investment can broaden the range of climate and fire
conditions under which savannas occur by reducing the range of conditions
leading to either complete tree loss or complete grass loss. Model-simulated
tree cover fraction present at different levels of mean annual precipitation
(in mm) for the model with bark
Both fire frequency and precipitation were important in maintaining
tree–grass coexistence and thus in controlling the distribution of
grasslands, savannas, and tropical forests (Fig. 5). At high fire
frequencies and low precipitation, we simulated grasslands with minimal tree
cover regardless of the model scheme (Figs. 5a, c, S7a, c). However,
simulations with bark investment as a fire survival strategy expanded
conditions under which tree crown area fraction ranged from 0.2 to 0.8. At fire
return intervals of 1–6 years and intermediate MAP (1150 mm yr
We documented that tree bark investment strategy interacts with precipitation and fire frequency to determine both (i) the stability of ecosystem carbon to fire and (ii) the coexistence of grasses and trees, illustrating that species traits, in addition to climate and fire, are critical for the stability of savanna and forest biomes. Bark investment strategy increased the stability of the carbon stock in large trees, which decreased ecosystem carbon losses with increased fire frequency under a range of precipitation conditions. Investment in bark was especially important in wetter savannas and forests, illustrating that the distribution of functional traits is fundamental to the resilience of wet forests to increased fire and changing rainfall regimes.
Our simulations illustrate that tree bark thickness as a fire survival trait substantially decreases fire-driven carbon losses, but the magnitude of the effect depends on precipitation regime (Fig. 4a–b). We found that bark investment was particularly important at reducing carbon losses at higher fire frequencies in locations with a high MAP (1660 mm). This is because trees had ample water availability, enabling smaller trees to grow rapidly and quickly become fire resistant. In contrast, the effect of bark investment offered little to no benefit in reducing carbon losses at the lowest MAP and highest fire frequency because small trees were not able to grow rapidly and accumulate thick enough bark to escape fire mortality. The decreasing utility of bark investment as a survival strategy at the lowest MAP and highest fire frequency likely occurred because bark investment slowed growth and prolonged the period during which trees were more susceptible to fire-driven mortality. Taken together, these results suggest that current models that do not account for bark investment strategies may underpredict Neotropical carbon resistance to fire in both savannas and forests. Such models would overpredict mortality of large thick-barked trees that make up a majority amount of aboveground woody biomass (Slik et al., 2013; Hanan et al., 2008). However, further work understanding the spatial distribution of tree species and their corresponding bark investment strategies is also critical (Pellegrini et al., 2017a; Rosell, 2016; Pausas, 2015; Dantas et al., 2013) because observations show that even large trees in rainforests have thin enough bark that they suffer substantial fire mortality (Uhl and Kauffman, 1990), although the forests with high mortality have precipitation values much higher than the maximum we consider (1660 mm).
Capturing savanna distributions globally has long been difficult for vegetation models, which overpredict the extent of either tropical forests or grasslands (Hickler et al., 2006; Cramer et al., 2001; Bonan et al., 2003; Hely et al., 2006; Schaphoff et al., 2006; Sato et al., 2007). A number of recent studies have focused on this issue: the adaptive dynamic global vegetation model (aDGVM) was able to capture savanna extent in Africa by including (i) trees with a higher fire mortality rate in small tree size classes, (ii) regenerative tree resprouting after fire events, and (iii) grass as super individuals (Scheiter and Higgins, 2009). The individual-based Populations-Order-Physiology model also included size-dependent tree mortality and was able to reproduce key vegetation structure and function along a rainfall and fire gradient in Australia (Haverd et al., 2013). Studies by Baudena et al. (2015) and Lasslop et al. (2016) have proposed several key mechanisms for capturing savannas in models: (1) water limitation on tree growth, (2) competition for water between grasses and trees, and (3) a grass–fire feedback.
The results from this study show a strong dependence of tree growth on precipitation and fire frequency, supporting observations (Pellegrini et al., 2017b; Higgins et al., 2007) and other modeling studies (Baudena et al., 2015). However, we found that including bark investment as a fire survival mechanism broadened the range of climate and fire conditions under which savannas occur by reducing the range of conditions leading to either complete tree loss or complete grass loss. This is due to the shift towards fewer, larger trees that store relatively more AGB per unit crown area than smaller trees, but are also resilient to frequent fire, resulting in ecosystems with intermediate crown cover in a wider range of precipitation and fire regimes. Thus, fire and precipitation as well as species-specific bark thickness traits have the potential to affect tree–grass coexistence, suggesting that inclusion of bark investment in models at broader spatial scales has the potential to substantially enhance our ability to accurately project changes in the tropical carbon sink with changes in fire and rainfall over the upcoming century.
Several avenues exist for future model improvement. Tree re-sprouting after fire has been shown to be essential in predicting the range of conditions for which tree–grass coexistence is possible (Higgins et al., 2000). Currently in our model, trees reproduce only through seedling recruitment. This has the potential to affect model benchmarking and simulation predictions because initial aboveground growth rates of re-sprouts are substantially higher compared to seedlings due to (a) access to belowground carbohydrate stores and (b) elimination of the need to allocate carbon for roots. Thus, re-sprouting may allow for better agreement between model predictions at higher growth rates and fire frequencies (Figs. 2, S1–2). Enabling tree re-sprouting may also stabilize savannas under frequent fire and low precipitation where all model versions currently simulate grassland (Fig. 5).
Additionally, the current fire model does not interact dynamically with climate or the nitrogen cycle. ED2 is capable of resolving nitrogen dynamics (Trugman et al., 2016) and we anticipate that coupling nitrogen and climate feedbacks to fire will be an important step in accurately modeling the carbon cycle. Despite these current limitations, the updated model accurately predicts growth, demographics, and AGB. We believe that these results can provide important insight into tree–grass coexistence and carbon resistance with changing fire frequency in the Neotropics with global change.
In conclusion, our results highlight that carbon storage in tropical savannas and forests depends not only on changing environmental drivers but also on tree fire survival strategy. Thus, we can improve projections of the tropical carbon sink with global climate change by increasing our understanding of the distribution of bark investment and incorporating this knowledge of bark investment into future vegetation models. Further, an increased understanding of the interaction between bark investment strategy and environmental drivers promises to increase our ability to project the distribution of savanna in regions previously simulated as grasslands or forests by reducing the range of conditions leading to either complete tree loss or complete grass loss.
Model codes are included in the Supplement.
Data presented here can be acquire from the corresponding author.
The supplement related to this article is available online at:
ATT, AFAP, DM, and WAH designed the research. ATT performed numerical experiments. ATT and AFAP analyzed the data. AFAP and WAH contributed field data. ATT drafted the paper and all authors contributed to writing of the manuscript.
The authors declare that they have no conflict of interest.
The authors gratefully acknowledge support from the Program in Latin American Studies at Princeton University (Adam F. A. Pellegrini), the Princeton Environmental Institute Walbridge Fund (Anna T. Trugman), the National Science Foundation Graduate Research Fellowship (Anna T. Trugman and Adam F. A. Pellegrini), the National Science Foundation Award 1151102. (David Medvigy) and DEB1354943 (William A. Hoffmann), and US Department of Energy, Office of Science, Office of Biological and Environmental Research, Terrestrial Ecosystem Science (TES) Program under award number DE-SC0014363 (David Medvigy). Edited by: Kirsten Thonicke Reviewed by: two anonymous referees