Tropical riparian forests in danger from large savanna wildfires

Abstract

1. Tropical savannas are known for the fire-prone ecosystems, yet, riparian evergreen forests are another important landscape feature. These forests usually remain safe from wildfires in the wet riparian zones. With global changes, large wildfires are now more frequent in savanna landscapes, exposing riparian forests to unprecedented impact. 2. In 2017, a large wildfire spread across the Chapada dos Veadeiros National Park, an iconic UNESCO site in central Brazil, raising concerns about its impact on the fire-sensitive ecosystems. By combining remote sensing analysis of Google Earth images (2003-2019) with detailed field information from 36 sites, we assessed wildfire impacts on riparian forests. For this, we measured the structure of trees, saplings and herbaceous plants, as well as topsoil variables. 3. Since 2003, all riparian forests had canopy cover above 90%, but after 2017, canopy cover dropped to 20% in some forests, indicating large variation in wildfire damage. A closer look in the field revealed that, on average, the wild-fire killed 52% of adult trees and 87% of tree saplings in flooded forests. In non-flooded forests, impacts on adult trees were negligible, but fire killed 75% of tree saplings. Opportunistic vines and the invasive grass Melinis minutiflora were already present in severely disturbed flooded forests. In all forests, impacts on many ecosystem variables were related to canopy damage, a variable measurable from satellite. Overall, seasonally flooded riparian forests were the most severely impacted, possibly due to the relatively thinner barks of their trees.

Full text

J Appl Ecol. 2020;00:1–12. wileyonlinelibrary.com/journal/jpe
|
1© 2020 British Ecological Society
Received: 24 July 2020
|
Accepted: 23 October 2020
DOI: 10.1111/1365-2664.13794
RESEARCH ARTICLE
Tropical riparian forests in danger from large savanna wildfires
Bernardo M. Flores1,2 | Michele de Sá Dechoum2,3 | Isabel B. Schmidt4 |
Marina Hirota1,2,5 | Anna Abrahão1,6 | Larissa Verona1 | Luísa L. F. Pecoral1 |
Marcio B. Cure2 | André L. Giles1,7 | Patrícia de Britto Costa1,8,9 |
Matheus B. Pamplona10 | Guilherme G. Mazzochini1 | Peter Groenendijk1 |
Géssica L. Minski2 | Gabriel Wolfsdorf1,7 | Alexandre B. Sampaio11 | Fernanda Piccolo1 |
Lorena Melo1 | Renato Fiacador de Lima3 | Rafael S. Oliveira1,8
1Depar tment of Plant Biology, Universit y of Campinas, C ampinas, Brazil; 2Gra duate Program in Ecology, Federal Un iversity of Santa Cat arina , Floria nópolis,
Brazil; 3D epar tment of Ecology and Zoology, Feder al Unive rsity of Santa C atar ina, Florianó polis, Brazil; 4D epar tment of Ecology, University of Brasília,
Brasí lia, Br azil; 5Department of Physics, Federal University of Santa Cat arina , Florianópolis, Bra zil; 6Institute of S oil Scie nce and L and Evalu ation, University
of Hohenheim, Stuttgart, G ermany; 7Grad uate Program in Ecology, Universit y of Campinas, C ampinas, Brazil; 8Scho ol of Biological Sciences, Universit y
of Western Australia, Per th, WA, Australia; 9Gra duate Program in P lant Biology, Universit y of Camp inas, C ampinas, Brazil; 10De part ment of Mat hematics,
University of E xeter, Exete r, UK and 11National Centre for Biodiver sity A ssess ment and Research and Conserv ation of the Brazi lian Cer rado, C hico Men des
Instit ute for Biologic al Conservat ion, Br asilia , Brazil
Correspondence
Bernardo M. Flo res
Email: mflores.bernardo@gmail.com
Funding information
Coordenação de Aperfeiçoamento de
Pessoal de Nível Superio r; Instituto
Serrapilheira, Grant/Award Number:
Serra-1709–18983; Fun dação Grupo
Boticá rio de Proteção à Naturez a, Gra nt/
Award Number: 1114-20181; Fundação de
Amparo à Pesquis a do Est ado de São Paulo,
Grant /Award Number: 18/01847-0 and
2016/25086-3
Handling Editor: Cécile Remy
Abstract
1. Tropical savannas are known for the fire-prone ecosystems, yet, riparian ever-
green forests are another important landscape feature. These forests usually
remain safe from wildfires in the wet riparian zones. With global changes, large
wildfires are now more frequent in savanna landscapes, exposing riparian forests
to unprecedented impact.
2. In 2017, a large wildfire spread across the Chapada dos Veadeiros National Park,
an iconic UNESCO site in central Brazil, raising concerns about its impact on the
fire-sensitive ecosystems. By combining remote sensing analysis of Google Earth
images (2003–2019) with detailed field information from 36 sites, we assessed
wildfire impacts on riparian forests. For this, we measured the structure of trees,
saplings and herbaceous plants, as well as topsoil variables.
3. Since 2003, all riparian forests had canopy cover above 90%, but after 2017,
canopy cover dropped to 20% in some forests, indicating large variation in
wildfire damage. A closer look in the field revealed that, on average, the wild-
fire killed 52% of adult trees and 87% of tree saplings in flooded forests. In
non-flooded forests, impacts on adult trees were negligible, but fire killed 75%
of tree saplings. Opportunistic vines and the invasive grass Melinis minutiflora
were already present in severely disturbed flooded forests. In all forests, im-
pacts on many ecosystem variables were related to canopy damage, a variable
measurable from satellite. Overall, seasonally flooded riparian forests were the
most severely impacted, possibly due to the relatively thinner barks of their
trees.

2
|
Journal of Applied Ecology
FLORES E t aL.
1 | INTRODUCTION
Globally, interactions between climate change and human activi-
ties are exposing tropical ecosystems to new stressing conditions
(Barlow et al., 2018). Arguably, tropical savanna landscapes are
among the most threatened, in part because of misinformed land
management. Agribusiness expansion over savannas is often asso-
ciated with habitat loss and the introduction of non-native grasses
(Strassburg et al., 2017; Veldman et al., 2015). These grasses disperse
from planted pastures and become invasive in well-preser ved sites,
where they often outcompete native species, altering ecosystem
functioning (Damasceno et al., 2018; Zenni et al., 2019). They pro-
duce more fuel biomass than native grasses, increasing the risk of
large wildfires (Damasceno et al., 2018; D'Antonio & Vitousek, 1992;
Fusco et al., 2019). In addition, fire exclusion practices are also in-
creasing landscape flammability (Schmidt et al., 2018; Veldman
et al., 2015). As a result, tropical savannas are losing resilience to
withstand climatic changes, with potentially negative ecological and
societal consequences (Bengtsson et al., 2019).
Tropical savannas are well-known for their biodiverse open
habitats dominated by fire-resistant plant species. However,
fire-sensitive evergreen forests are another important feature of
these landscapes, often restricted to wet riparian zones (Bueno
et al., 2018; Kellman & Meave, 1997; Natta et al., 2002; Pettit &
Naiman, 2007; Ribeiro & Walter, 2008). Combined, these ecosys-
tem mosaics provide numerous services for societies. Grasslands
and savannas provide forage for herbivores, medicine and wild food
for local peoples, and contribute to the recharge of underground
water reservoirs (Bengtsson et al., 2019). Riparian forests reduce
soil erosion, enhancing water quality and water security (Wantzen
et al., 2006). They may act as fire breaks, reducing the spread of
wildfires (van Nes et al., 2018). Vertebrate species also benefit from
the water and shelter provided by riparian forests, including top
predators (Redford & Fonseca, 1986). Therefore, by connecting for-
es t ha bit at s and pr ov iding vita l re sou rc es for keys to ne sp ec ies , ri par-
ian forests contribute to stabilize trophic networks, and enhance the
overall resilience of tropical savanna landscapes (Estes et al., 2011).
Although riparian forests are surrounded by fire-prone ecosystems,
their humid microclimate and wet soils usually suppress wildfires
(Hoffmann et al., 2012). As a result, most tree species are fire-
sensitive, with relatively thin barks, compared to trees in the open
savanna (Dantas & Pausas, 2013). However, when rare large wild-
fires spread during extreme drought events, they are more likely to
penetrate riparian forests with potentially negative impacts (Pet tit
& Naiman, 2007).
In recent decades, tropical savannas world-wide have been
experiencing a lengthening of the fire weather season (Jolly
et al., 2015). In tropical South America, the season is now 33 days
longer than 35 years ago, which implies a higher wildfire risk (Jolly
et al., 2015). In 2017, a delayed onset of the rainy season, coupled
with land conflicts, resulted in numerous wildfires throughout Brazil
(Fidelis et al., 2018). A particularly large one spread across the
iconic Chapada dos Veadeiros National Park (CVNP), burning not
only savannas but also the fire-sensitive riparian forests (Figure 1;
Figures S1 and S2; Text S1). Here, we combine satellite image anal-
ysis with detailed field assessments, to quantify the impact caused
by this large wildfire to riparian forests of the CVNP. First, to con-
firm the actual timing of the wildfire and assess its damage to for-
est cover, we used ver y high spatial resolution Google Earth images
(2003 through 2019, ~0.5 m), and produced a time series of canopy
cover change for 16 riparian forest fragments. We then tested if
wildfire damage to the canopy, measured from satellite, was a good
predictor of impac ts on other ecosystem variables. Using field data
from 36 sites randomly spread across the study area (Figure 1), we
measured the structure of adult trees and saplings, the cover of her-
baceous plants, and several topsoil variables. We expected to find
that canopy damage was related to tree size and bark thickness dis-
tributions; traits previously shown to influence tree mortality (Balch
et al., 2011; Cochrane, 2003). Moreover, we hypothesized that in the
open burnt sites, invasive grasses and opportunistic plants would be
expanding, and that topsoils would be changing due to ash deposi-
tion and erosion processes.
4. Synthesis and applications. Our findings reveal how riparian forests embedded in
tropical savanna landscapes are in danger from large wildfires. The destruction of
some forests has opened space for new plant species that may propel a shift to
an alternative ecosystem state. Riparian forests are habitat of large savanna ani-
mals and their loss could affect entire trophic networks. Managing wildfires and
invasive grasses locally is probably the best strategy to maintain riparian forests
resilient. As wildfire regimes intensify in tropical savanna landscapes, our findings
stress the need for an integrated management that considers riparian forests as a
vulnerable element of the system.
KEYWORDS
Cerrado, climate change, drought, global change, invasive grasses, resilience, resistance,
tropical ecosystems

|
3
Journal of Applied Ecolog
y
FLORES E t aL.
2 | MATERIALS AND METHODS
2.1 | Study area
We studied flooded and non-flooded riparian forest ecosystems
at the CVNP, Br azil (Figure 1). The la nd sc ape is formed by mos aics
of different vegetation types (Ribeiro & Walter, 2008). Wet and
dry grasslands and savannas occur in between streams, covering
most of the landscape. At the northwest edge of the park, dry
deciduous forests are found, whereas at the southwest edge, ri-
parian evergreen forests are most common (Figure 1a). Riparian
forests are closed-canopy ecosystems, which largely contrast to
the open palm swamp savannas, locally known as veredas. Along
riparian zones, the vegetation alters abruptly between closed-
canopy flooded forest and open palm swamp savannas, where na-
tive grasses coexist with the large monodominant palm Mauritia
flexuosa. Some riparian forests in our study area can be season-
ally flooded by streams in the wet season, whereas others remain
above the water level throughout the year (hereafter flooded and
non-flooded forests). The region receives a mean annual rainfall
of 1,500 mm, has a mean dry season of 130 days, and a mean
temperature of 21°C (Oliveira & Marquis, 2002). The park is sur-
rounded by rural areas with pastures as the main activity, and
small touristic towns.
FIGURE 1 The 2017 wildfire at the Chapada dos Veadeiros National Park. (a) Map of the study region, in Central Brazil, showing
in green forested areas, in grey the area affected by the wildfire, and in light green circles, our field study sites. (b) Photos show: (left)
savanna landscape with riparian forests and palm swamp savannas along streams; (centre) fires at night, burning riparian ecosystems; (right)
landscape burnt by wildfire, except for a fire break. Photo credits to B.M. Flores (lef t) and F. Tatagiba (ICMBio) (centre and right)
(a)
Alto
Paraiso
(b)

4
|
Journal of Applied Ecology
FLORES E t aL.
2.2 | Google Earth image analyses
To obtain a first overview of the actual damage caused by the 2017
wildfire on riparian forests of the CVNP, we analysed time series
of canopy cover produced with Google Earth images (Image data:
©2020 CNES/Airbus & Maxar Technologies) of ver y high spatial res-
olution (~0.5 m), freely available for different years in Google Earth
Pro software v 7.3.3.7699. Images prior to 2013 had 1.5 m spatial
resolution. We included images from 2003, 2014, 2015, 2016, 2017
and 2019, which had good quality (few clouds) for visual inspection
and detection of canopy cover. To quantify canopy cover, we first
used the image from 2014 to create polygons delimiting the contours
of 16 riparian forest fragments within the study area (Figure 1). We
classified each fragment as flooded or non-flooded, according to the
micro-she d to which they belonged (Table S1). Among all riparian for-
ests in the study area, 54% is seasonally flooded, whereas the other
46% does not flood in the wet season. We then added a spatial grid
of points, spaced by 0.0001 degrees (~11 m), totalling 7,425 sample
points within 90 ha of riparian forest (Figure S3). For each sample
point, in each year, we visually classified as covered by forest (1) or
non-forest (0), from a viewpoint of 1.5 km in height. We classified
areas with bare ground, shrub or herbaceous vegetation as non-
forest. For each year, we calculated canopy cover as the proportion of
points inside eac h fra gm ent that were covered by fo re st. This met ho d
has proven effective for detecting disturbances in the Amazon forest
(Flores et al., 2014), which was confirmed at the CVNP. We analysed
ca no py cover cha ng e over the 16-y ea r per io d us in g a ‘l oe ss’ smoot he r
with the ‘ggplot2’ package (Wickham, 2016), in R software (R Core
Team, 2019).
2.3 | Field sampling
Six months after the wildfire, in April 2018, we sampled 36 field
plots of 20 m × 10 m (0.02 ha) in riparian forests spread across
five micro-sheds (streams) within the CVNP burnt area (Figure 1;
Table S1; Figure S3a). With Landsat 8 (OLI) imagery from 29
January 2017, scene 221-070, using false colour composites
with the bands 7 (SWIR-2), 5 (NIR) e 4 (red) and the channels red,
green and blue, we first identified 82 micro-sheds that could be
used in the study. Among those possibilities, we chose five mi-
cro-sheds based on two criteria: (a) that they were embedded in
the landscape affected by the wildfire and (b) their accessibility
(Figure 1a). The plots were spaced by at least 30 m within for-
est fragments in each micro-shed, avoiding stream channels and
edges. Riparian forests in four of the five streams are seasonally
flooded (flooded forests). In contrast, forests along one stream
are not flooded because the river channel is at least 2 m deep,
keepin g the fore st soil we ll-dr ai ned even in th e rainy season (n on-
flooded forests). Among the 36 study sites, six were located in
unburnt forests, including one in the non-flooded area and five in
flooded areas. These sites were considered our ground reference
for the pre-fire st ate. The other 30 study sites were located in
forests burnt by the 2017 wildfire, including eight in non-flooded
forests, and 22 in flooded forests. We classified forests as burnt
or unburnt in the field using fire signs, such as charred trees and
ashes in the super ficial soil. We also confirmed our classification
with Google Earth , using three sampl e point s for each field si te , to
analyse their canopy cover change in time series.
In each 20 m × 10 m field plot, we measured the diameter at
breast height (1.3 m, DBH) of all trees ≥10 cm. We also measured
tree bark thickness using a bark corer. Because bark thickness in-
creases with tree age, we worked with relative thickness, calcu-
lated as 100 × (thickness/DBH), according to Lawes et al. (2013).
Moreover, in three evenly spaced positions along the 20 m central
line of each plot, we dug trenches to measure root mat depth, in-
cluding fine roots and hummus. In these same trenches, beneath
the root mat, we collected superficial soil samples (0–20 cm) that
formed one single compound sample per plot. These soil samples
were analysed for tex ture and available (exchangeable) nutrients at
the Soil Dep ar tment Lab oratory at the Fed eral Un ive rsity of Viçosa,
Brazil (see Text S2 for details).
For each 20 m × 10 m field plot, we fixed four subplots of
1 m × 1 m in each corner, in which we sampled tree saplings, na-
tive herbaceous cover, native vine cover, non-native (exotic) grass
cover, bare soil cover and canopy cover. All measures taken from
the four subplots were averaged to produce a single value per plot.
In each subplot, we measured the density of woody saplings with
DBH between 1 and 5 cm. We visually estimated the cover of her-
baceous plants, vines and non-native grasses within one of seven
classes; class-0 for 0% cover, class-1 for cover 0%–5%, class-2 for
cover 5%–25%, class-3 for cover 25%–50%, class-4 for cover 50%–
75%, class-5 for cover 75%–95% and class-6 for cover 95%–100%.
We also visually estimated bare soil cover from 0% to 100% in each
subplot, considering bare soil as the absence of root mat, litter or
living plants. We estimated canopy cover using a Lemmon Spherical
Concave Densiometer, taking measures to the north, south, east and
west from each subplot, at 1.2 m above the ground. We obtained a
total of 16 canopy cover measures per 20 m × 10 m field plot, which
we averaged to obtain a single estimate.
2.4 | Field data analyses
Our analyses involved three main steps: (a) detect canopy cover
changes over 16 fragments (~90 ha) of riparian forest; (b) test if fire
damage to the canopy, measured with satellite and in the field, was
a good predictor of other impacts on vegetation and soil variables
measured in the field; (c) test if fire damage could be explained by
tree diameter and bark thickness.
Step 1 involved a landscape-scale analysis to understand the
magnitude of the 2017 wildfire. This approach also allowed us to
extrapolate our field observations. We validated our remote sens-
ing analysis by comparing canopy openness measured with Google
Ear th (image of 2019) with canopy ope nne ss measured in th e field
(in 2018), in both cases af ter the wildfire. Satellite estimates were

|
5
Journal of Applied Ecolog
y
FLORES E t aL.
based on observations from the three sample points nearest to
each field plot (see Figure S3). Field estimates of canopy open-
ness were based on 16 observations per plot. We used a Pearson
correlation analysis to compare both estimates of mean canopy
openness.
For step 2, first we quantified mean fire impacts on the vegeta-
tion and topsoil in field sites. For this, we first divided values found
in burnt forests, by the mean value of unburnt reference forests:
(1 − ‘burnt-site’/‘reference-mean’) × 100; (in the cas e of non -flooded
forests, we used the values from the single reference site). For each
variable, based on the impact observed in burnt sites, we estimated
means and 95% confidence intervals. We then assessed in detail
whether canopy openness could be used as proxy for other eco-
system impacts. This hypothesis is consistent with previous stud-
ies, showing that severe fires cause high tree mortality in tropical
forests, increasing canopy openness and consequently the risk of
reburning, grass invasion and topsoil erosion (Brando et al., 2014;
Cochrane, 2003; Flores, Staal, et al., 2020). We related ‘canopy
openness’ with tree basal area, sapling density, native and non-na-
tive herbaceous cover, vine cover, bare soil cover, root mat depth,
as well as soil texture and nutrient availability. We analysed our
data using linear mixed models (LMMs) with the r package ‘lme4’,
function ‘lmer’ (Bates et al., 2015). For each response variable, we
used ‘canopy openness’ as a fixed factor, and ‘micro-shed’ as a ran-
dom factor, to control for landscape heterogeneity (Table S1). As
a complementary analysis, we compared fire impacts between
burnt and unburnt flooded and non-flooded fores ts, again using the
function ‘lmer’ (Bates et al., 2015), with ‘forest type’ as fixed factor
and ‘micro-shed’ as a random factor. We visually analysed the re-
sidual-plots from each model to check for normality, and log-trans-
formed the variables: ‘sapling density’, ‘native vine cover’ and ‘P
concentration’, to approach normal distribution. For all analyses, we
tested for spa ti al autocorrel at io n us in g the fun ction ‘cor re log’, in the
r package ‘ncf’ (Bjornst ad, 2020), with distance classes of 0.005 de-
grees (~550 m). We did not find any spatial autocorrelation among
our field study sites in any of the analyses (Figures S4 and S5), and
thus assumed that all field sites were independent replicates. Due
to an imbalance in the number of replicates for flooded and non-
flooded forests (Table S1), we compared burnt forests of the single
non-flood ed micro-she d, with bur nt fore st s gr ouped in fo ur floode d
micro-sheds. For this group comparison, we used the nonparametric
Mann–Whitney (Wilcoxon) test, with the function ‘wilcox.test’ in R
software (R Core Team, 2019), which uses ranked data and hence is
rather robust to imbalanced sample sizes.
To understand whether tree diameter and bark thickness
could explain differences in wildfire damage between flooded
and non-flooded forests, we analysed the density distributions
of DBH and relative bark thickness for all trees in each forest
type, using data from unburnt forest reference sites only. We
compared both forest types using a Mann–Whitney test with
the function ‘wilcox.test’ in the R sof tware. Both traits have
been shown to influence fire resistance (Balch et al., 2011;
Cochrane, 2003).
3 | RESULTS
3.1 | Wildfire damage to the canopy from satellite
Our analysis of canopy cover change in riparian forests at the CVNP,
using Google Earth imagery, suggest s that at least since 2003, can-
opy cover in all 16 fores t fragment s remain ed high above 90%. After
2017, however, canopy cover decreased in many forests, reaching
20% in some cases, and showing that wildfire impac ts varied from
mild to highly destructive. Wildfire damage to the canopy varied ac-
cording to the micro-shed where forests are located (Figure S6), and
also to the local flooding conditions (Figure 2). We found a striking
difference between flooded and non-flooded forests, with the first
appearing to be more fire-sensitive (Figure 2). We validated our sat-
ellite estimates in the field, and found that post-fire canopy openness
measured with Google Earth and in field plots were strongly corre-
lated (r = 0.80; Figure 2). Moreover, the analysis of canopy cover in
the 16 forest fragments, before and after the wildfire, demonstrated
how the damage was unrelated with pre-wildfire levels (Figure S7).
3.2 | Wildfire impacts assessed in the field
Field assessments show that (Table 1), on average, in flooded for-
ests, the wildfire decreased canopy cover by 38 (±13)%, killed 52
FIGURE 2 Temporal changes in canopy cover on 16 forest
fragments in the study area (~90 ha), from 2003 through 2019,
derived from Google Earth images. Before 2017, canopy cover was
generally high, above 90%, but after the 2017 wildfire (vertical
grey dashed line), canopy cover became variable, revealing a
gradient of fire damage. We applied a small jitter to the horizontal
(year) axis to reduce data overlap. Inner plot shows the correlation
between canopy cover measured in the field in 2018 and canopy
cover measured with Google Earth in 2019. For details on sampling
method, see Figure S3. Satellite image credits to Google Earth
(Image data: ©2020 CNES/Airbus & Maxar Technologies)
0.00
0.25
0.50
0.75
1.00
2005 2010 2015 2020
Canopy cover
Forest type
Flooded
Non−flooded
Sample points (n)
0
500
1,000
1,500
2,000
03366 100
020406080 100
r = 0.80
Canop
y
openness in field
(
%
)
Canopy openness
in Google Earth (%)
– Fire damage +
2017
Field vs. Google Earth (after fire)
Year

6
|
Journal of Applied Ecology
FLORES E t aL.
(±15)% of the adult trees and 87 (±9)% of tree saplings, causing a
47 (±13)% decrease in tree basal area, while native herbaceous and
vine cover had increased by 20%–25% and <5% respec tively. On
average, invasive grasses increased <5%, but were already present
in t wo severely burnt flooded forest s. Bare soil cover increased by
39 (±14)%, and root mat depth decreased by 65 (±18)% . Soil clay
decreased by 25 (±9)%, and phosphorus availability increased by
18-fold. In contrast, in non-flooded forests, the wildfire did not sig-
nificantly alter canopy cover, tree density, invasive grass cover, root
mat depth, bare soil cover and soil clay. However, it killed 75 (±26)%
of tree saplings, increased native herbaceous and vine covers by
20%–25%, and slightly decreased phosphorus concentration. Burnt
non-flooded forests also had more tree basal area than the single
reference forest. These patterns and the values obser ved in refer-
ence forests also show how, for most ecosystem variables, wildfire
impacts were severer on flooded forests, compared to non-flooded
forests (Figure S8), regardless of which micro-shed they belonged
to (Table S2).
When then related ‘canopy openness’ to other ecosystem vari-
ables (Figure 3), and found that wildfire damage to the canopy sig-
nificantly predicted reductions in tree basal area (Figure 3a) and
root mat depth (Figure 3f), as well as increases in bare soil cover
(Figure 3e) and soil available phosphorus (Figure 3h). Damage to
the canopy was also significantly related to an increase in soil pH
and decrease in soil aluminium (Figure S9). We found a near-signifi-
cant (p = 0.06) reduction of sapling density (Figure 3b). Native vine
cover, non-native grass cover and clay fraction did not change with
fire damage (Figure 3). We could not estimate changes in non-na-
tive grass cover because most sites had zero cover, although we
found that two burnt sites were already colonized by the African
grass species Melinis minutiflora (<5% cover in Figure 3d). In gen-
eral, we found the strongest wildfire damages in flooded forests,
TABLE 1 Fire impacts on vegetation and topsoil variables
of flooded and non-flooded riparian forests. We show per cent
changes (mean ± CI), relative to reference sites. Red downward
facing triangles indicate reduction. Blue upward facing triangles
indicate increase
Ecosystem variable
Change (%) af ter the wildfire
Flooded forests
Non-flooded
forests
Vegetation
Canopy cover −38 ± 13 −4 ± 2
Tree density −52 ± 15 +4 ± 17
Tree basal area −47 ± 13 +83 ± 53
Sapling density −87 ± 9 −75 ± 26
Native herbaceous cover +20–25 +20–45
Native vine cover +0–5 +20 –25
Non-native grass cover +0–5 0
Topsoil
Root mat depth −65 ± 18 +4 ± 11
Bare soil cover +39 ± 14 0
Clay fraction −25 ± 9 −17 ± 16
Silt fraction +63 ± 19 −17 ± 20
Sand fraction −10 ± 22 +45 ± 45
P concentration +1,808 ± 473 +36 ± 20
Ca2+ concentration −6 ± 40 −59 ± 30
pH +8 ± 6 −4 ± 4
Sum of bases −2 ± 34 −54 ± 27
Note: Fire impacts were estimated by dividing each burnt forest value
(N = 22 for flooded and N = 8 for non-flooded) by the forest reference
mean (N = 5 for flooded and N = 1 for non-flooded). Confidence
Intervals (CI) are based on alpha = 0.05. The cover of herbaceous, vines
and grasses changed between classes on a sc ale from 0 to 6, with 0 as
0% cover, 1 as 0%–5% cover, 2 as 5%–25% cover, 3 as 25%–50% cover,
4 as 50%–75% cover, 5 as 75%–95% cover and 6 as 95%–100% cover.
FIGURE 3 Wildfire impacts on riparian forests of the Chapada
dos Veadeiros National Park. We used data on canopy openness
as a proxy for fire damage. We then related canopy openness to
vegetation and soil variables: (a) tree basal area, (b) tree sapling
density, (c) native vine cover, (d) non-native grass cover, (e) bare soil
cover, (f) root mat depth, (g) clay fraction and (h) topsoil phosphorus
concentration. Field data were collected 6 months after the
wildfire. The vertical grey dashed line indicates the mean canopy
openness of reference sites. Statistically significant effects (based
on LMM) are shown above each plot. The cover of plants in (c) and
(d) varied on a scale from 0 to 6, with 0 as 0% cover, 1 as 0%–5%
cover, 2 as 5%–25% cover, 3 as 25%–50% cover, 4 as 50%–75%
cover, 5 as 75%–95% cover and 6 as 95%–10 0% cover
020406080
0
10
20
30
40
50
60
Tree basal area (m2/ha)
●
●
●
●
●
●
●
●
p = 0.005
020406
08
0
Tree saplings per m2
p = 0.06
●●●
●
●●
●
●
0123
020406080
0
1
2
3
4
5
6
Native vine cover
p = 0.81
●
●
●
●
●
●
●
●
020406
08
0
0.0
0.2
0.4
0.6
0.8
Non−native grass cover
●●●●●●●●
p not estimated
020406080
0
20
40
60
80
Bare soil cover (%)
●●●●●●●●
p < 0.001
020406
08
0
0
5
10
15
Root mat depth (cm)
●
●
●
●
●
●
●
●
p < 0.001
020406080
20
30
40
50
Clay fraction (%)
●
●
●
●
●
●
●
●
p = 0.58
020406
08
0
0
20
40
60
80
100
P available (mg/kg)
p < 0.001
●
●
●
●
●
●
●
●
●
●
Flooded burnt
Non−flooded burnt
Flooded reference
Non−flooded reference
Canopy openness (%)
Fire damage
(a) (b)
(c) (d)
(e) (f)
(g) (h)

|
7
Journal of Applied Ecolog
y
FLORES E t aL.
whereas non-flooded forests suffered mild damages to most eco-
system variables (Figure 3). Further analyses indicated that the cover
of native vines and non-native (exotic) grasses did not change with
topsoil phosphorus (P) concentration, neither with bare soil cover
(Figure S10). Both sites where non-native grasses were present,
however, had high P concentration.
Our analyses of density distributions of tree DBH and relative
bark thickness indicated that, although trees in both forest types
have similar DBH values (Figure 4a; p = 0.53), trees in the non-
flooded forests have relatively thicker barks, compared to trees in
flooded forest s (Figure 4b; p = 0.002).
4 | DISCUSSION
4.1 | Wildfire impacts on riparian forests
Our result s reveal how large wildfires can be destructive for ripar-
ian forests embedded in tropical savanna landscapes. In riparian
forests of the CVNP, the 2017 wildfire killed on average half of all
adult trees and 88% of tree saplings. It consumed half of topsoil root
mats, exposing 28% of bare soils, which boosted phosphorus avail-
ability, probably as a result of ash deposition. These impacts were
stronger on seasonally flooded forests than on non-flooded forests,
with some sites suffering complete mortality and topsoil combustion
(Figures 3 and 5a,b). In severely burnt forests, favourable light and
nutrient conditions may have facilitated the expansion of opportun-
istic plants, such as vines and ferns, which already covered par ts of
our field plots only 6 months after the wildfire (Figure 5d,e). In two
of these sites, invasive grasses were already present at low covers.
Overall, our findings confirm our hypothesis that riparian forests are
fire-sensitive ecosystems, and indicate that seasonally flooded for-
ests, which represent 54% of all riparian forests in this landscape,
are the most vulnerable to wildfires. Although our field assessments
are imbalanced with relatively more sites representing flooded for-
ests, our analyses were robust in suggesting their higher sensitivit y,
compared to non-flooded forests (Table S2).
A similar pattern has been obser ved in the Amazon, where sea-
sonally flooded forests are more sensitive to wildfires than non-
flooded forests (Flores et al., 2014; Nogueira et al., 2019; Resende
et al., 2014). Flooding conditions are known to cause hypoxia, and
trees often invest in above-ground root systems to overcome this
stress (Parolin et al., 2004). Slow litter decomposition in these sys-
tems also causes humus to accumulate in the topsoil (dos Santos
& Nelson, 2013). Root mats retain soil humus, reducing nutrient
leaching and erosion (Stark & Jordan, 1978), while enhancing oxy-
gen acquisition under water (Parolin et al., 2004). Yet, during the dry
season of extremely dry years, root mats may act as fuel for deadly
smouldering wildfires (Flores et al., 2014; Resende et al., 2014). In
most tropical forests, a single wildfire usually kills between 23% and
44% of the trees (Cochrane, 20 03). In floodplain forests, however,
one single wildfire event can kill 60%–10 0% of all adult trees, which
places them among the most fire-sensitive tropical forests (Flores
et al., 2014, 2016; Resende et al., 2014). Our findings at the CVNP
reveal that, while the non-flooded forests suffered negligible tree
mortality, flooded forests suffered on average 52% tree mor tality
from a single wildfire event.
Root mats may have contributed to increase forest flammabilit y
during the 2017 drought. However, non-flooded forests also have
root mats (Figure 3f), which raises the question of why they only
suffered mild damage. Tree size and bark thickness are traits known
to enhance internal protection from fire damage in tropical forest
trees (Balch et al., 2011; Cochrane, 2003). We found that, while both
forest types had similar DBH distributions, trees in the non-flooded
forest had higher bark thickness, compared to trees in the flooded
forest (Figure 4). This pattern suggests that both forest types may
FIGURE 4 Comparing (a) DBH and (b) relative bark thickness of individual trees in unburnt flooded and non-flooded forest s. The Mann–
Whitney test confirmed that both forests did not differ in terms of DBH (p = 0.53), but were significantly dif ferent in terms of relative bark
thickness (p = 0.0 02). In flooded forests, we show 166 individual trees from five distinct reference sites, whereas in non-flooded forests, we
show 12 trees from the single non-flooded forest site. Trees in non-flooded forests have higher relative bark thickness than trees in flooded
forests, which may help explain the differences in wildfire severity
0.00
0.01
0.02
0.03
0.04
0.05
20 40 60
Diameter at breast height (cm)
Density
FOREST TYPE
Flooded unburnt (n = 166)
Non−flooded unburnt (n = 12)
0.00
0.03
0.06
0.09
01020
Relative bark thickness
(a) (b)

8
|
Journal of Applied Ecology
FLORES E t aL.
FIGURE 5 Riparian forests of the Chapada dos Veadeiros National Park af ter the 2017 wildfire. (a, b) Two burnt flooded forest s with
forest structure and organic soils severely disturbed. (c) Burnt flooded forest with low tree mortalit y. (d) Severely burnt flooded forest
dominated by the opportunistic fern Pteridium arachnoideum. (e) Severely burnt flooded forest dominated by vines. (f–h) Three unburnt
flooded forest s

|
9
Journal of Applied Ecolog
y
FLORES E t aL.
have experienced different fire regimes in the past (Pellegrini
et al., 2017). Another possibilit y is that the thicker barks of trees in
non-flooded forests are an adaptation to reduce water loss during
the dry season (Loram-Lourenco et al., 2020), which contributed to
increase their wildfire resistance.
4.2 | Post-wildfire ecosystem response
An emerging question is whether severely disturbed riparian for-
ests will recover back to their original state or remain arrested
by self-perpetuating vines, ferns and invasive grasses. In a similar
tropical savanna landscape of Belize, riparian forests were shown
to recover well from small wildfires (Kellman & Meave, 1997). At
the CVNP, however, the 2017 wildfire was a rare event that killed
most trees and saplings, and consumed most of the organic soil in
some forests (Figure 3), potentially destroying tree seed banks.
Remnant trees of ten contribute to attract animal dispersers and
fac ilitate forest recover y af te r di st urbance s, yet , wh en mos t tree s
have been killed, dispersal may be limited. Even if seeds are able
to arrive in burnt forests, they will have to overcome multiple re-
cruitment limitations, such as competition with herbaceous plants
(Figure 5d ,e). In bur nt for est s, in cr ea se in soil pho sp horus con ce n-
tration, as well as reduction in soil acidity and toxicity may now
boost the growth of opportunistic and invasive plants. In fact,
6 months after the fire, the non-native C4 grass Melinis minuti-
flora was already present in two severely burnt flooded forests
(Figure 3d), pro ba bly be nefiting from the impro ved soil cond itions
(Bustamante et al., 2012). Because of its high biomass and flam-
mable compounds, the expansion of M. minutiflora may reduce
tree s ee dl in g su rvi va l (H of fmann & Har id as an, 2008) an d enhance
overall ecosystem flammability (Hoffmann et al., 2004). Hence,
the spread of M. minutiflora and other opportunistic plants in
disturbed riparian forests may contribute to arrest forest succes-
sion, as previously shown in forests of the Cerrado and Amazonia
(Flores et al., 2016; Hoffmann & Haridasan, 2008; Veldman &
Putz, 2011).
An alternative possibility is that Mauritia flexuosa palms, as
well as native grass species may colonize burnt riparian forests,
as these sites are connected to palm swamp savannas along
streams (Figures 1b and 5). Mauritia flexuosa palms often survive
from wildfires and increase their seed production, potentially
leading to mass recruitment in burnt sites ( Arneaud et al., 2017).
Transitions from riparian forest to Mauritia swamp savanna have
been shown across the Neotropics by palaeoecological evidence
(Rull & Montoya, 2014). For instance, at the Venezuelan Gran
Sabana region, riparian forests were replaced by Mauritia swamp
savannas following an increase in fire activit y 2000 years ago
(Montoya et al., 2011). Fires initially arrested the ecosystem in a
state dominated by ferns for 200 years, until palm swamp savan-
nas expanded permanently (Rull et al., 2013). Interestingly, some
of the burnt forests we studied are also dominated by the oppor-
tunistic fern Pteridium arachnoideum (Figure 5d), a species known
to outcompete tree seedlings and arrest forest succession (Pivello
et al., 2018).
Our findings imply that, as large wildfires become more fre-
quent in tropical savannas, riparian forests will be increasingly
exposed to the risk of collapse (Scheffer et al., 2001; van Nes
et al., 2018). The complete destruction of some forests (Figure 5)
may open space for new plant species with contrasting functions
th at pr op el the ecos yst em to an alt ern at ive veg eta tio n sta te, whic h
could potentially be: (a) a degraded state with invasive grasses and
opportunistic plants or (b) a palm swamp savanna state with M.
flexuosa and native grasses. To reduce such risk, it is necessary
to manage landscape flammability. Some protected areas of the
Brazilian Cerrado, for instance, have already started implement-
ing an Integrated Fire Management program, with the use of pre-
scribed controlled fires (Schmidt et al., 2018). Small prescribed
fires act as micro-disturbances that help restore landscape hetero-
geneity, reducing wildfire spread (Mistry et al., 2005). The CVNP
only adopted this strategy at a larger scale after the 2017 event.
Additionally, the persistent control of non-native grasses is fun-
damental to reduce their invasion in disturbed sites. Managing
wildfires and invasive grasses at the landscape scale is challenging,
yet probably the best strategy to maintain riparian forests resil-
ient. These forests are an important habitat for large animals that
move across the savanna landscape (Redford & Fonseca, 1986),
implying that they contribute to stabilize trophic networks (Estes
et al., 2011).
In summary, our findings reveal how riparian forests embed-
ded in tropical savanna landscapes are in danger from large wild-
fires. Seasonally flooded forests were the most impacted by the
2017 wildfire at the CVNP, raising concerns about whether they
will recover or shift into an alternative ecosystem state. In our study
area, long fire-free periods allowed grass fuel to build-up, causing
wildfires to be intense and uncontrollable. In addition, a synergis-
tic combination of climate change (Jolly et al., 2015) and environ-
mental governance loss (Levis et al., 2020) is causing large wildfires
to happen more often in savannas landscapes of the Cerrado and
Pantanal, in Brazil (Mega, 2020). The most promising solution to re-
duce such risk probably lies in combining the ancient indigenous fire
management knowledge with recent scientific discoveries (Durigan
& Ratter, 2016; Mistry et al., 2005). New forms of management using
prescribed fires are already being applied in many fire-prone eco-
systems, restoring landscape heterogeneity and reducing the risk of
large wildfires (Buisson et al., 2019; Schmidt et al., 2018). Evidence
we present here contribute to these initiatives by stressing the need
to consider riparian forests as a vulnerable element of tropic al sa-
vanna systems.
ACKNOWLEDGEMENTS
The work was supported by the Fundação Grupo Boticário de
Pr ote çã o à Natur eza , grant 1114 -2 018 1. B. M . F. is fu n d e d by São Pa ulo
Research Foundation FAPESP grant 2016/25086-3. P.G. acknowl-
edges FAPESP grant 18/01847-0. B.M.F., M.H., P.G. and R.S.O. ac-
knowledge the grant from Instituto Serrapilheira/Serra-1709–18983.

10
|
Journal of Applied Ecology
FLORES E t aL.
A.A., A.L.G., P.d.B.C. and G.W. acknowledge the Coordenação de
Aperfeiçoamento de Pessoal de Nível Superior (CAPES, Brazil) Code
001. We thank the CVNP and ICMBio for logistics. B.M.F. and A.A.
acknowledge Sisbio for licences 6 4171-1 and 43511.
AUTHORS' CONTRIBUTIONS
B.M.F., M.H., M.d.S.D., I.B.S., A.A., M.B.P., A.B.S. and R.S.O.
conceived and designed the study; B.M.F., M. S.D., M.B.C., A.A .,
A.L.G., P.d.B.C., M.B.P., P.G., G.L.M., G.W., L.V., L.L.F.P., F.P. and
R.S.O. collected the field data; B.M.F. and G.G.M. analysed the
field data; L.V. and L.L.F.P. collected satellite data, and together
with B.M.F., analysed the data. All authors contributed to result
interpretation. B.M.F. led the writing and all authors contributed
substantially.
DATA AVAILAB ILITY STATE MEN T
Data available via the Dryad Digital Repository https://doi.org/
10.5061/dryad.rr4xg xd72 (Flores, de Sá Dechoum, et al., 2020;
Flores, Staal, et al., 2020). Satellite data are freely available from
Google Earth.
ORCID
Bernardo M. Flores https://orcid.org/0000-0003-4555-5598
Michele de Sá Dechoum https://orcid.org/0000-0002-3484-2498
Isabel B. Schmidt https://orcid.org/0000-0001-9420-6509
Anna Abrahão https://orcid.org/0000-0001-9295-2292
André L. Giles https://orcid.org/0000-0002-1973-400X
Peter Groenendijk https://orcid.org/0000-0003-2752-6195
REFERENCES
Arneaud, L . L ., Farrell, A. D., & Oatham, M. P. (2017). Mar ked reproduc-
tive plasticity in response to contras ting fire regimes in a neotropical
palm. Tropical Ecology, 58(4), 693–703.
Balch, J. K., Nepstad, D. C., Curran, L. M., Brando, P. M., Portela, O.,
Guilherme, P., Reuning-Scherer, J. D., & de Car valho, O. (2011).
Size, species, and fire behavior predic t tree and liana mor tality from
experimental burns in the Brazilian Amazon. Forest Ecology and
Management, 261(1), 68–77. https://doi.or g/10.1016/j.for eco.2010.
09. 0 29
Barlow, J., Franca, F., Gardner, T. A., Hicks, C. C., Lennox, G. D., Berenguer,
E., Castello, L., Economo, E. P., Ferreira, J., Guénard, B., Leal, C . G.,
Isaac, V., Lees, A. C., Parr, C. L., Wilson, S . K., Young, P. J., & Graham,
N. A. J. (2018). The future of hyp erdiverse tropical ecosystems.
Nature, 559(7715), 517–526.
Bates, D., Maechler, M., Bolker, B., & Walker, S. (2015). Fitting linear
mixed-effects models using lme4. Journal of Statistical Soft ware, 67,
1–4 8 .
Bengt sson, J., Bullock, J. M., Egoh, B., Everson, C., Everson, T., O'Connor,
T., O' Farrell, P. J., Smith, H. G., & Lindborg, R. (2019). Grasslands—
More important for ecosystem services than you might think.
Ecosphere, 10(2), e02582. https://doi.org/10.1002/ecs2.2582
Bjornstad, O. N . (2020). ncf: Spatial covariance functions. R package ver-
sion 1.2-9. Retrieved from https://CRA N.R-proj e ct.org/pack a ge=ncf
Brando, P. M., Balch, J. K., Nepstad, D. C ., Morton, D. C., Putz, F. E.,
Coe, M. T., Silverio, D., Macedo, M. N., Davidson, E. A., Nobrega,
C. C., Alencar, A., & Soares-Filho, B. S. (2014). Abrupt increases
in Amazonian tree mortality due to drought–fire interactions.
Proceedi ngs of the National Academy of Sciences of the United States of
America, 111(17), 6347–6352 . ht tp s://doi .or g/10.1073/pna s.13 05 4
99111
Bueno, M. L., Dexter, K. G., Penning ton, R . T., Pontara, V., Neves,
D. M., Ratter, J. A., & de Oliveira-Filho, A. T. (2018). The envi-
ronment al triangle of the Cerrado Domain: Ecological fac tors
driving shif ts in tree species composition between forests and
savannas. Jou rnal of Ecology, 106(5), 2109–2120. ht tps://doi.
org /10.1111/1365-2745 .12969
Buisson , E., Le Stradic, S., Silveira, F. A., Durigan, G., Overbeck, G. E.,
Fidelis, A., Fernandes, G. W., Bond, W. J., Hermann, J.-M., Mahy, G.,
Alvarado, S. T., Zaloumis, N. P., & Veldman, J. W. (2019). Resilience
and restoration of tropic al and subtropical grasslands, savannas, and
grassy woodlands. Biological Reviews, 94(2), 590–609.
Bustamante, M. M., de Brito, D. Q., Kozovit s, A. R., Luedemann, G., de
Mello, T. R., de Siqueira Pinto, A., Munhoz, C. B., & Takahashi, F. S.
(2012). Effec ts of nutrient additions on plant biomass and diversity
of the herbaceous-subshrub layer of a Brazilian savanna (Cerr ado).
Plant Ecology, 213(5), 795–808. ht tps://doi.org/10.1007/s1125
8-012- 0 042-4
Cochrane, M. A. (20 03). Fire science for rainforests. Nature, 421(6926),
913–919.
Damasceno, G., Souza, L., Pivello, V. R., Gorgone-Barbosa, E., Giroldo, P.
Z., & Fidelis, A. (2018). Impac t of invasive grasse s on Cerrado under
natural regeneration. Biological Invasions, 20(12), 3621–3629. https://
doi.org/10.10 07/s1053 0- 018-1800 -6
Dantas, V. D. L., & Pausas, J. G. (2013). The lanky and the corky: Fire-
escape strategies in savanna woody species. Journal of Ecology,
101(5), 1265–1272. https://doi.org/10.1111/1365-2745.12118
D'Antonio, C. M., & Vitousek , P. M. (1992). Biological invasions by ex-
otic grasses, the grass/fire cycle, and global change. Annual Review of
Ecology and Systematic, 23(1), 63–87. https://doi.org/10.1146/annur
ev.es.23.110192.000431
dos Santos, A . R., & Nelson, B. W. (2013). Leaf decomposition and fine
fuels in floodplain forests of the Rio Negro in the Brazilian Amazon.
Journal of Tropical Ecology, 29(5), 455–458. ht tps://doi.org/10.1017/
S 0 2 6 6 4 6 7 4 1 3 0 0 0 4 8 5
Durigan, G., & Ratter, J. A . (2016). The need for a consistent fire policy
for Cerrado conservation. Journal of Applied Ecology, 53(1), 11–15.
https ://doi.or g/10.1111/1365 -266 4.1 2559
Estes, J. A., Terborgh, J., Brashares, J. S., Power, M . E., Berger, J., Bond,
W. J., Carpenter, S. R., Essington, T. E., Holt, R. D., Jackson, J. B. C.,
Marquis, R. J., Oksanen, L., Oksanen, T., Paine, R. T., Pikitch, E. K.,
Ripple, W. J., Sandin, S. A., Schef fer, M., Schoener, T. W., … Wardle, D.
A. (2011). Trophic downgrading of planet Earth. Science, 333 (6040),
301–306.
Fidelis, A ., Alvarado, S., Barradas, A., & Pivello, V. (2018). The year 2017:
Megafires and management in the Cerrado. Fire, 1(3), 49. https://doi.
org /10.3 390/fire1 030049
Flores, B. M., de Sá Dechoum, M., S chmidt , I. B., Hirota, M., Abr ahão, A .,
Verona, L., Pecoral, L., Cure, M., Giles, A., Costa, P., Pamplona, M.,
Mazzochini, G., Groenendijk, P., Minski, G., Wolfsdorf, G., Sampaio,
A., Piccolo, F., Melo, L., Fiacador, R., & Oliveira, R . (2020). Data from:
Tropical riparian forests in danger from large savanna wildfires.
Dryad Digital Repository, https://doi.org/10.5061/dryad.rr4xg xd72
Flores, B. M., Fagoaga, R., Nelson, B. W., & Holmgren, M. (2016).
Repeated fires trap Amazonian blackwater floodplains in an open
vegetation state. Journal of Applied Ecology, 53(5), 1597–1603.
https ://doi.or g/10.1111/1365 -266 4.1 2687
Flores, B. M., Piedade, M. T. F., & Nelson , B. W. (2014). Fire disturbance
in Amazonian blackwater floodplain forests. Plant Ecology & Diversit y,
7(1–2), 319–327. ht tps://doi .or g/10 .1080/1755 0 874.2012.716 086
Flores, B. M., Staal, A., Jakovac, C. C., Hirota, M., Holmgren, M., & Oliveira,
R. S. (2020). Soil erosion as a resilience drain in disturbed tropical
forest s. Plant and Soil, 450(1) , 11–25. https ://doi.or g/10.10 07/s1110
4-019-04097 -8

|
11
Journal of Applied Ecology
FLORES E t aL.
Fusco, E. J., Finn, J. T., Balch, J. K ., Nagy, R. C., & Bradley, B. A. (2019).
Invasive grasses increase fire occurrence and frequency across
US ecoregions. Proceedings of the National Academy of Sciences of
the United States of America, 116(47), 23594–23599. https://doi.
org /10.1073/p nas.19082 53116
Hoffmann, W. A., & Haridasan, M. (2008). The invasive grass, Melinis
minutiflora, inhibits tree regeneration in a Neotropical savanna. Austral
Ecology, 33(1), 29–36. ht tps://doi .org/10.1111/j.14 42-9993. 2007.
01 787. x
Hoffmann, W. A., Jaconis , S. Y., Mckinley, K. L., Geiger, E. L.,
Gotsch, S . G ., & Franco, A . C . (2012). Fuels or microclimate?
Understanding the drivers of fire feedback s at savanna–forest
boundaries. Austral Ecolog y, 37(6), 634–64 3. https://doi.org/
10.1111/ j.14 42-9 993 .2011.02324.x
Hoffmann, W. A., Lucatelli, V. M. P. C ., Silva, F. J., Azeuedo, I. N. C., Marinho,
M. D. S., Albuquerque, A. M. S., Lopes, A. D. O., & Moreira, S. P. (2004).
Impact of the invasive alien grass Melinis minutiflora at the savan-
na-forest ecotone in the Brazilian Cerrado. Diversity and Distributions,
10(2), 99–103. https://doi.org/10.1111/j.1366-9516.2004.00 063.x
Jolly, W. M., Cochrane, M. A., Freeborn, P. H ., Holden, Z. A., Brown,
T. J., Williamson, G. J., & Bowman, D. M. (2015). Climate-induced
variations in global wildfire danger from 1979 to 2013. Nature
Communications, 6(1), 7537. ht tps://doi.org/10.1038/ncomm s8537
Kellman, M., & Meave, J. (1997). Fire in the tropical gallery for-
ests of Belize. Journal of Biogeography, 24 (1), 23–34. https://doi.
org /10.1111/j .136 5-269 9.1997.tb 000 47.x
Lawes, M. J., Midgley, J. J., & Clarke, P. J. (2013). Costs and benefits
of relative bark thickness in relation to fire damage: A savanna/
forest contrast. Journal of Ecology, 101(2), 517–524. https://doi.
org /10.1111/1365-2745 .12035
Levis, C., Flores, B. M., Mazzochini, G. G., Manhães, A. P., Campos-Silva,
J. V., de Amorim, P. B., Peroni, N., Hirota, M., & Clement , C. R. (2020).
Help restore Brazil's governance of globally important ecosystem
services. Nature Ecology & Evolution, 4(2), 172–173. https://doi.
org /10.103 8/s4155 9-019-1093-x
Loram-Lourenço, L., Farnese, F. D. S., Sousa, L. F. D., Alves, R. D. F. B.,
Andrade, M. C. P. D., Almeida, S. E. D. S., Moura, L. M. D. F., Costa, A.
C., Silva, F. G., Galmés, J., Cochard, H., Franco, A. C., & Menezes-Silva,
P. E. (2020). A structure shaped by fire, but also water: Ecological
consequences of the variability in bark proper ties across 31 species
from the Brazilian Cerr ado. Frontiers in Plant Science, 10(1) , 1718.
https://doi.org/10.3389/fpls.2019.01718
Mega, E. R. (2020). ‘Apocaly ptic’ fires are r avaging the world's largest
tropical wetland. Nature, 586, 20–21. https://doi.org/10.1038/d4158
6-020-02716 -4
Mistr y, J., Berardi, A ., Andrade, V., Krahô, T., Krahô, P., & Leonardos, O.
(2005). Indigenous fire management in the cerrado of Brazil: The
case of the Krahô of Tocantíns. Human Ecology, 33(3), 365–386.
htt ps://doi.org/10.10 07/s1074 5-0 05- 4143-8
Montoya, E., Rull, V., Stansell, N. D., Abbot t, M. B., Nogué, S., Bird, B.
W., & Díaz, W. A. (2011). Forest–savanna–morichal dynamic s in re-
lation to fire and human occupation in the southern Gran Sabana
(SE Venezuela) during the last millennia. Quaternary Research, 76 (3),
335–344.
Natta , A. K., Sinsin, B., & van der Maesen, L. J. G. (2002). Riparian for-
ests, a unique but endangered ecosystem in Benin. Botanische
Jahrbücher, 124 (1), 55–69. https://doi.org/10.1127/0 006 -8152/2002/
0124- 0 055
Nogueira, D. S., Marimon, B. S., Marimon-Junior, B. H., Oliveir a, E. A .,
Morandi, P., Reis, S. M., Elias, F., Neves, E. C., Feldpausch, T. R., Lloyd,
J., & Phillips, O. L. (2019). Impac ts of fire on forest biomass dynam-
ics at the southern amazon edge. Environmental Conservation, 46(4),
285–292. https://doi.org/10.1017/S0376 89291 9000110
Oliveira, P. S., & Marquis, R. J. (20 02). The cerrados of Brazil: Ecology an d
natural history of a neotropical savanna. Columbia University Press.
Parolin, P., De Simone, O., Haase, K., Waldhoff, D., Rottenberger, S.,
Kuhn, U., Kesselmeier, J., Kleiss, B., Schmidt, W., Piedade, M. T. F.,
& Junk, W. J. (2004). Central Amazonian floodplain forests: Tree ad-
aptations in a pulsing system. The Botanical Review, 70(3), 357–380.
Pellegrini, A. F. A., Andereg g, W. R. L., Paine, C. E. T., Hoffmann, W.
A., Kartzinel, T., Rabin, S. S., Sheil, D., Franco, A. C., & Pacala, S. W.
(2017). Convergence of bark investment according to fire and climate
structures ecosystem vulnerability to future change. Ecology Letters,
20(3), 307–316. https://doi.org/10.1111/ele.12725
Pettit , N. E., & Naiman , R . J. (2007). Fire in the riparian zone:
Characteristics and ecological consequences. Ecosystems, 10(5),
673–687. https://doi.org /10.10 07/s1002 1-007-9048-5
Pivello, V. R., V ieira, M . V., Grombone-Guar atini, M. T., & Matos, D. M.
S. (2018). Thinking about super-dominant populations of native spe-
cies–examples from Brazil. Perspectives in Ecology and Conservation,
16(2), 74–82. ht tps://doi.org/10.1016/j.pecon.2018.04.001
R Core Team. (2019). R: A language and environment for statis tical comput-
ing. R Foundation for Statistical Computing. Retrieved from https://
www.R-proje ct.org/
Redford, K. H., & da Fonseca, G. A . (1986). The role of gallery forests
in th zoogeography of the cerrado's non-volant mammalian fauna.
Biotropica, 18(2), 126–135. https://doi.org/10.2307/2388755
Resende, A . F., Nelson, B. W., Flores, B. M., & de Almeida, D. R. (2014).
Fire damage in seasonally flooded and upland forests of the Central
Amazon. Biotropica, 46(6), 64 3–6 46. htt ps://doi. org/10 .1111/
btp.1 2153
Ribeiro, J. F., & Walter, B. M. T. (2008). As principais fitofisionomias do
bioma Cer rado. Cerrado: Ecologia E Flora, 1(1), 151–212.
Rull, V., & Montoya, E. (2014). Mauritia flexuosa palm swamp commu-
nities: Natural or human-made? A palynological study of the Gran
Sabana region (northern South America) within a neotropical context.
Quaternary Science Reviews, 99(1), 17–33 . ht tp s://doi .org/10.1016/j.
quasc irev.2014.06.007
Rull, V., Montoya, E ., Nogue, S., Vegas-Vilarrubia, T., & Safont, E . (2013).
Ecological palaeoecology in the neotropical Gran Sabana region: Long-
term rec or ds of ve ge tatio n dy namic s as a basis for ecol og ic al hyp ot h-
esis testing. Perspectives in Plant Ecolog y, Evolution and Systematics,
15(6), 338–359. https://doi.org/10.1016/j.ppees.2013.07.004
Scheffer, M., Carpenter, S., Foley, J. A., Folke, C., & Walker, B. (20 01).
Catastrophic shifts in ecos ystems. Nature, 413 (6856), 591–596.
Schmidt, I. B., Moura, L. C., Ferreira, M. C ., Eloy, L ., Sampaio, A . B., Dias,
P. A., & Berlinck, C. N. (2018). Fire management in the Brazilian sa-
vanna: First steps and the way forward. Journal of Applied Ecology,
55(5) , 20 94–210 1. h tt ps ://doi.or g/10.1111/1365-266 4.13118
Stark, N. M ., & Jordan, C . F. (1978). Nutrient retention by the root mat
of an Amazonian rain forest. Ecology, 59(3), 434–437. https://doi.
org /10. 23 07/1936571
Strassburg, B. B. N., Brooks, T., Feltran-Barbieri, R., Iribarrem, A.,
Crouzeilles, R., Loyola, R., Latawiec , A. E., Oliveira Filho, F. J. B.,
Scaramuzza, C. A. D. M., Scarano, F. R., Soares-Filho, B., & Balmford,
A. (2017). Moment of truth for the Cerrado hotspot. Nature Ec ology &
Evolution, 1(4), 1–3. https://doi.org/10.1038/s4155 9-017-0099
Van Nes, E. H., Staal, A., Hantson, S ., Holmgren, M., Pueyo, S., Bernardi,
R. E., Flores, B. M., Xu, C., & Scheffer, M. (2018). Fire forbids fif-
ty-fifty forest. PLoS ONE, 13(1), e0191027. htt ps://doi.org /10.1371/
journ al.pone.0191027
Veldman, J. W., Buisson, E., Durigan, G ., Fernandes, G. W., Le Str adic, S.,
Mahy, G., Negreiros, D., Overbeck, G. E., Veldman, R. G., Zaloumis,
N. P., Putz, F. E., & Bond, W. J. (2015). Toward an old-grow th concept
for grasslands, savannas, and woodlands. Frontiers in Ecology and the
Environment, 13(3), 154–162. https://doi.org/10.1890/140270
Veldman, J. W., & Putz, F. E. (2011). Grass-dominated vegetation, not
species-diverse natural savanna, replaces degraded tropic al forests
on t he southern edge of the Amazon Basin. Biological Conservation,
144(5), 1419–1429. https://doi.org/10.1016/j.biocon.2011.01.011

12
|
Journal of Applied Ecology
FLORES E t aL.
Wantzen, K. M., Siqueira, A., Cunha, C. N. D., & Pereira de Sá, M. D. F. (2006).
Stream-valley systems of the Brazilian Cerrado: Impact assessment
and conservation scheme. Aquatic Conservation: Marine and Freshwater
Ecosys tems, 16 (7), 713–732. https://doi.org/10.1002/aqc .807
Wickham, H. (2016). ggplot2: Elegant graphics for data analysis. Springer-
Verlag. Retrieved from https://ggplo t2.tidyv erse.org/. ISBN
978-3-319-24277-4.
Zenni, R. D., Sampaio, A. B., Lima, Y. P., Pessoa-Filho, M., Lins, T. C.,
Pivello, V. R., & Daehler, C. (2019). Invasive Melinis minutiflora out-
performs native species, but the magnitude of the effect is contex t-
dependent. Biological Invasions, 21(2), 657–667. https://doi.org/10.1007/
s1053 0-018-1854-5
SUPPORTING INFORMATION
Additional supporting information may be found online in the
Supporting Information section.
How to cite this article: Flores BM, de Sá Dechoum M,
Schmidt IB, et al. Tropical riparian forests in danger from
large savanna wildfires. J Appl Ecol. 2020;00:1–12. ht t p s ://
doi .org /10.1111/1365-26 64.13794