Drought reduces the growth and heath of tropical rainforest understory plants

We 104 hypothesized that relative to non-droughted control plants, drought affected tree saplings and 105 shrubs would exhibit decreases in aboveground biomass, physiological performance

1 Drought reduces the growth and health of tropical rainforest understory plants

2 DAVID Y P TNG1,3,*, DEBORAH M G APGAUA1,3, CLAUDIA P PAZ2, RAYMOND W

3 DEMPSEY3, LUCAS A CERNUSAK3, MICHAEL J LIDDELL3, SUSAN G W

4 LAURANCE3

5

6 1Centre for Rainforest Studies, School for Field Studies, Yungaburra, Queensland 4872,

7 Australia

8 2Department of Ecology, Institute of Biosciences, São Paulo State University, Av 24A 1515,

9 Rio Claro, SP 13506-900, Brazil

10 3Centre for Tropical, Environmental and Sustainability Sciences, College of Science and

11 Engineering, James Cook University, 14-88 McGregor Rd, Smithfield Qld 4878, Australia

20 Abstract

21 Tree saplings and shrubs are frequently overlooked components of tropical rainforest

22 biodiversity, and it may be hypothesized that their small stature and shallow root systems

23 predisposes them to be vulnerable to drought However, these purported influences of

24 drought on growth, physiological performance and plant traits have yet to be studied in

25 simulated drought conditions in the field We simulated drought using a rainfall exclusion

26 experiment in 0.4 ha of lowland tropical rainforest in northeast Australia in 2015 After six

27 months, we compared the average change in aboveground biomass and plant health of

28 drought-affected tree saplings and understory shrubs with control individuals We also

29 assessed photosynthetic function, plant health and leaf traits in eight target species Both tree

30 saplings and shrubs had significantly lower aboveground biomass in the drought treatment

31 compared to the control Drought-affected individuals of target species exhibited a

32 significantly higher incidence of disease and insect attack, reduced photosynthesis, and a

33 range of leaf trait changes compared to control individuals We conclude that reduced growth

34 and photosynthetic capability, an increased susceptibility to insect attack, and leaf trait

35 changes constitute a near immediate drought response in tropical rainforest tree saplings and

36 shrubs Our results show that these often-overlooked components of tropical rainforest

37 biodiversity are likely to be the most rapidly and negatively impacted component of the plant

38 community in drought conditions

39 Keywords: drought, leaf economic spectrum, plant functional traits, tropical plant life forms,

40 tropical rainforest, throughfall exclusion

41

42 1 Introduction

43 An understanding of how plants respond to drought is an important cornerstone in the

44 study of how plants deal with environmental stresses and has real-world implications in

45 agricultural and ecological systems While the effects of drought on plants are relatively well

46 characterized in laboratory conditions and in particular for crop plants in agricultural settings

47 (Valladares & Pearcy 1997; Apgaua et al 2019), investigation of plant performance under

48 field conditions is fragmentary (Martínez-Ferri et al 2000; Schuldt et al 2011; Meir et al

49 2015a; Binks et al 2016; Tng et al 2018) Also complicating such studies is the fact that

50 plant response to multiple stresses (e.g drought, excessive light, heat, etc.) are usually not

51 predictable from single-factor studies (Valladares & Pearcy 1997; Corlett 2011, Rowland et

52 al 2015a)

53 Reductions in growth and widespread plant mortality are among the most worrisome

54 consequence of drought (Allen et al 2010: Liu et al 2015) However, susceptibility to

55 drought can vary across and within species, and moreover, drought-induced mortality is

56 thought to result from one or a combination of three processes: hydraulic failure, gradual

57 carbon starvation and/or invertebrate or pathogen attack (Adams et al 2017; Gely et al

58 2020) The relative contribution of these processes to mortality under drought conditions,

59 however, is poorly understood (McDowell et al 2008, 2013) For instance, droughts may

60 promote natural enemy attacks in water-stressed plants by reducing hosts’ natural chemical

61 defences and elevating nitrogen, sugars and secondary metabolites in foliage (Mattson et al

62 1987; Larsson 1989; Koricheva et al 1998) The level of damage to plants from these enemy

63 attacks appears to depend on the type of feeding substrate for insects and fungi, and the level

64 of water stress severity Jactel (2012) found taxa that attack both healthy and stressed plants

65 caused significantly more damage to foliage than wood in water-stressed trees irrespective of

66 drought severity

67 Plant responses to drought are often measured in terms of physiological performance

68 (Rennenberg et al 2006) Traits such as photosynthesis and stomatal conductance are

69 routinely measured when studying the effects of water deficit on plants, and most studies

70 show a decrease in these measures when plants are exposed to drought (Rennenberg et al

71 2006; Apgaua et al 2019) However, functional trait-based approaches to tracking plant

72 response to drought can also be helpful, providing another aspect to the story Leaf and wood

73 traits such leaf mass per unit area, leaf dry matter content, and wood density are important

74 components of the economic spectra in plants (Wright et al 2004; Chave et al 2009) While

75 plant functional traits are often used in ecosystem-scale studies as predictors of the

76 vulnerability or performance of plants when exposed to environmental stressors (Greenwood

77 et al 2017), it is also instructive to examine how these traits respond to environmental

78 changes, particularly when the question relates to responses of individual species (Bjorkman

79 et al 2018; Yue et al 2019; Tng et al 2018) For instance, it may be hypothesized that plants

80 exposed to drought will exhibit a decrease in leaf traits such as leaf fresh weight, leaf fresh

81 weight to dry weight ratios, leaf toughness and leaf mass per unit area, due to changes in leaf

82 cell turgor pressure and nutrient changes (Chen et al 2015; Delzon 2015) In turn, these leaf

83 functional trait changes may serve as the mechanism that leads to lower physiological

84 performance and vulnerability to natural enemies Quantifying the link between plant

85 functional traits and the environment is therefore important for understanding the potential

86 impacts of climate change on plant communities

87 Most field studies examining the effects of droughts in tropical rainforest have

88 focused on mature trees (Meir et al 2015a; Schuldt et al 2011) However, tree saplings and

89 understory shrubs can play important roles in maintaining rainforest diversity and vegetation

90 dynamics (Royo & Carson 2006), and their responses to drought therefore deserve closer

91 examination Tree saplings, whilst regarded as being more susceptible than mature trees to

92 the negative impacts of drought (Niinemets 2010), have rarely been studied under

93 experimental field conditions Likewise, there are also few studies on how drought affects

94 smaller plant lifeforms such as understory shrubs (Condit et al 1995)

95 Rainfall exclusion or throughfall infrastructures represent a robust way to

96 experimentally induce a drought on a forest stand to investigate plant responses in situ (Meir

97 et al 2015b; Rowland et al 2015b) However, due to the sheer scale of such endeavours,

98 there have only been four tropical rainforest throughfall exclusion infrastructures established

99 to date: two in eastern Amazon, both each one ha in size (Nepstad et al 2007; da Costa et al

100 2010); one in Sulawesi (Schuldt et al 2011); and, one in tropical Australia (the Daintree

101 Drought Experiment: Laurance 2015; this paper) The establishment of the Daintree Drought

102 Experiment in tropical Australia provided us with an opportunity to examine the effects of a

103 short-term drought (six months) on tropical rainforest tree saplings and shrubs We

104 hypothesized that relative to non-droughted control plants, drought affected tree saplings and

105 shrubs would exhibit decreases in aboveground biomass, physiological performance

106 measures such as photosynthesis and stomatal conductance, and leaf traits (discussed earlier)

107 We also hypothesized that droughted plants would be subjected to higher levels of leaf

108 herbivory, insect attack and diseases

114 Our study site is located at the Daintree Rainforest Observatory (16°06′20′′S

115 145°26′40′′E, 50 m a.s.l.; Tng et al., 2016; Fig 1a) in a lowland rainforest adjacent to the

116 Daintree National Park in Cape Tribulation, north-eastern Australia The Daintree research

117 site commenced in 1998 with the installation of an industrial crane (Liebherr 91C) and the

118 establishment of a 1 -ha census plot The site experiences a tropical climate, with mean

119 temperatures of 24.4oC and a relatively high annual average rainfall of 4900 mm annum-1

120 (Bureau of Meteorology, 2015) The rainfall is highly seasonal with 66% falling between

121 January and April, the wet season The forest type at the site has a complex vertical profile,

122 with canopy heights ranging from 24 to 33m (Liddell et al., 2007), and a wide variety of plant

123 lifeforms (Tracey, 1982) Soils are developed over metamorphic and granitic colluvium and

124 are of relatively high fertility (Bass et al., 2011)

125

126

127 Fig 1 Study location (a) in the Daintree Rainforest Observatory, north Queensland, Australia

128 and (b) schematic, (c) top-down view with the throughfall exclusion panels visible under the

129 tree canopy, and (d) cross-section of the throughfall exclusion experimental setup, showing

130 the arrangement of panels and the gutters used respectively to intercept and channel rainfall

136 A throughfall infrastructure to exclude rainfall was implemented in May 2015 in two

137 rectangular 0.2 -ha patches within the 1-ha crane plot, with the remaining 0.6 ha of the plot

138 serving as a control experimental patch (Fig 1b; Laurance 2015) The rainfall exclusion

139 infrastructure consists of two 50 x 40 m clear-panel roofing structures which capture and

140 remove water from the 0.4 -ha (Fig 1c) The roofing panels are installed in between rows of

141 raised aluminium sheet gutters used to funnel rainwater away The panels taper at a height of

142 c 2.8m (Fig 1d), and therefore completely cover all trees sapling, shrub and herb lifeforms

143 under that height Where needed, slits were made in the roofing panel to accommodate all

144 stems above 2.8m height, such that their crowns are allowed to emerge through the roofing

145 panels

146

147 2.3 Understory microclimate and soil moisture

148

149 The presence of roofing structures might lead to modifications in microclimate that

150 need to be addressed To do this we recorded microclimate data from the drought and control

151 patches using a portable custom-made manifold This manifold consisted of a pyranometer

152 (Apogee SP-215-L) which measures solar radiation flux density, a temperature and relative

153 humidity probe (Model CS215, CMOSens®), and a datalogger (CR200X, Campbell

154 Scientific®) mounted on a pole and affixed to a tripod at a height of 1.7m We set the

155 datalogger to log light intensity (W/m2), relative humidity (%) and temperate (˚C)

156 measurements every minute for 15 minutes from 36 random spots (18 random spots each in

157 the control- and drought-treatment sectors), resulting in 15 data points for each variable per

158 spot Because we were limited by having only one manifold, we collected microclimatic data

159 between 1000hrs to 1500hrs over two days in November 2015, alternating between control-

160 and drought-treatment sectors after making measurements at any given spot This enabled us

161 to randomize locations during the period of measurements

162 We obtained volumetric soil water content from soil moisture censors installed at eight

163 soil pits stratified across both control and drought treatments (four pits each) Within each

164 soil pit, volumetric soil water content (cm3 cm-3) was measured continuously using time

165 domain reflectometry (TDR) probes (CS616, Campbell Scientific, UK) installed to log soil

166 moisture at four soil depths: 10, 50, 100, and 150 cm

167

168 2.4 Plant growth responses

169

170 To obtain an assessment of overall growth or mortality since the throughfall

171 infrastructure was implemented, we used nine established 10 m x 2 m rectangular subplots to

172 conduct demographic assessments of saplings and shrubs, six of which are now within the

173 drought treatment areas of the 1-ha plot and three in the control The subplots were

174 established in May 2015 where every tree sapling (individuals >1cm diameter at a stem

175 height of 1.3 m height) and shrub (individuals >0.4 cm diameter at a stem height of 5 cm)

176 was tagged, identified, and measured with a calliper at those respective stem heights (Tng et

177 al 2016) To ensure the accuracy of subsequent measurements, we used white liquid paper

178 ink to mark the point of measurement on the shrub of sapling individual The subplots were

179 marked out and established whilst the foundations of the throughfall-exclusion infrastructure

180 were being installed, so an effort was made to ensure that subplots established in the areas to

181 be droughted were situated in-between and parallel to the rows of gutters (inter-gutter width

182 of five meters) During the installation of the trough-drainage system of the

throughfall-183 exclusion infrastructure, a number of tree saplings and shrub stems had to be trimmed but this

184 damage was limited mostly to narrow strips of area just beneath the aluminium gutters and

185 did not impact plants within our subplots However, there was a difference in density

186 distribution of saplings and shrubs (excluding palms and tree stems with crowns above the

187 panels) within the 1-ha plot due to natural variability Therefore, the three control and six

188 drought treatment subplots respectively had 29 and 22 sapling species (37 spp total) and 7

189 and 6 shrub species (9 spp total) These species were comprised of 90 and 81 sapling and 65

190 and 60 shrub individuals within the control and drought treatment subplots respectively

191 (Supplementary Material Table S2)

192 We distinguished between tree and shrub life-form for the species within our subplots

193 based on their well-documented life history (Hyland et al 2010) and demographic data from

194 the 1-ha long term monitoring plot (Tng et al 2016) The tree sapling and shrubs we censused

195 within the subplots were restricted to individuals within the 0.5-2.5 m height class, which

196 ensured that each individual had their crown wholly under the rainfall-exclusion panels This

197 also circumvented any bias due to possible irrigation, albeit minimal, that might occur from

198 stem flow in individuals with crowns emerging out above through slits in the panels The

199 same 2.5 m height limit was applied for the target species on which we made trait

200 measurements (see later)

201 In November 2015, six months after our initial census, we censused and

re-202 measured the stem diameter and heights of the tree saplings and shrubs within the nine

203 subplots, and also visually estimated plant health (see later) on all individuals Initially, we

204 had intended to re-census the sapling and shrub growth after an additional six months (in

205 May 2016) but during a field assessment 11 months into the experiment in April 2016, the

206 rainfall exclusion panels had begun to develop a layer of algal growth which conspicuously

207 reduced the light conditions under the panels and would therefore confound further growth

208 analyses

209

210 2.5 Plant health and physiological performance

211

212 For a more targeted within species examination of plant responses to drought, we used

213 a number of non-destructive methods to parameterize drought responses, following

214 Niinemets (2010) These included: (i) quantitative visual estimates of plant health (herbivory,

215 disease symptoms and presence of insect pests); (ii) physiological performance measures,

216 and; (iii) leaf traits

217 We selected eight target species of common tree saplings and shrubs for which we

218 could locate replicates with ease within the overall 0.4 and 0.6 ha drought and control patches

219 respectively Our target species consist of the saplings of five species of mature-phase trees,

220 Argyrodendron peralatum (Malvaceae), Cleistanthus myrianthus (Phyllanthaceae),

221 Endiandra microneura (Lauraceae), Myristica globosa subsp muelleri (Myristicaceae),

222 Rockinghamia angustifolia (Euphorbiaceae); and three shrubs, Amaracarpus nematopodus,

223 Atractocarpus hirtus (Rubiaceae) and Haplostichanthus ramiflorus (Annonaceae) (Table 1)

224 For brevity, we henceforth use only genus names when referring to these species

225 Although these targeted species occurred within the nine subplots, we sampled

226 individuals outside the subplots for leaf traits to minimize impacts to the long-term

227 monitoring setup that may result from collecting leaf material for functional trait analysis

228 Pertinently also, some of the target shrub species occurred only sparingly within the subplots

229 and so for this targeted species analysis it was expedient for us to sample outside of the

230 subplots to obtain sufficient replication (n = 5-12 individuals per species within each

231 treatment) of these species to provide reliable trait estimates

232 Plant health was visually estimated on replicate plants of each target species both

233 within and outside the subplots in terms of the overall percentage of the leaves on each

234 individual plant with signs of herbivory, disease, and insect attack by at least two observers

235 (Table 1) Herbivory was defined as obvious holes or areas of the leaves that had been

236 predated on; disease as observable patches of yellow, white or dark discolouration, or

237 necrosis on leaves, and; insect attack as the presence of sap sucking insects such as

238 mealybugs or scale insects on leaves and/or shoots Both top and bottom leaf surfaces were

239 inspected for symptoms of disease and presence of sap-sucking insects

240 Table 1 Species of targeted tree saplings and shrubs sampled in the control and drought

241 treatment for disease symptoms, herbivory, and insect attack after six months of drought

242 treatment in a throughfall exclusion experiment at the Daintree Rainforest Observatory, Cape

Myristica globosa subsp muelleri

(Warb.) W.J.de Wilde

245 For plant physiological performance indicators, we used leaf photosynthetic rate (A:

246 µmol CO2 m-2 s-1) and stomatal conductance (g s: mol H2O m-2 s-1), which we measured

247 between 1000hrs to 1500hrs using a LI-6400 Portable Photosynthesis System (LI-COR,

248 Lincoln, Nebraska, USA) For this purpose, we took point measurements on one fully

249 expanded leaf per individual for five replicate individuals of each of the targeted species

250 within the control and drought treatments Photosynthesis and stomatal conductance

251 measurements were conducted in November 2015

252

253 2.6 Leaf functional traits

254

255 To obtain a measure of leaf functional trait responses, we sampled 5-12 leaf replicates

256 per species from each treatment following a standard protocol (Pérez-Harguindeguy et al

257 2013) Leaf fresh mass (g), dry mass (g), fresh mass: dry mass ratio (g g-1), leaf mass per unit

258 area (LMA: g cm-2) were measured from 20 leaf discs per individual collected with a 0.6mm

259 hole punch Leaf toughness was measured using a penetrometer to determine the amount of

260 force (in grams: g) needed to penetrate the leaf lamina when applied to three random spots on

261 the leaf, avoiding visible secondary and tertiary veins We deviated from the standard

262 protocol of measuring leaf fresh mass: dry mass ratio by measuring the leaf fresh weights

263 immediately after collection and without rehydration as we wanted to obtain a more realistic

264 measure of leaf hydration status of samples under field conditions

265

266 2.7 Data analysis

267

268 To summarize the microclimate data, we averaged the 15 data points at each spot for

269 solar irradiance flux density, relative humidity and temperature, and calculated the means of

270 these variables for the control- and drought-treatment plots Because the experiment was

271 designed for analysis as a pairwise comparison between the control- and drought-treatments,

272 we compared the means of all the microclimate variables using one-tailed t-tests (α = 0.05)

273 We examined soil volumetric water content differences between drought and control areas

274 using a linear mixed effects model using the package lmerTest with the daily estimates of soil

275 volumetric water content considered repeated measures and accounted for as a random factor

276 We then run an analysis of variance on the lmer model to obtain F and P values for the

277 contrasts and their interactions The least square means for the model are presented in

278 Supplementary Material Table S1 For visualization purposes, data were averaged for each

279 depth at each pit over a six-month period from 1/5/2015

280 For the analysis of the growth data, we pooled the individuals from subplots within

281 each treatment, and analyzed the sapling and shrub dataset separately To parameterize the

282 growth response of the saplings and shrubs, we first calculated the aboveground biomass

283 (AGB, kg) of each individual sapling or shrub for each census using stem diameters (D: cm),

284 plant height (H: cm) and wood density (WD: kg) following an equation by Chave et al

285 (2014), where: AGB = 0.0673 x (D2 x H x WD)0.976 The choice of Chaves equation was

286 based on the widespread use of this equation in rainforest tree biomass estimates and the lack

287 of any parametric equation for tropical rainforest saplings/shrubs Where individual plants

288 were represented by multiple stems, the AGB for each stem was calculated and then summed

289 to obtain the AGB for the individual Wood density values for most of the species in our

290 subplots were obtained from Apgaua et al (2015, 2017) and supplemented with our

291 unpublished data We then calculated the percentage change in AGB (%ΔAGB) for each

292 individual by the following equation: %ΔAGB = [(AGBfinal-AGBinitial)/AGBinitial] x 100,

293 where AGBinitial and AGBfinal refers to the aboveground biomass of each individual in the first

294 (May 2015) and final census (Nov 2015) respectively

295 To test our hypothesis of whether sapling and shrub individuals within the drought

296 subplots in general showed a greater magnitude of responses in terms of insect incidence,

297 disease symptoms and herbivory relative to the control, we fitted generalized linear models

298 individually for saplings and shrubs For insect incidence and disease symptoms, we fitted

299 zero-inflated generalized linear mixed models, using the glmmTMB package (Brooks et al

300 2017), which fit zero-inflated Poisson models with a single zero-inflation parameter applying

301 to all observations For percentage change in aboveground biomass, insect incidence, disease

302 symptoms and herbivory in tree saplings and shrubs in the subplots, we used linear mixed

303 effects models with the restricted maximum likelihood estimation, using the nlme package In

304 all models, we used treatment (drought or control) as an explanatory variable and individual

305 aboveground biomass in the initial census as a random effect (to account for any size

306 dependent effects)

307 To test whether there were species specific responses within our eight target species

308 in plant health, plant performance, leaf functional traits, and physiological measures, we

309 fitted generalized linear models using treatment as the explanatory variable In the case of

310 insect incidence and disease symptoms, we fitted zero-inflation regression models using the

311 zeroinfl function in the pscl package, which fits zero-inflated data via the maximum

312 likelihood estimation (Zeileis et al 2008) All analyses were performed in R 3.0 following a

313 standard protocol of data exploration (Zuur et al 2010)

319 Microclimate measures in the control and drought treatments ranged respectively

320 between 3.85–99.8 W m-2 and 3.89–166.3 W m-2 for light intensity (Fig 2a); 78.6–96.6% and

321 78.9–98.1% for relative humidity (Fig 2b), and; 26.1–31.0⁰C and 26.6–31.6⁰C for

322 temperature (Fig 2c) T-tests comparing the means of these measures between control- and

323 drought-treatments showed no significant differences (all P > 0.05).

325

326 Fig 2 Boxplots of microclimate variables of (a) solar irradiance flux density, (b) relative

327 humidity, and (c) temperature for random point samples (n = 18 points each) in the control-

328 (green symbols) and drought-treatments (brown symbols) in a throughfall exclusion

329 experiment at the Daintree Rainforest Observatory, Cape Tribulation, Australia The (d) soil

330 volumetric soil water content (VWC) was measured from 1.5 m long soil probes installed

331 within the control and the drought areas (n = 4 soil probes in each) of the study plot during

332 the 6-month experimental period Each box encompasses the 25th to 75th percentiles; the

333 median is indicated by the boldest horizontal line and the other horizontal lines outside the

334 box indicate the 10th and 90th percentiles Pairwise differences are indicated (ns = not

335 significant; P < 0.05*, P < 0.01**, P < 0.001***).

336

337 Over the six-month study period, the thoroughfall exclusion infrastructure succeeded

338 in significantly drying the soils of the top 100 cm of the soil profile compared to the control

339 treatment (ANOVA F1,4601 = 1228.57, P < 0.0001) This interaction between drought

340 treatment and depth was significant at surface and subsurface depths (ANOVA F3,4599 =

341 213.4, P < 0.0001) Soils in the drought experiment were on average 28.6% drier at the

342 surface (10cm), 20.2% drier at the subsurface (50 cm), and 9.3% and 3.5% drier at 100 and

343 150 cm depths, respectively than in the controls (Fig 2d) However, at soil depths of 1.5m,

344 differences in volumetric water content were not significant (Fig 2d; See also Supplementary

345 Material Table S1 for the least square means for the ANOVA run on the soil pit data)

346

347 3.2 Plant growth responses of tree saplings and shrubs

348

349 During the November 2015 census, we found two dead sapling individuals and one

350 dead shrub in the drought treatment subplots, and no dead individuals in the control subplots

351 (Table 2; Supplementary Tables S2) Individuals of tree saplings and shrub individuals

352 exhibited aboveground biomass increments, reductions or lack of change in both control and

353 drought treatment subplots, but in general more individual stems in the control subplots

354 exhibited increases (Table 2) There was a net increase in tree sapling aboveground biomass

355 in both the control (14.51%) and drought treatment (+5.77%) subplots (Table 2) For shrubs,

356 the control subplots exhibited a net increase in aboveground biomass (+17.01%) but the

357 drought treatment subplots showed a reduction (-2.56%) (Table 2) Consequently, the net

358 percentage change in aboveground biomass for both tree saplings and shrubs transects was

359 significantly higher in the control than in the drought treatment subplots (Fig 3a)

360

361 Table 2 Percentages of tree saplings and shrubs in the subplots showing increases, decreases

362 or no changes in aboveground biomass (AGB) between 2015 to 2016 in the control and

363 drought treatment in a throughfall exclusion experiment at the Daintree Rainforest

364 Observatory, Cape Tribulation, Australia

367 Fig 3 Boxplots showing changes in lowland tropical rainforest tree saplings (top panel) and

368 shrubs (bottom panel) in terms of (a) percentage change in aboveground biomass (AGB) and

369 the differences in plant health measures: (b) insect incidence (c) disease symptoms, and: (d)

370 herbivory The number of individuals (n) of tree saplings and shrubs in the three control and

371 six drought subplots was 90 and 65, and 81 and 60 respectively Each box encompasses the

372 25th to 75th percentiles; the median is indicated by the horizontal line within the box and the

373 other horizontal lines outside the box indicate the 10th and 90th percentiles Open circles

374 indicate outliers and dots represent individual data points Significant differences between

375 treatments are indicated (See Table 3 for statistics)

376

377

378 Table 3 Parameter estimates (±SE) and random effect variances for linear mixed models

379 fitted for change in aboveground biomass (ΔAGB), percentage of insect incidence, disease

380 symptoms and herbivory on sapling and shrubs as responses, and treatment (control vs

381 drought) as fixed effects and initial plant aboveground biomass as random effects For the

382 percentage of insect incidence and disease symptoms, we fitted a zero-inflated generalized

383 linear mixed models, and for herbivory we fitted linear mixed effects models (See Methods)

Parameter Statistics Intercept Drought Random residual

p-value 3.67E-11 <2.00E-16***