Interactions among fire, grazing, harvest and abiotic conditions shape palm demographic responses to disturbance


Correspondence author: E-mail:


1. Determining the drivers of plant demography is integral to understanding the processes that shape plant species abundances and distributions. Despite recognition that interactions among drivers have important effects on demographic processes, few demographic studies test for interactions among multiple drivers in plants.

2. We used a factorial-design experiment to study the interactive effects among three common forms of disturbance in the tropics (fire history, grazing and leaf harvest by humans) on the vital rates of Phoenix loureiri (mountain date palm) in South India. In addition, we tested for interactive effects among these disturbances, abiotic conditions and plant size. We also tested for non-consumptive effects of grazing and harvest, such as trampling, by measuring the intensities of grazing and harvest in plots open to these disturbances.

3. Intensities of leaf harvest and grazing varied with abiotic conditions and disturbance. Leaf harvest decreased with increasing grazing intensity, suggesting that the net effect of harvest on palm populations is less where it co-occurs with grazing. In areas without fire, plots with lower soil moisture had higher grazing intensities.

4. We found multiple significant main and interactive effects of disturbance on palm vital rates. Palm mortality increased with fire and grazing. Grazing and harvest reduced growth, but growth increased following fire. The negative impact of harvest on palm individuals was reduced when harvest occurred in plots with fire.

5. We found evidence of non-consumptive effects of grazing and harvest on palm growth, likely from trampling. Studies inferring the effects of grazing by comparing grazed and ungrazed individuals within an area where grazing occurs will likely underestimate grazing effects.

6.Synthesis. Our findings reveal that P. loureiri demographic rates are driven by interactive effects among multiple forms of disturbance and abiotic factors and that the intensities of disturbance are themselves driven by interactions between other forms of disturbance and abiotic factors. These results illustrate that understanding the effects of, and interactions among, multiple drivers will be key in attempts to mitigate the effects of environmental change on plant species declines.


Understanding the processes that shape patterns of species abundances and distributions is a fundamental goal of ecology. Determining the drivers of population dynamics is a critical step towards achieving this goal. Several recent studies have demonstrated that plant populations of the same species exhibit substantial spatial and temporal demographic variation (Buckley et al. 2010; Jongejans et al. 2010), but we still lack an understanding of the factors responsible for this observed variation.

Moreover, despite a recognized need to understand interactions among drivers to determine the mechanisms underlying ecological processes (Agrawal et al. 2007; Didham et al. 2007), a limited number of studies have examined the interactive effects of drivers on plant demography. The few studies that have explicitly tested for interactions suggest that they are likely to be common and that overlooking them impairs our ability to understand population dynamics (e.g. Elderd & Doak 2006; Schleuning, Huamán & Matthies 2008; Farrington et al. 2009; Martínez-Ramos, Anten & Ackerly 2009). Interactive effects may be subadditive, with less of an impact in combination than would be predicted from each alone. For example, root harvest of Panax quinquefolius (American ginseng) reduced long-term population growth rates less when harvest co-occurred with deer browsing because browsed Panax individuals were less visible to harvesters and therefore had higher rates of survival (Farrington et al. 2009). Interactive effects may also be synergistic, having a greater impact in combination. For example, population growth rates of the boreal shrub Vaccinium myrtillus declined with both increasing herbivory and increasing resources, but herbivory had a stronger negative effect in high resource conditions (Hegland, Jongejans & Rydgren 2010).

With this complexity, experimental tests of the effects of drivers on plant demography are especially needed, but limited in number (Crone et al. 2011). Furthermore, if this ecological understanding is to be used to mitigate the effects of disturbance and climate change, it is necessary to differentiate among interaction chain and interaction modification effects (Didham et al. 2007). An interaction chain is a series of directly linked drivers, whereas an interaction modification effect occurs when the per-unit effect of one driver depends on the environmental context of other drivers. With interaction chains, manipulation of a single driver might effectively produce the desired outcome. With interaction modification effects, managing the multiple interacting drivers simultaneously is likely to be necessary (Didham et al. 2007). Determining the relative importance of demographic drivers and the nature of their interactions is critical to understanding and managing the effects of global environmental change.

Disturbance is recognized as an important driver of the demography of many plant populations (Sousa 1984). Disturbances can have both consumptive effects (due to the removal of biomass) and non-consumptive effects on plant demography. In the case of grazing, for example, non-consumptive effects may include direct effects such as trampling, as well as indirect effects such as changes in soil fertility or reduced competition (e.g. Hobbs 1996; Rooney & Waller 2003; Heckel et al. 2010). Separating consumptive from non-consumptive effects in demographic studies has proved difficult (Maron & Crone 2006), and non-consumptive effects are often unaccounted for (see review by Schmidt et al. 2011). However, ignoring non-consumptive effects can lead to under- or over-estimation of the effects of disturbance.

We used an experimental study, integrated with observational data for important covariates, to test for interactions among drivers of plant demography of wild Phoenix loureiri Kunth (mountain date palm) in a savanna woodland in South India. Phoenix loureiri populations are subject to multiple forms of disturbance, including frequent anthropogenic ground fires, herbivory by ungulate grazers (both domesticated and wild) and leaf harvest by local human communities. These three forms of disturbance – fire, grazing and plant harvest – are common and frequent sources of disturbance in tropical forests and savannas (FAO 2010). Furthermore, these disturbances are similar to naturally occurring disturbances with which the palm has evolved (i.e. wildfires, grazing by wild ungulates and leaf loss to falling overstorey branches). However, due to increasing anthropogenic activities, these disturbances occur today with greater frequency and intensity than they have in the past, as is the case in many other tropical systems (Oesterheld, Sala & McNaughton 1992; Olff & Ritchie 1998; Bond, Woodward & Midgley 2005). With these increases, understanding the effects of fire history, leaf harvest and grazing is especially important to understanding how environmental change is likely to influence plant demography, and therefore species distributions and abundances.

Vegetation, disturbance regime and abiotic conditions are known to have reciprocal and interactive effects on each other (e.g. Bezemer et al. 2006; Kirkpatrick, Marsden-Smedley & Leonard 2011). However, the implications of these interactions for plant demography have not been well studied (Hawkes & Sullivan 2001; Maron & Crone 2006; Wisdom et al. 2006). Fire and grazing may have synergistic effects, as grazers have been shown to prefer burned areas (Sensenig, Demment & Laca 2010). There are also likely to be synergistic interactions among fire, grazing and leaf harvest based on evidence that resilience to disturbance in palms decreases with multiple concurrent forms of disturbance (Chazdon 1991). The negative effects of harvest and grazing may be lower in high light conditions – a subadditive interaction – as understorey palms are often light limited and better able to compensate for disturbance in high light conditions (Anten, Martínez-Ramos & Ackerly 2003). Previous studies of palm demography have demonstrated that loss of palm leaves can reduce palm survival, growth and reproduction and that these vital rates are also affected by abiotic conditions (e.g. Ratsirarson, Silander & Richard 1996; Endress, Gorchov & Berry 2006; Martínez-Ramos, Anten & Ackerly 2009). The effect of grazing on palm demography has also been found to vary with substrate type (Berry, Gorchov & Endress 2011). However, experimental tests of the interactive effects among multiple forms of disturbance have not been carried out in wild palm populations. As individual vital rates and fitness are often correlated with plant size, determining how the effects of disturbance and abiotic conditions vary with plant size is also important to understanding demographic responses to these drivers.

To test the effects of and interactions among these three forms of disturbance on palm demographic rates and their relationship with abiotic environmental factors, as well as the non-consumptive effects of grazing and harvest, we established a split-plot fully crossed three-way factorial experiment. The actual intensities of grazing and harvest on plants exposed to these disturbances were allowed to vary. We used this study design to ask:

  • 1 Do the intensities of grazing and leaf harvest differ across abiotic conditions?
  • 2 What are the effects of fire history, grazing and leaf harvest on rates of palm mortality, growth and reproduction? Are there interactive effects between fire history, grazing and leaf harvest?
  • 3 Do the effects of and interactions between fire history, grazing and leaf harvest depend on plant size?
  • 4 Do the effects of and interactions between fire history, grazing and leaf harvest differ with abiotic conditions (soil moisture and light availability)?
  • 5 Are there non-consumptive effects of grazing and leaf harvest on palm vital rates?

We expected to find higher intensities of grazing in plots with recent fire. Given palm harvesters’ reported preference for leaves from plants recovering from fire, we also expected to find higher intensities of harvest in areas with recent fire. In addition, we expected synergistic interactions among harvest, grazing and fire, as well as subadditive interactions between light levels and harvest and grazing. We expected to find evidence of non-consumptive effects of grazing and leaf harvest such that palm growth rates would be lower in plots open to harvest and grazing, even on individual plants that escaped harvest and grazing.

Materials and methods

Study species and site

Phoenix loureiri is widely distributed across sub-Himalayan Asia, from India through southern China into Taiwan and the Philippines, where it occurs from sea level to 1700 m in open grasslands and scrublands or forest understorey (Barrow 1998). The species is dioecious and can reproduce clonally by basal suckers. Within our study site, nearly all genets had multiple genetically identical stems (ramets). Individual palms can grow up to 5 m tall, but in our study site they remain shrubby with stems <30 cm tall, as is common in dry and disturbed areas (Barrow 1998). In South India, leaves of c.60 cm in length or longer are harvested for hand brooms.

Our study took place in the Sathyamangalam Reserve Forest, Tamil Nadu state, India, part of the Western Ghats Biodiversity Hotspot (Mittermeier et al. 2005). Our study site was located in a savanna woodland with c.900 mm of annual rainfall, on mountain slopes at 1400 m. Here, P. loureiri occurs abundantly in the understorey, and its leaves are harvested by local people during the dry season, generally from February through May. Commercial palm leaf harvest has occurred in the study area since 1975. Livestock from local villages graze in the study site and may browse on palm leaves, although the palm is not preferred fodder. Wild ungulates, including gaur (Bos gaurus) and sambar deer (Rusa unicolor), are also present at the site and browse on palm leaves. In addition, the Asian elephant (Elephas maximus) consumes palm leaves and the stems of young plants and may uproot palms (Sukumar 1990). No elephant herbivory occurred within our study plots. From the perspective of the palm, ungulate grazing and harvest are similar in that both reduce the photosynthetic area of the palm. However, ungulate grazing removes only the upper part of the palm leaf, whereas harvest removes the full leaf and part of the petiole. Ungulates preferentially graze from leaves that have not fully expanded, while harvesters collect larger, fully developed leaves (L. Mandle, personal observation). Ground fires, often set by local people to manage for fodder and harvested plant species, are a frequent occurrence. It is rare to find areas that have not burned within the past 3 years. The frequency of fire has likely increased over the past century, based on trends in similar vegetation in a neighbouring area (Kodandapani, Cochrane & Sukumar 2004).

Experimental design and treatments

We established five replicate blocks of 8 plots each, for a total of 40 plots. Blocks were located in areas that had partially burned in February–March during the dry season just prior to the establishment of the study in August 2009. We used a split-plot design, with four fire plots per block located within a burned patch and four corresponding no-fire plots located outside and adjacent to the burned patch. The burned split plots in the five blocks were produced by two separate fires that burned a larger area. We established blocks to control for potential spatial variation in environmental conditions including soil and fire properties. Within each fire and no-fire area of each block, we randomly selected four palm genets of 5–21 ramets (mean = 10) located a minimum of 5 m apart (see Fig. S1 in Supporting Information). These four genets were randomly assigned to the four harvest-grazing treatment combinations (harvest and grazing, harvest and no grazing, no harvest and grazing, no harvest and no grazing), yielding a 3 × 3 fully crossed factorial design. A 2 × 2 m plot was demarcated around each focal genet, and all ramets belonging to that genet were tagged. In nearly all cases, palms were sparse enough that each plot contained a single genet and it was clear which ramets made up that genet. Otherwise, ramets separated by more than 25 cm without a visible connection were considered separate genets.

Plots assigned to the no-grazing treatment were enclosed by 2.5 × 2.5 × 1.5 m tall wooden fences to exclude ungulate grazers. Fences were checked monthly for damage, and repairs were made if needed.

All focal genets were flagged, and local harvesters were asked not to harvest from flagged plants. Field assistants from a nearby village harvested palm leaves from plots assigned to the harvest treatment just prior to re-monitoring in 2010 and 2011, which was during the normal harvest season. Consistent with local harvesting practices, field assistants harvested all the leaves from a plant that would have been taken if they were harvesting for commercial broom-making. Usable leaves (> c.60 cm) were cut at the base of the petiole with a machete; leaves that were too small were left intact on the plant.

This study design allowed us to establish a gradient of harvest and grazing intensities, and also to determine how these intensities varied within the environment with abiotic conditions and plant size, as well as covaried with each other and with fire. This provided us with detailed information on the relationships among environmental drivers of demographic variation in a wild palm population.

Data collection

Palms were monitored at the time of the establishment of the experiment in August 2009 (after fire occurred, but before the harvest and grazing treatments were implemented), and again in May 2010 and May 2011. For every ramet, at the start of the study we measured the stem height, and at each census we measured the width of the petiole of longest leaf, the number of grazed leaves, the number of harvested leaves and the total number of leaves (including harvested or grazed leaves that still had green petioles). We used the petiole width of the longest leaf as an indicator of plant size because most palms did not have above-ground stems and because petiole width correlates with survival, growth and reproduction in this (Mandle 2012) and other palm species (e.g. Joyal 1996). For reproductive ramets, we recorded the number of flowering or fruiting stalks produced. During re-monitoring in 2010 and 2011, new vegetative sprouts were tagged and measured, and plots were examined for seedlings. We monitored over 400 palm ramets.

Due to similarities in dentition between domestic and wild ungulates present at the study site, we were unable to differentiate herbivory on palms by livestock from herbivory by wild grazers. Based on a year of camera trapping in the area, 85% ungulates sighted were livestock, compared with 15% wild (L. Mandle, T. Ticktin, unpublished data), suggesting observed grazing is primarily from livestock.

To account for possible environmental differences among plots that could explain differences in palm demographic rates, we measured canopy openness at the start of the study using hemispherical photographs taken 1 m off the ground, analysed with gap light analyzer version 2 (Frazer et al. 1999). Soil moisture (m3 water per m3 soil) was measured on each monitoring date using a Dynamax TH2O Theta soil moisture metre with a ML2 Theta probe by averaging over five points within each plot.

Data analysis

To incorporate both fixed and random effects, as well as the hierarchical nature of the data with covariates measured at different levels, we used linear and generalized linear mixed-effects models (LMM and GLMM). We modelled the variation in harvest and grazing intensities across study plots, as well as variation in palm vital rates (mortality, growth, flowering and vegetative reproduction; Table S1).

Random effects included ramets nested within genets (plots) within split plots (fire treatment) within blocks. Because random effects were based on the spatial configuration of the experimental design, all random effects were retained in all models (Littell et al. 2006). Fixed explanatory variables included the main effects of fire, harvest, grazing and year, as well as covariates – measures of abiotic conditions and of ramet and genet size. For the model of plant growth, a quadratic term for ramet size (petiole width) was added based on checks of model residuals. When harvest and grazing were used as explanatory variables, we used the proportion of leaves harvested or grazed per ramet or genet in the initial full model to account for variation in actual levels of grazing and harvest among plots in the grazing (unfenced) and harvest treatments. We use the terms harvest or grazing treatment level to refer to the plot-level treatment, and harvest or grazing intensity to refer to the proportion of leaves grazed or harvested at the level of the genet or ramet. Year was treated as a fixed effect because it represented time since establishment of the experiment; however, it also includes the effect of inter-annual environmental variation.

We modelled harvest and grazing intensity using data from the 20 plots open to either harvest or grazing. All other models were based on data from all 40 plots. To avoid lack of fit and heteroscedasticity of residuals when grazing intensity was modelled across both years of the study, we modelled grazing intensity separately for 2010 and 2011. Because our experimental harvest of leaves occurred just prior to monitoring, the observed intensity of leaf harvest could not have affected grazing intensities during the previous year. Therefore, we did not include leaf harvest intensity as a predictor of grazing intensity.

Full models were reduced in a backwards stepwise process, sequentially dropping the least significant fixed effect term in the model, testing for significance with likelihood ratio tests with a threshold of = 0.05. After model reduction, we tested for possible non-consumptive effects of grazing and harvest on plant growth (i.e. effects of grazing or harvest beyond that caused by the direct removal of biomass). We did this by adding the plot-level treatments (open to grazing versus fenced and harvest versus no harvest) as explanatory variables, along with two-way interactions between treatment and year, and two measures of ramet size (starting petiole width, and stem height). A significant effect of the plot-level grazing or harvest treatment after accounting for actual levels of grazing and harvest was interpreted as indicating a non-consumptive effect. Limited sample sizes prevented us from testing for non-consumptive effects on rates of mortality or flowering. The two genets with the highest number of ramets occurred by chance in plots without harvest or fire. Plants in these plots had high leverage in the model when the number of ramets per genet was included as an explanatory variable, so this variable was omitted from the initial model of growth and added only after model reduction to test for possible density-dependent effects.

It was not possible to identify the sex of the 18 genets which did not flower during the study; therefore, we modelled the probability of flowering independent of sex. Sex was also not included as a factor in models of other vital rates, as we found no evidence for different rates of growth or survival by sex or interactions between sex and fire, grazing or harvest for the subset of genets with known sex (L. Mandle, T. Ticktin, unpublished data). All analyses were completed in r 2.13.1 (R Development Core Team 2011) using the lmer function in the lme4 package (Bates, Maechler & Bolker 2011).


Intensities of grazing and harvest and relationship to abiotic conditions

Canopy openness and soil moisture varied widely across plots and were not significantly correlated with any type of disturbance overall (details in Appendix S1). Nearly one-third (30%, n = 566) of leaves in unfenced plots were grazed in 2010; a smaller proportion (20%, n = 585) were harvested from plots open to harvest. The amount of grazing dropped in 2011, with 11% (n = 500) of leaves grazed, while both the proportion and number leaves harvested (27%, n = 462) increased (Fig. S2). Across the plots open to grazing, grazing intensity (the probability of a ramet being grazed) varied significantly with fire history, soil moisture, ramet height and the number of leaves per genet in 2010 (Table 1a). Larger plants with taller stems were more likely to be grazed. There was a significant interaction between fire history and soil moisture such that grazing intensities were greater in plots with fire than without fire under relatively moist soil conditions, but lower in plots with fire than without fire in plots with drier soil conditions (Fig. 1). There was no significant effect of soil moisture or number of leaves per genet in plots with fire and no significant effect of fire overall (see Table S2 for estimates and standard errors of non-significant main effects and covariates for all models). In 2011, when grazing levels were lower, ramet size was the only significant predictor of grazing, again with the probability of being grazed increasing with plant size (Table 1b).

Table 1.   Effects of disturbance and abiotic factors on the intensity of grazing on Phoenix loureiri ramets in (a) 2010 and (b) 2011 from binomial generalized linear mixed-effects models
Fixed effectsEstimateSE Z value P-value
(a) 2010
Stem height0.2320.0574.098<0.001
Total leaves per genet−0.0180.031−0.5750.565
Soil moisture (centred)−35.50715.917−2.2310.025
Fire × Soil moisture (centred)39.62018.2552.1700.030
Fire × Total leaves per genet−0.1230.0493−2.5030.012
Random effectsSD   
Split plot (Block)0.000   
Genet [Split plot (Block)]1.101   
Fixed effectsEstimateSE Z value P-value
(b) 2011
Intercept 3.3100.759 4.362<0.001
Starting size (petiole width)2.3601.0362.2790.0226
Random effectsSD   
Split plot (Block)0.190   
Genet [Split plot (Block)]0.914   
Figure 1.

 The probability of Phoenix loureiri ramets being grazed depended on soil moisture (mean between start and end of the annual interval) and fire history (fire in black, no fire in grey) in 2010. Lines are based on the estimated parameters presented in Table 1a. Points (jittered) show the observed pattern of grazing. Probabilities of grazing are graphed at mean values for canopy openness and number of leaves per genet.

Across ramets in harvest treatment plots, the proportion of leaves harvested varied significantly with fire history, grazing intensity, ramet size (largest petiole width), number of leaves per genet, soil moisture and year (Table 2). Harvest increased greatly with ramet size. Overall, increased grazing intensity reduced the proportion of leaves harvested from a ramet, and this was especially the case for ramets in no-fire plots (Fig. 2). As the fire by leaves-per-genet interaction shows, a greater proportion of leaves were harvested from genets with many leaves in plots with fire; the opposite was the case in plots without fire. Based on the year by soil moisture interaction, the proportion of leaves harvested was greater for ramets in drier plots in 2010, but there was no effect of soil moisture in 2011.

Table 2.   Effects of disturbance and abiotic factors on the intensity of harvest of mountain date ramets from a binomial generalized linear mixed-effects model
Fixed effectsEstimateSE Z value P-value
Starting size (petiole width)3.2070.4397.301<0.001
Soil moisture (centred)−11.5933.318−3.493<0.001
Leaves per genet (centred)−0.0310.011−2.6920.007
Grazing intensity−4.5300.521−3.618<0.001
Year (2011)0.1070.1760.6070.544
Fire × Grazing intensity2.6441.3931.8990.058
Fire × Leaves per genet0.0480.0133.608<0.001
Year (2011) × Soil moisture11.3544.1532.7350.006
Random effectsSD   
Split plot (Block)0.000   
Genet [Split plot (Block)]0.049   
Ramet (Genet [Split plot (Block)])<0.001   
Figure 2.

 The proportion of Phoenix loureiri leaves harvested per ramet decreased with increased grazing, especially in plots without fire. Lines show the estimated proportion leaves harvested based on the parameters presented in Table 2 for 2010 at mean values for canopy openness, soil moisture, number of leaves per genet and starting size (0.6 cm). Points (jittered) show the observed proportion of leaves harvested from plants 0.6 cm in size in 2010.

Effects of and interactions among disturbance and abiotic factors on palm vital rates

The palm mortality rate in 2010 was 3.5%, with 14 ramets (of 397) from eight genets dying. In 2011, mortality declined to 1.3%, when five ramets (of 389) died from three genets. Palm ramet mortality was significantly affected by grazing intensity and fire history (Table 3). Mortality was higher among ramets in fire plots. Ramet mortality was greater in genets that experienced higher grazing, especially in the second year of the study. In addition, ramet mortality was higher in genets with more ramets. In contrast, ramets in genets with more leaves had decreased mortality.

Table 3.   Effects of disturbance and abiotic factors on the mortality of Phoenix loureiri ramets from a binomial generalized linear mixed-effects model
Fixed effectsEstimateSE Z value P-value
Ramets per genet0.3880.1143.392<0.001
Total leaves per genet−0.1790.048−3.748<0.001
Grazing intensity0.1011.7930.0570.955
Year (2011)−1.8920.914−2.0700.038
Grazing intensity × year (2011)8.4184.0152.0970.036
Random effectsSD   
Split plot (Block)0.000   
Genet [Split plot (Block)]<0.001   

Fire history, harvest intensity and grazing intensity all had significant effects on palm ramet growth (Table 4, Fig. 3). The effects of harvest and grazing increased with palm size, such that harvest and grazing on average reduced growth of plants above a certain size (greater than c.0.6 cm for harvest and c.0.5 cm for grazing). While the estimated effects of harvest and grazing on smaller plants were positive, actual rates of harvest and grazing experienced by plants of this size were very low so realized effects were essentially zero. In addition, the interaction between fire and harvesting intensity was significant, indicating that increasing levels of harvest reduced growth, especially for plants not recently exposed to fire. Except for the smallest plants, growth was reduced in the second year of the study. When the number of ramets per genet is added to the model presented in Table 4, the number of ramets had a significant negative effect on ramet growth (χ2 = 6.110, d.f. = 1, = 0.013).

Table 4.   Effects of disturbance and abiotic factors on the growth of Phoenix loureiri ramets from a linear mixed-effects model
Fixed effects*EstimateSEChi-squared P-value
  1. *Estimates and standard errors are reported from the model fitted with restricted maximum likelihood. Chi-squared statistics and P-values are from likelihood ratio tests with each parameter removed from the maximum likelihood-based model, with all other parameters retained. It was not possible to test the significance of harvest intensity or the intercept because of the higher-order interaction between fire and harvest intensity.

Starting size (largest petiole width)−0.4100.12211.088<0.001
(Starting size)20.2260.1024.7720.029
Stem height0.0100.00145.337<0.001
Harvest intensity0.5020.110NANA
Grazing intensity0.2190.0729.349<0.001
Year (2011)0.0960.03311.894<0.001
Fire × Harvest intensity0.1040.0523.8760.049
Harvest intensity × Starting size−0.9630.14642.325<0.001
Grazing intensity × Starting size−0.4290.11414.202<0.001
Year (2011) × Starting size−0.3470.05344.242<0.001
Random effectsSD   
Split plot (Block)<0.001   
Genet [Split plot (Block)]0.036   
Ramet [Genet (Split plot (Block)]<0.001   
Figure 3.

Phoenix loureiri ramet growth was reduced with increasing harvest intensity, especially in plots without fire. The negatives effects of harvest and grazing increased with plant size, whereas fire had a consistently positive effect on palm growth. Lines show growth by starting size (0.2, 0.6 and 1.0 cm petiole width) in 2010 as predicted from the model presented in Table 4 with the addition of plot-level harvest and grazing treatment effects. For each starting size, we show the effect of mean grazing and harvest intensities for plants with the mean number of leaves and median stem height.

Sixteen genets flowered in 2010 and 13 in 2011. Eight genets flowered in both years. One genet flowered only prior to applying the harvest and grazing treatments. Nine genets were identified as male and 13 as female. Eighteen did not flower over the three monitoring periods. Flowering increased under more open canopies and declined with increasing grazing intensities (Table 5). None of the six genets with more than 40% of their leaves grazed were observed flowering. Genets in fire plots had a significantly lower probability of flowering in the second year, but there was no effect of year on genets in unburned plots, and no significant effect of fire history on flowering overall.

Table 5.   Effects of disturbance and abiotic factors on flowering of Phoenix loureiri genets from a binomial generalized linear mixed-effects model
Fixed effectsEstimateSE Z value P-value
Canopy openness0.04900.01912.5710.010
Grazing intensity−7.9922.917−2.7400.006
Year (2011)−0.00450.774−0.0060.995
Fire × year (2011)−2.5871.217−2.1250.034
Random effectsSD   
Split plot (Block)<0.001   
Genet [Split plot (Block)]0.817   

There were no new palm seedlings observed in any of the plots over 2 years. Over 2 years, six new vegetative sprouts were produced from six different genets. None of the factors tested were significant predictors of vegetative reproduction.

Additional effects of grazing and harvest

To test for non-consumptive effects of harvest and grazing on plant growth such as from trampling, we added two sets of factors to the model in Table 4: (i) the intensities of harvest and grazing per genet, and (ii) the harvest and grazing treatment levels (e.g. plots open to harvest or grazing versus those that were not). Each set of factors was significant even after accounting for the effect of harvest and grazing intensity on individual ramets (the consumptive effect of harvest and grazing), and therefore explained additional variation in ramet growth rates. Ramet growth declined as the proportion of leaves grazed in the genet increased (χ2 = 4.787, d.f. = 1, = 0.02867); this effect did not vary significantly between years. Ramet growth also declined with an increasing proportion of leaves harvested in the genet, and this effect was greater on plants with taller stems (χ2 = 21.092, d.f. = 2, < 0.001). Ramets in unharvested plots grew more than those in harvested plots, and the benefit of protection from harvest increased with stem height (χ2 = 16.528, d.f. = 2, < 0.001). Ramets in fenced plots grew more than those in open plots in 2010, but not in 2011 when grazing was lower (χ2 = 11.162, d.f. = 2, = 0.004). When both sets of factors (genet-level grazing and harvest intensities and treatment levels) were included as predictors of palm growth in the same model, the grazing treatment level remained marginally significant by likelihood ratio test (grazing + grazing:year interaction, χ2 = 5.295, d.f. = 2, = 0.071), whereas genet grazing intensity was no longer significant (χ2 = 0.594, d.f. = 1, = 0.441). When both harvest treatment level and genet harvest intensity were included in the same model, genet harvest intensity remained marginally significant (χ2 = 4.995, d.f. = 2, = 0.082), whereas harvest treatment level was not (χ2 = 3.217, d.f. = 2, = 0.200).


We find that P. loureiri demographic rates are driven by multiple interactive effects among abiotic conditions, plant size and disturbance and that the intensities of disturbance are themselves driven by interactions between other forms of disturbance and abiotic factors (Fig. 4). We also find evidence of non-consumptive effects of harvest and grazing on palm growth. Our results suggest that understanding and predicting the effects of environmental change on palm demography is only possible when interactive effects are included. Our integration of a manipulative field experiment with data on environmental covariates was integral to untangling the interactive effects we found, but this approach is rare, especially in tropical plant demography. We expect this approach will be valuable to determining the interactive effects among environmental drivers that are likely common to plant demographic responses to disturbance.

Figure 4.

Phoenix loureiri vital rates are affected by disturbance, plant traits, abiotic conditions as well as interactions among these factors. The intensity of grazing and harvest a plant experiences depends on abiotic conditions, plant traits and disturbance. Black arrows show factors significantly affecting vital rates, with dashed lines for mortality, solid lines for growth, and dotted lines for flowering. Grey arrows show factors significantly associated with grazing (solid) and harvest (dashed) intensities. Arrows point from predictor variables to response variables as modelled in this study and should not be interpreted as indicating causality.

Interactive effects shape harvest and grazing intensities

Harvest and grazing did not occur randomly across palm individuals, but instead the intensities of these disturbances were associated with local abiotic conditions as well as the size of the palm ramets and genets. Because broom-making requires palm leaves of at least 60 cm in length, harvest intensity was greater on ramets with larger leaves. There was less harvest from genets with more leaves in no-fire plots, which, given the species’ clumped growth form, may be a result of reduced ease of access to leaves in the middle of large genets. Harvesters report that one benefit of fire is that it burns off the especially spiny older leaves, making harvest easier. Local harvesters favour areas with recent fire because it induces a new flush of brighter green leaves. Increased greenness of vegetation following fire has been shown in other systems (Henry et al. 2006). Fewer leaves were harvested from more heavily grazed ramets, suggesting that herbivory limits harvest. Our finding that this pattern was stronger in plots without fire may be a result of the reduced leaf quality (from the perspective of broom-making) in areas without fire.

The clumped growth form of P. loureiri may also be responsible for our finding that smaller palms may experience reduced grazing because their leaves are less accessible. In contrast to our expectations, and to findings in other ecosystems with both fire and ungulate grazing (Fuhlendorf et al. 2009; Sensenig, Demment & Laca 2010), the intensity of grazing was not significantly higher in plots with recent fire. This may be a result of the small scale of our experiment: within a block, plots with and without fire were separated only by metres or tens of metres, which may be a finer scale than that at which the grazers and herders operate. In a concurrent study, we did detect increased grazing on palm leaves with recent fire across 14 populations separated on the order of kilometres (Mandle 2012), suggesting fire may be a driver of grazing intensity, but only at larger spatial scales.

The increased grazing on genets with more leaves in plots without fire in 2010 was unexpected, but may be because genets with more leaves are more visible to grazers in these areas. The source of the interaction between soil moisture and fire on the intensity of grazing in 2010 is also not clear. The complex patterns of spatial and temporal variation in grazing that we observed emphasize the need for a better understanding of the drivers of this variation, especially because grazing affects multiple palm vital rates. In particular, the cause of the difference in grazing intensity between the two study years is unknown but has important implications for palm demography.

Further investigation of the role of herbivore community composition on plant responses to grazing is also warranted. Different herbivore species may have different effects on plant vital rates depending on their grazing habits (e.g. Holdo, Holt & Fryxell 2009). In addition, the presence of certain grazing species can alter rates of grazing by other species (e.g. Lagendijk, Page & Slotow 2012). Data from our camera traps (L. Mandle, T. Ticktin, unpublished data) indicated that most of the grazing we observed was due to livestock. This might suggest that reducing livestock grazing would benefit palm populations. However, if reduced livestock grazing led to increased grazing by wild ungulates, as has been found elsewhere in South India (Madhusudan 2004), reductions in livestock grazing might not have the anticipated effect.

Effects of and interactions among disturbance and abiotic conditions on palm vital rates

The interaction between fire and harvest on palm growth suggests that the effects of harvest on palms in areas with fire may be less negative than would have been predicted from studying the effects of fire and harvest separately. This interactive effect may result from the tendency of palms in areas with fire to have more leaves than in areas without fire (Poisson GLMM, likelihood ratio test, χ2 = 3.682, d.f. = 1, = 0.055): after having the same proportion of leaves harvested, palms in no-fire areas may be left with fewer intact leaves for photosynthesis. This can be considered an interaction modification effect (Didham et al. 2007), in which the per-unit effect of leaf harvest depends on the fire treatment level. The reductions in growth and increases in mortality in genets with more ramets show that palm vital rates are density dependent. This suggests that disturbances that increase palm mortality – such as grazing – can be in part compensated for by increased survival and growth of remaining individuals (Maron & Crone 2006).

Despite the prevalence of palms in fire-prone vegetation, we know of no studies that compare the demography of burned and unburned palms (but see Souza & Martins 2004 for a study of palm demography before and after fire). We find that fire history affected multiple vital rates of the P. loureiri in contrasting ways. Palms in plots with recent fire had higher rates of mortality, but surviving individuals had higher rates of growth. The population growth rates of understorey palms and other long-lived species are generally more sensitive to rates of survival than growth, though the relative sensitivity may be size dependent (Franco & Silvertown 2004; Zuidema, de Kroon & Werger 2007). This suggests that the increases in ramet growth following fire are likely to mitigate – but not fully compensate for – increased mortality in P. loureiri. Our results show significant impacts of fire up to 2 years after the fire event; however, data on the direct effects of fire events on palm vital rates are still required to fully understand the implications of fire for population dynamics. Although this has not been studied in palms, multi-year effects of fire on plant vital rates have been found in other plant species in fire-prone systems (Hartnett & Richardson 1989; Quintana-Ascencio & Morales-Hernández 1997).

The effects of grazing on palm vital rates were almost uniformly negative. Grazing – which occurred primarily on larger plants – reduced ramet growth rates, with a greater negative effect on larger plants. The increased effect of grazing on larger palms is consistent with findings from a neotropical dioecious understorey palm (Endress, Gorchov & Noble 2004). Also consistent with these previous findings, grazing in our study was associated with reduced P. loureiri flowering and, in the second year, with increased mortality. The increased mortality with increased grazing in 2011, when overall levels of grazing were lower, could be a result of the drier conditions in that year.

Except for the smallest plants, which did not produce leaves large enough to be harvested, harvest reduced ramet growth, with increasingly negative effects on larger plants. The lack of a significant effect of harvest on reproduction is surprising, as reduced vegetative and sexual reproduction due to leaf harvest has been found in other palm species (e.g. Ratsirarson, Silander & Richard 1996; Endress, Gorchov & Berry 2006; Zuidema, de Kroon & Werger 2007; Martínez-Ramos, Anten & Ackerly 2009). The lack of an effect of leaf harvest on palm mortality is not unexpected: other studies of palm demography have detected effects only at the highest intensities of harvest (Endress, Gorchov & Berry 2006; Zuidema, de Kroon & Werger 2007). However, we note that a lack of statistical significance cannot be interpreted as an acceptance of the null hypothesis of no true effect. Here, the lack of statistically significant effects of harvest and other factors and interactions could also be due to limited power given our relatively small sample size and a potentially small true effect size. Lack of significance could also be due to the limited duration of our study. With the long history of harvest at the site, more than 2 years of protection from harvest may be required for differences in rates of flowering or mortality to become apparent.

Increased flowering under more open canopies, as we found for P. loureiri, is common among understorey palm species (de Steven 1989; Cunningham 1997). Because the open canopies of the ecosystems where P. loureiri occurs are partly maintained by frequent fires that reduce tree recruitment (Ratnam et al. 2011), fire may in this way indirectly benefit palm populations. In contrast with our expectations, the effects of fire, grazing and harvest did not vary significantly with light availability. The significant effect of year and its interactions with disturbance and abiotic factors suggests that additional demographic drivers in this system remain unaccounted for.

Grazing limits palm leaf harvest

While we did not find a significant interactive effect between harvest and grazing intensity on the vital rates of palm individuals, the negative correlation between actual rates of harvest and grazing intensity suggests that the overall effect of harvest depends on the rates of grazing. This form of interaction between harvest and grazing represents an interaction chain effect (Didham et al. 2007), in which the per-unit effect of leaf harvest is constant across levels of grazing, but the total amount of leaves harvested is less in areas with grazing than without. As a result, and in contrast to our expectations, harvest then likely has less of an effect on palm populations in areas with grazing. As more leaves are grazed, fewer intact leaves remain that can be harvested. This subadditive effect of harvest and grazing is similar to that found for American ginseng, where deer grazing reduces plant mortality due to harvest by hiding plants from harvesters (Farrington et al. 2009). Given that P. loureiri leaf harvest across South India occurs in areas with variable amounts of grazing, and because rates of grazing and harvest are linked, understanding the impacts of harvest in this system necessitates understanding patterns of grazing as well. Acquiring a better understanding of the interactions between harvest and grazing is likely to be important for many other species, including other palms (e.g. Berry, Gorchov & Endress 2011; Lopez-Toledo, Horn & Endress 2011), that are subjected to both harvest and grazing simultaneously.

Additional effects of grazing and harvest

A common approach to simulating the effects of ungulate herbivory in plant demographic studies has been to compare demographic rates across all individuals within a grazed population to demographic rates of the subset of individuals within the same population that have escaped herbivory (e.g. Knight 2004; McGraw & Furedi 2005; Farrington et al. 2009). This approach does not account for any non-consumptive effects associated with herbivory, such as trampling or soil compaction. We find evidence that protection from grazing and harvest increases the growth even of P. loureiri individuals that are not directly grazed or harvested. These non-consumptive effects could be physical, if trampling affects ramets in genets that have been harvested or grazed, even when biomass is not removed from the ramet. These effects could also be physiological – resulting from increased carbon export to, or reduced carbon subsidies from, other grazed or harvested ramets within the same genet, as has been found in other clonal species (e.g. Chapman, Robson & Snaydon 1992; Zhang, Yang & Dong 2002). We were unable to test for these possible effects independently within our study – while some genets in plots open to grazing or harvest did not have any leaves grazed or harvested, it is likely that these genets also would have experienced fewer physical effects as well. However, we found that grazing treatment level (fenced or open) explained additional variation in ramet growth, even after including actual levels of grazing on the ramet and genet in the model, which would have accounted for physiological effects. This suggests that physical effects like trampling reduce ramet growth in addition to the direct consumptive effects of herbivory from biomass removal.

Our results suggest that studying herbivory using ungrazed plants within a grazed population can lead to inaccurate conclusions about the effects of this form of disturbance from at least two sources. First, this approach can bias results if the conditions that led to ungrazed plants escaping herbivory (such as smaller size or reduced leaf quality) are also associated with differences in vital rates for those individuals. In our study, smaller palms – the ones most likely to escape grazing – grew more than larger palms, even after controlling for differences in grazing, which would lead to an overestimation of the direct effects of grazing. Secondly, this approach will underestimate the negative impacts of grazing if, as we found, ungrazed plants are still affected by non-consumptive effects of grazing. The likely role of physical effects in our study, combined with findings of negative impacts of deer herbivory even on unpalatable plant species (Heckel et al. 2010), suggests that negative non-consumptive effects of herbivory are common even in non-clonal species. Studies using simulated herbivory such as clipping (e.g. Oba, Mengistu & Stenseth 2000; Vandenberghe, Freléchoux & Buttler 2008) may also underestimate the effects of herbivory if the non-consumptive effects of simulated herbivory differ from those of actual herbivores.

The importance of interactions among drivers to understanding plant demographic processes

Most studies of plant demography to date have been observational (Crone et al. 2011), and this is especially the case in tropical ecosystems. Very few experimental studies exist of interactions among multiple drivers on plant demography. Our use of a manipulative experiment integrated with measurements of disturbance intensity and abiotic factors was critical to disentangling multiple pathways by which environmental drivers can affect plant vital rates.

The prevalence of interactions we found that influenced both the intensity and the outcome of disturbance at the ramet and genet levels suggests that interactions among drivers are likely to be important for P. loureiri at the population level as well. As fire, livestock grazing and wild plant harvest are very common co-occurring forms of anthropogenic disturbance throughout the tropics, our findings from P. loureiri can provide insights into how other tropical species are likely to respond to similar disturbances.

With the short time frame of our study, the high interannual variation in grazing intensity we observed and the likelihood of temporal variation in, and interactions between, other factors such as soil moisture and fire, we expect that long-term palm population dynamics are also affected by interactions not detected here and that interactive effects themselves vary temporally. With growing evidence to suggest that interactions among drivers are the norm rather than an exception (e.g. Schleuning, Huamán & Matthies 2008; Martínez-Ramos, Anten & Ackerly 2009; Hegland, Jongejans & Rydgren 2010), further research is needed to determine in which situations the interactive effects of environmental drivers are most important to understanding and predicting plant demography. The number of possible interactions in any system is large and attempting to study them all simultaneously is neither feasible nor desirable. This is especially the case in human-managed tropical ecosystems, where interacting forms of disturbance commonly co-occur and where understanding and mitigating species declines is particularly important to conserving remaining biodiversity (Chazdon et al. 2009). Further studies focusing on the effects of drivers and their interactions on high-sensitivity vital rates and on drivers likely to vary over spatial or temporal scales of interest will be critical to meeting this need.


We thank Mahadesh, Mahadevan and Basavanna for field assistance; Anita Varghese and Keystone Foundation staff for providing critical insights and logistical support; and the Tamil Nadu Forest Department for their cooperation. We are grateful to Curtis Daehler, Rebecca Ostertag, Andrew Taylor, Journal of Ecology editors and two anonymous referees for comments that greatly improved this manuscript. This work was supported by a U.S. National Science Foundation (NSF) Graduate Research Fellowship under Grant No. 0822443 to L.M. and NSF Doctoral Dissertation Enhancement Project Grant No. OISE-1104989 to T.T. & L.M., as well as support from the University of Hawai‘i Botany Department and Ecology, Evolution and Conservation Program and NSF DGE05-38550 to K.Y. Kaneshiro.