El Niño impacts on human‐modified tropical forests: Consequences for dung beetle diversity and associated ecological processes

Our knowledge of how tropical forest biodiversity and functioning respond to anthropogenic and climate‐associated stressors is limited. Research exploring El Niño impacts are scarce or based on single post‐disturbance assessments, and few studies assess forests previously affected by anthropogenic disturbance. Focusing on dung beetles and associated ecological functions, we assessed (a) the ecological effects of a strong El Niño, (b) if post‐El Niño beetle responses were influenced by previous forest disturbance, and (c) how these responses compare between forests impacted only by drought and those affected by both drought and fires. We sampled 30 Amazonian forest plots distributed across a gradient of human disturbance in 2010, 2016, and 2017—approximately 5 years before, and 3–6 and 15–18 months after the 2015–16 El Niño. We found 14,451 beetles from 98 species and quantified the beetle‐mediated dispersal of >8,600 seed mimics and the removal of c. 30 kg of dung. All dung beetle responses (species richness, abundance, biomass, compositional similarity to pre‐El Niño condition, and rates of dung removal and seed dispersal) declined after the 2015–16 El Niño, but the greatest immediate losses (i.e., in 2016) were observed within fire‐affected forests. Previous forest disturbance also influenced post‐El Niño dung beetle species richness, abundance, and species composition. We demonstrate that dung beetles and their ecological functions are negatively affected by climate‐associated disturbances in human‐modified Amazonian forests and suggest that the interaction between local anthropogenic and climate‐related stressors merits further investigation.


| INTRODUC TI ON
El Niño Southern Oscillation (ENSO) events are becoming increasingly frequent and severe (Timmermann et al., 2018), potentially due to human-induced climate change (Cai et al., 2014). Over the last 40 years, strong ENSO events have been associated with changes in precipitation patterns in tropical rain forests, leading to extreme droughts and forest fires (Juárez-Orozco, Siebe, & D. Fernández y Fernández., 2017).
The frequency and extent of forest fires are further exacerbated by decadal-scale increases in dry season lengths (Fu et al., 2013), the spread of fire-dependent agriculture, and the increases in forest flammability that results from human-driven disturbances such as logging and fragmentation (Hardwick et al., 2015;Uhl & Kauffman, 1990).
Despite advances in our knowledge of El Niño consequences for the carbon cycle Malhi et al., 2018), our understanding of the impacts for biodiversity and related ecosystem functions remains limited, especially in human-modified tropical forests. While detailed studies have been conducted following experimental fires (e.g., ≤50-ha forest plots; Balch, Massad, Brando, Nepstad, & Curran, 2013;Brando et al., 2014;Oliveras et al., 2014), these may underestimate the effects of large-scale megafires which can affect millions of hectares (e.g., Withey et al., 2018). In these fires, burned forests may be tens of kilometers away from source populations in unburned forests or isolated by a matrix of agricultural land-uses. Furthermore, large-scale fires may have higher fire intensities as severe droughts also result in a drier fuel layer  and increased fuel loads (Brando et al., 2008).
Where large-scale studies assessing the ecological consequences of wildfires have taken place, they often lack pre-fire information (e.g., Barlow & Peres, 2004) and rely on space-for-time approaches that may underestimate biotic changes in tropical forests (Christie et al., 2019;. Furthermore, it is not clear how previous anthropogenic forest disturbance, such as selective logging, influences the response of biodiversity and associated functions to ENSOmediated droughts and wildfires, or whether changes in biodiversity result in further changes in ecosystems functioning. Addressing these knowledge gaps is critically important given the increased likelihood of severe dry seasons (Duffy, Brando, Asner, & Field, 2015) and the increased rates of human-driven forest modification that is expected for tropical regions (Lewis, Edwards, & Galbraith, 2015).
We address these knowledge gaps by evaluating changes in biodiversity and some ecological processes in a region of the Amazon affected by a mega fire and intense drought in the 2015-16 El Niño event ( Figure 1). We focus on dung beetles (Coleoptera: Scarabaeinae), because they (a) had been sampled in >200 forest plots in 2010, 5 years before the 2015-16 El Niño, encompassing a gradient of pre-El Niño forest disturbance, from undisturbed primary forests to logged primary forests and logged-and-burned primary forests Gardner et al., 2013); (b) are a cost-effective indicator group ; and (c) perform a number of important ecological functions that can be readily assessed in the field (e.g., França, Louzada, & Barlow, 2018;Nichols et al., 2008;. We returned to 30 of these forest plots between 3-6 and 15-18 months after the ENSO drought and related wildfires, repeating the sampling techniques used in 2010. This design allowed us to have a full-factorial design (Table S1) to assess the following questions:

| Study region
We conducted our study in three municipalities in the Brazilian Amazon: Belterra, Santarém, and Mojuí dos Campos in the state of Pará ( Figure 1). The climate in this region is characterized as hothumid (Köppen's classification), and the annual average temperature and precipitation are 25°C and 1,920 mm, respectively, with short dry seasons between August and October ( Figure S1), which are longer and drier during El Niño years (Jolly et al., 2015).

| Sampling design
We sampled dung beetles and beetle-associated ecological functions within 30 forest plots (Figure 1) distributed along a pre-El Niño disturbance gradient, including undisturbed primary forests (n = 10), logged primary forests (n = 10), and logged-and-burned primary forests (n = 10). Between October and December 2015, half of these forests plots were impacted by understory fires that occurred during the exceptionally dry weather caused by the extreme 2015-16 El Niño event, while unburned controls were preserved in all of our previous forest disturbance classes (hereafter fire-affected and drought-only forests, respectively; Table S1).

| Data collection
Dung beetles and ecological functions were surveyed in exactly the same locations ( Figure 1) and following the same sampling techniques in all three surveys ( Figure S2). The first data collection occurred in June-July 2010, around 5 years before the 2015-16 El Niño. The second and third surveys took place in June-July 2016 (as in 2010, end of the rainy season; Figure S1) and March-April 2017 (in the rainiest months in the study region; Figure S1), approximately 3-6 and 15-18 months after the El Niño-associated uncontrolled fires that affected our study region. At each of the 30 forest sites, beetles and their ecological functions were sampled at three sampling points (0, 150 and 300 m) along a 300-m transect. We used nine dung-baited pitfall traps (three traps per sampling point; Figure S2c) to sample dung beetles, resulting in a total of 810 dung-baited pitfall traps (270 pitfalls/year). All trapped dung beetles were identified to species or morphospecies. Dung beetle species-level average body mass was calculated from the dry body weight of 15 individuals using a Shimadzu balance with precision of 0.0001 g. Rates of dung removal and secondary seed dispersal were assessed between 3 and 4 weeks after the dung beetle surveys. At each sampling point, we placed a mesocosm arena with ~0.79-m 2 area ( Figure S2a,b) and containing, in the centre, a 200-g dung pile (4:1 pig to human ratio, following França et al., 2018) mixed with 50 seed mimics (3.5-mm diameter, as in Braga et al., 2017;Braga, Korasaki, Andresen, & Louzada, 2013). Further methodological details are described in Supporting Information.

| Statistical analyses
To address our research questions, we examined changes in six response variables: species richness, abundance, species composition, biomass, and rates of dung removal and seed dispersal. All analyses were performed within the R Studio version 3.3.1 (R Core Team, 2019) and conducted at the plot-level: community attributes (richness, abundance, and biomass) were the sum of values from each of the nine pitfall traps, and rates of ecological functions were the average of the values recorded in the three arenas.
Dung beetle biomass at the plot-level was calculated by multiplying the average body mass of each species by their abundance.
Species composition was measured as pairwise beta-diversity (Socolar, Gilroy, Kunin, & Edwards, 2016), based on the Bray-Curtis similarity index (1-dissimilarity) calculated for each forest plot and year through the vegdist function ("vegan"; Oksanen et al., 2015). Post-El Niño species composition therefore represents the F I G U R E 1 Map showing the location of our study region in the eastern Brazilian Amazonia. (a) Around 8,072 km 2 of primary forests from the total area in the map (ca 27,418 km 2 ) were burned during the 2015-16 El Niño event. The inset shows the study region (light green) within Brazil (light gray) and state of Pará (dark gray). (b) The map within the study region-shown by the gray border in (a). Also shown in these panels are the locations of the 30 sampled forest plots (beige-filled circles) We conducted a three-way full-factorial (Table S1)   were statistically significant, we therefore used Tukey's Wholly Significant Difference through the "lsmeans" package to assess post hoc differences (Lenth, 2016). Plots were generated by using the ezPlot function ("ez"; Lawrance, 2016) and subsequently modified using "ggplot2" (Wickham, 2009).
Gaussian distributions for all response variables and model residuals were tested using the Shapiro-Wilk normality test through the Shapiro.test function ("stats"; Crawley, 2002). Data normality and homoscedasticity were achieved for biomass, abundance, and dung removal rates after rank-transformation. We used the package "dplyr" for data cleaning and the function pearson.test in "stats" to assess the residual independence from all RM-ANOVAs (Table S2). As sites that are closer together are expected to hold more similar communities (Kühn & Dormann, 2012), we also assessed spatial autocorrelation within our datasets by performing Pearson-based Mantel tests using the mantel function with 1,000 permutations ("vegan"; Oksanen et al., 2015). Mantel tests were made separately for dung beetle richness from each survey. Mantel tests of distance between forests showed a weak but signifi- by El Niño fires were still lower than the pre-El Niño condition (ttest, t-ratio = 3.6, p = .04; Figure S3). We discuss these results in light of the ecological consequences that interactions between climatic and local stressors can bring to tropical forest biodiversity and ecological functions.

| Direct drivers of change in dung beetle communities
We provide evidence that El Niño-induced droughts and fires can be strong direct drivers of change in dung beetle communities.
Although spatial patterns of movement can vary among species (Silva & Hernández, 2015), most dung beetles are poor fliers-for example, having an estimated movement of 90 m in 48 hr (Silva & Hernández, 2015) and may not be able to escape understory fires; F I G U R E 2 Dung beetle responses to El Niño-induced drought and fires in previously undisturbed and human-modified Amazonian forests. (a-c) Dung beetle species richness, (d-f) abundance, (g-i) compositional similarity, (j-l) biomass and rates of (m-o) dung removal, and (p-r) secondary seed dispersal were sampled within 30 forest plots (n = 5 plots per forest type and El Niño class) in the eastern Brazilian Amazon region, near Santarém in the State of Pará. Surveys were carried out in 2010 (i.e., pre-El Niño survey) and in 2016 and 2017around 3-6 months and 15-18 months after the 2015-16 El Niño fires affected half of these forest plots. Models were repeated-measures ANOVA treating "Year" as the repeated measure, and "forest classes" and "El Niño classes" as grouping factors for each response variable. To facilitate post hoc visual comparisons within the analyzed data, error bars depict Fisher's least significant difference on the three-way interaction (ezPlot, by default; Lawrance, 2016); thus non-overlapping bars can be interpreted as being significantly different  There is, however, a lack of understanding to what extent belowground nests are affected by droughts or fires. Although temperatures at the soil surface can be extremely high even during low-intensity fires (Kennard & Gholz, 2001), these decrease abruptly with increasing depth-for example, reaching only 22-25°C between 22 and 30 cm in depth after burning for 2 hr (Beadle, 1940). But most Amazonian dung beetles nest in shallower soil layers (Griffiths et al., 2015), raising the possibility that mortality within nests contributes to the immediate post-El Niño declines in dung beetle communities.
Alternatively, the belowground environment may represent a refuge for dung beetles (both larvae and adults) nesting in deeper soil layers (Gregory, Gómez, Oliveira, & Nichols, 2015;Griffiths et al., 2015), and post-drought and fire emerged beetles could help explain the time lag between the El Niño and dung beetle responses two years later.

| Indirect mechanisms underpinning post-El Niño changes in dung beetle communities
With some exceptions (e.g., Barlow et al., 2016;Cleary & Mooers, 2006), the current literature on drought-and fire-induced impacts on tropical forests is dominated by plant studies (e.g., Berenguer et al., 2018;Brando et al., 2008;Silva et al., 2018), which show increased tree mortality (Nakagawa et al., 2000), reduced carbon storage , and large physiological changes such as in flower and fruit production (Sakai et al., 2006). Given the scale of effects observed among the primary producers, it seems likely that invertebrate taxa would also be affected. Two obvious mechanisms could underpin these changes. First, lower post-El Niño fruit production can result in large-vertebrate famine Wright, Carrasco, Calderon, & Paton, 1999). These drought and fire-induced changes in vertebrate communities (Barlow, Peres, Henriques, Stouffer, & Wunderle, 2006;Peres, Barlow, & Haugaasen, 2003) are likely to result in cascading effects on dung beetles (Nichols, Gardner, Peres, & Spector, 2009), as co-declines in these two groups have been reported in other human-modified tropical forest landscapes (e.g., Bogoni, Silva, & Peres, 2019;. Second, high rates of tree mortality following droughts and forest fires result in more open canopies, which may affect communities through the hotter and drier forest microclimates (Brando, Oliveria-Santos, Rocha, Cury, & Coe, 2016;Hardwick et al., 2015).
Tropical dung beetles have been shown to respond to such forest modification, both indirectly through sublethal changes on their body conditions Salomão, González-Tokman, Dáttilo, López-Acosta, & Favila, 2018) and directly, by reducing species-specific relative abundances and the community diversity and biomass .

| Could pre-El Niño forest disturbance influence post-El Niño ecological communities?
Previous forest disturbance influenced the post-El Niño declines in dung beetle richness, abundance, and compositional similarity to pre-El Niño condition, while previously logged-and-burned forests had slightly faster recovery times for some dung beetle responses between 2016 and 2017. Both these findings suggest that previous forest disturbance could have acted as an environmental filter or selective force (Balmford, 1996;Nunes et al., 2016): Microclimatic changes relating to previous logging or fires (e.g., Hardwick et al., 2015;Lindenmayer, Hunter, Burton, & Gibbons, 2009) may have extirpated the most disturbance-sensitive species-as previously observed for dung beetle communities in primary forests converted to oil palm plantations in Southeast Asia (Edwards et al., 2014)-and favored the species that are more tolerant to drought and fires. This conjecture is supported by evidence showing that dung beetle species are highly susceptible to environmental modification (Beiroz et al., 2018), including changes in forest structure (Salomão et al., 2018) and microclimatic conditions (Birkett, Blackburn, & Menéndez, 2018). Perhaps more importantly, our results support that local human-driven disturbance and climateassociated stressors can act together and influence tropical forest biodiversity and functioning. Thus, focusing on a single stressor may fail to capture the magnitude of the threat faced by tropical forests and their fauna Newbold et al., 2019), which are increasingly threatened by local human-driven disturbances (Lewis et al., 2015) and are expected to have more frequent and extreme droughts in the next decades (Duffy et al., 2015).

| Exploring the resilience of dung beetlemediated processes
The lack of influence of pre-El Niño forest disturbance on beetlemediated processes is consistent with previous studies showing the disturbance resilience of invertebrate-mediated processes in tropical forests França et al., 2018) and confirms that community and functional attributes may be asymmetrically affected by human activities in tropical forests (Braga et al., 2013;Carvalho et al., 2020;França et al., 2018). However, forest structure is known to be a key determinant of insect communities (Basset, Charles, Hammond, & Brown, 2001), and there is presumably a threshold at which point changes in forest structure are so great that invertebrate communities and mediated processes are also affected (e.g., França, Frazão, Korasaki, Louzada, & Barlow, 2017). The lower rates of dung removal and seed dispersal in post-El Niño surveys (Figure 2m-r) suggest this threshold was surpassed by the severe impacts of the 2015-16 El Niño drought and wildfires on vegetation and forest structure in this region Silva et al., 2018;Withey et al., 2018). These lower rates of ecological functions-which occurred within both drought-only and fire-affected forests-could be attributed to the higher vulnerability of large-bodied dung beetles to forest disturbance (Larsen, Williams, & Kremen, 2005). For example, we found 61 individuals of Coprophanaeus lancifer (Linné, 1767) during the pre-El Niño survey in 2010, while only 11 and 5 specimens were sampled in 2016 and 2017, respectively. This is the largest dung beetle species found in the study region and belongs to one of the most important functional groups (large tunnelers) performing ecosystem functions of soil bioturbation (Gregory et al., 2015) and dung and seed removal (Slade, Mann, Villanueva, & Lewis, 2007).

| Dung beetle responses between El Niño drought-only impacted and fire-affected forests
We However, drought effects on dung beetles were surprisingly strong considering that tree mortality in drought-affected forests is only 1%-3% (Phillips et al., 2010) compared with 50% or more in drought and fire-affected forests (e.g., Barlow et al., 2012).

| Research limitations
While our findings are likely to reflect the short-term sensitivity of tropical invertebrates and associated processes in human-modified forests to El Niño drought and fires, they are not without limitations.
One key issue is that our pre-El Niño plots were sampled nearly 5 years before the event. Although assessments of the vegetation suggest minimal differences between 2010 and pre-El Niño samples in 2016, at least some of the influence of El Niño on invertebrate communities may be obscured by pre-El-Niño changes in biodiversity through processes such as succession (e.g., Lennox et al., 2018), longer-term disturbance responses (e.g., Silva et al., 2018), or ecological drift and competition (e.g., Levi et al., 2019;Ulrich, Puchałka, Koprowski, Strona, & Gotelli, 2019). As such, while before-after studies hold many advantages when the before assessment is immediately before the disturbance (e.g., Christie et al., 2019;, there is a risk that the ecological signal will become degraded with greater temporal disconnection. Another potential limitation relates to seasonality. We sampled dung beetles at the end of the rainy seasons in 2010 and 2016, and in the rainiest months in 2017 (March-April; Figure S1). Thus, our research does not take seasonality into account, which can play a significant role in dung beetle responses to forest disturbance (Andrade et al., 2011; but see Gardner, Hernández, Hernández, . Furthermore, the post-El Niño samples only extended to 15-18 months after the fires occurred in our forest plots, and patterns may be dominated by non-equilibrium processes, including high levels of instability and stochasticity, which often dominate the short-term responses of ecological communities to disturbance (Mori, 2011) and can result in black-swan events in animal populations (Anderson, Branch, Cooper, & Dulvy, 2017). As our assessments are restricted to a single taxon and their associated functions, more long-term research is therefore needed to understand how reproducible our results are in other taxa performing key ecosystem processes in tropical forests, such as ants (Griffiths et al., 2018), termites (Ashton et al., 2019), and seed-disperser vertebrates . This would foster a better knowledge of how resilient tropical forest fauna and ecosystem functioning are to the interactions between human-and climate-associated stressors.

| CON CLUS IONS
By exploring the impacts of the 2015-16 El Niño on dung beetle communities and associated ecological processes across a gradient of previous forest disturbance in the Amazon, we confirm the threat posed by extreme drought and fire events for biodiversity and functioning of human-modified tropical forests. We found that undisturbed forests were more sensitive than logged-and-burned forests for most dung beetle responses, and that El Niño drought alone and/ or combined with fires can result in drastic losses in beetle diversity, abundance, biomass, and rates of dung removal and seed dispersal that can last for at least 18 months. Our results, therefore, suggest that local human-driven disturbances and climate-associated stressors can interact in different ways and that these interplays may asymmetrically affect the community and functional attributes of tropical forest invertebrates. However, future investigations, with more tightly controlled pre-disturbance conditions and longer-term tracking of recovery, are needed to better understand the interactions between multiple forest stressors.

ACK N OWLED G M ENTS
We are grateful to all the farmers for collaborating to the Long-Term

DATA AVA I L A B I L I T Y S TAT E M E N T
The data used in this study are archived at the Environmental Information Data Centre (NERC-EIDC; https ://doi.org/10.5285/ 799db 965-3ce7-4e9b-8590-de6a8 624d652).