2.1 Study site

The study was conducted in a 130-ha study plot at the Bilsa Biological Station (hereafter BBS; 7900450W, 000220N; 436–615m elevation in our study site), a 3500-ha reserve of humid Chocó rain forest in northwest Ecuador (Figure S1). Average monthly temperatures at our study site vary between 26 and 28°C, while rainfall is markedly seasonal, with a 5-month dry season between July and November characterized by relatively low precipitation but persistent cloud cover.

2.2 Study species

Oenocarpus bataua is a hyperdominant, slow-growing, monoecious, protandrous Neotropical palm species reaching up to 35m in height (Henderson, 1995; ter Steege et al., 2013). O. bataua produces mass inflorescences over 2m long, with over 200 racemes bearing more than 300 flowers each. Following the appearance of an inflorescence bud, flowers take an average of 5 months to reach anthesis (Núñez-Avellaneda & Rojas-Robles, 2008; Rojas-Robles & Stiles, 2009). Individuals can bear multiple inflorescences simultaneously, and fertilized flowers develop large, lipid-rich, single-seeded drupes, comprising large crops of over 2000 fruits per individual. Fruits take approximately 13 months to develop following pollination (Rojas-Robles & Stiles, 2009). The abundance and nutritional content of its fruits and seeds make O. bataua's phenology an important determinant of food availability for large-bodied vertebrates, including humans (Henderson, 1995).

2.3 Phenological monitoring

We identified all adult O. bataua trees within our study plot (n = 178; Figure S1) and evaluated their reproductive status in monthly censuses (May 2008 to April 2017) where we counted the number of developing inflorescences, mature inflorescences (presence of open flowers), developing infructescences, and mature infructescences (presence of mature fruits) in each tree. We did not differentiate between male and female flowering phases. We added 64 individuals to our monitoring effort as we gradually expanded the study plot from 30ha to 130ha between January and September of 2010, and another 12 new individuals that reproduced for the first time through 2015.

2.4 Tree height and crown emergence

During a single census in 2013, we recorded the height of each Oenocarpus bataua tree in our study site by measuring the distance from its base to the tallest point of its crown using a laser range finder. Following the same procedure, the forest canopy height around each focal individual was measured as that of the tallest tree within a 10-m radius from the focal individual. We then assessed “crown emergence” for each focal tree by comparing its height with that of its surrounding canopy. A tree was categorized as “emergent” if taller than the canopy around it, and “non-emergent” otherwise. Because the presence of a taller tree within a 10-m radius does not necessarily guarantee that the focal tree will experience reduced access to sunlight, this approach has the limitation of underestimating the “true” number of focal trees with “emergent” crowns. However, this limitation only increases the probability of type II errors (“false negatives”), making our “emergence” metric a conservative estimate suitable for analysis.

2.5 ENSO and local climate

We evaluated whether ENSO plausibly drives climate variability in our site by correlating publicly available temperature, precipitation, humidity, and cloud cover data from TerraClimate (Abatzoglou, Dobrowski, Parks, & Hegewisch, 2018) and CRU TS v4.01 (Kobayashi et al., 2015), and the three ENSO indices most commonly employed to define El Niño/La Niña events: the Oceanic Niño Index (ONI), the Southern Oscillation Index (SOI), and the Multivariate ENSO Index (MEI) (https://www.esrl.noaa.gov/psd/enso, Smith, Reynolds, Peterson, & Lawrimore, 2008) (see Appendix S1 for a detailed description of the methods). Overall, El Niño conditions were significantly correlated with anomalously high temperature and vapor pressure levels in our study region, and moderate increases in rainfall and decreases in cloud cover (Table S1).

2.6 Individual-level analyses

Given O. bataua's long reproductive cycle, trees observed for fewer than 70 months were excluded from individual-level analyses to avoid unrepresentative phenological records, resulting in a sample size of 153 individuals. We calculated inflorescence production as the total number of inflorescences initiated by each tree throughout the study divided by the number of years of observations for that tree. We then assessed the effects of tree height and crown emergence on the average number of inflorescences initiated annually (making each tree a single data point, n = 153) using a multiple linear regression formulated as follows:

(1)

The model included an interaction term of tree height and crown emergence (Height x Emergence), whose coefficient represents the difference in the effect of tree height for trees with emergent crowns compared to trees with non-emergent crowns. All predictors were centered at 0 and standardized (SD = 1) to aid comparison of effect sizes (Schielzeth, 2010) (summary statistics in Table S2). O. bataua's developing inflorescences are frequently depredated by insects, which can result in abortion (Pedersen & Balslev, 1992). In our study population, abortion rates averaged approximately 60 percent among individuals, and we estimate that an average of 35 percent of inflorescences per tree were attacked by insects (data not shown). However, if inflorescence abortion affected individual-level inflorescence production, we would expect abortion rates to be correlated with inflorescence initiation rates, yet we found no such association (r = −0.17, NS). Moreover, restricting the analysis to include only inflorescences that successfully flowered did not qualitatively affect our results.

To assess whether the presence of developing fruits affected the probability of producing new inflorescences among individuals, we excluded all trees that never produced developing infructescences over the study period, resulting in a sample of 122 individuals. We modeled the probability of inflorescence production each month across all individuals using a generalized mixed-effects model (GLMM), with a binomial distribution for the response and a logit link function for the linear component of the model. The model was formulated as follows:

(2)

We used production/no production of new inflorescence by focal individual i during month j as a binary response (n = 10,365), and the presence of developing infructescences in individual i in month j as a fixed effect factor. To test for temporal autocorrelation in the response, we computed partial autocorrelation correlation coefficients for 30 lags in the inflorescence production time series of all individuals in our sample, finding no significant patterns (see Figure S2 for details). Additionally, tree ID and year (indexed through individual i and year k, n = 122 and n = 10, respectively) were included as random intercepts, where and . Parameter estimates were generated using Laplace approximation, and the model was implemented using the “lme4” package version 1.1–21 in R (Bates, Mächler, Bolker, & Walker, 2014).

We assessed seasonal constraints in reproduction by testing for clustering of reproduction within a year at the individual level. To do this, we computed 95% confidence intervals of the expected number of months in which a given number of reproductive events would occur if flowering onset was equally likely each month of the year (see Figure S3 for details and a schematic description of the process). We complemented this analysis by evaluating how the probability of inflorescence production among individuals (n = 153) varied by month of the year (January–December) using a GLMM with production/no production of inflorescences each month as a response, month of the year as a fixed effect factor, and tree ID as a random effect (see Table S3 for details).

2.7 Population-level analyses

We calculated the proportion of trees in each phenophase every month between May 2008 and April 2017. Prior to analyzing the effects of ENSO anomalies on population-level flowering phenology, we temporally downscaled our data from monthly observations (n = 108) to non-overlapping trimesters (4-month period averages, n = 27) for two reasons. First, the monthly proportion of trees bearing developing inflorescences showed a complex autocorrelation structure with significant lags of up to 20 months. Decreasing the temporal resolution of our data yielded significant partial autocorrelation coefficients only for the first two lags. Second, monthly variation in ENSO indices showed weaker correlations with climate than did 4-month periods (Table S1). Because the proportion of trees producing new inflorescences was strongly autocorrelated between consecutive months (r = .92), we imputed missing observations in December 2008, October and November 2012, December 2014, and April 2015, by assuming a linear rate of change between two months of observation bounding a missing value. We then computed the non-overlapping trimestral time series using the full, imputed monthly time series.

The relationship between rates of reproductive onset in the population and ENSO anomalies was quantified using an autoregressive linear model formulated as follows:

(3)

We used the proportion of trees bearing developing inflorescences in the focal trimester as a response (Inflor_t) and included the value of the response for the previous trimester as a predictor to account for temporal autocorrelation (Inflor_t-1). We selected the ENSO index most strongly associated with climate in our site (MEI) and selected between the two time lags of the MEI most correlated with local climate (lag 0: MEI_t, lag 1: MEI_t-1) by running two models, including only one of them as a predictor, and selecting the model with the lowest AIC score (MEIt). We used ENSO indices instead of the climate variables from TerraClimate because ENSO simultaneously affects multiple climate variables, providing an integrated “climate package” (Stenseth et al., 2003). To evaluate whether the prevalence of individual-level resource sinks affects population-level flowering phenology, the model included the proportion of trees bearing developing fruits in the population (GF_t) as a predictor. We also included an interaction term between MEI_t and GF_t, which models the linear dependence of the phenological effects of MEI_t on GF_t or (alternatively) the dependence of GF_t effects on MEI_t. We visualized this interaction by computing a cross-sectional plot of the effect of MEI_t on the response at three reference (standardized) values of GF_t (−1, 0, 1) using the “visreg” package in R (Breheny & Burchett, 2013). To do so, each standardized observation of GF_t (a continuous variable) was assigned to the reference value closest to it. We standardized all predictors on Eqn. 3 to a mean of 0 and standard deviation of 1 except for MEI, which is a standardized index (summary statistics in Table S2). We detected no significant autocorrelation in model residuals (Box–Pierce test: p > .05 for all lags; Godfrey, 1979).

All data manipulation, visualization, and analysis for this study were carried out using R version 3.6.1 (R Core Development Team, 2018).

3.1 Individual-level predictors of phenological variation

Our model (Equation 1) explained a significant proportion of the variation in reproductive rates in our sample (R² = .22, df = 149, F = 13.67, p < .001). For average tree heights, canopy emergence was estimated to increase yearly inflorescence production by 0.32 (i.e., 34.7 percent) compared with trees whose crowns were under the canopy (β₁ = 0.32, SE = 0.08, p < .001; Figure 1a). Inflorescence initiation showed a significant positive relationship with tree height for trees with non-emergent crowns, with an average increase of 3.5m in tree height (equal to 1SD) predicted to increase yearly inflorescence production by 0.21 inflorescences per year (β₂ = 0.21, SE = 0.07, p = .002; Figure 1b). In contrast, we detected a significant negative interaction of similar magnitude between tree height and crown emergence (β₃ = −0.23, SE = 0.08, p = .006), indicating that tree height was only associated with higher reproductive rates for trees with canopy-covered crowns (Figure 1b). The presence of developing infructescences significantly reduced the probability that an individual would produce new inflorescences each month from 0.117 to 0.094 (i.e., a 20.4 percent decrease) (exp(β₁) = 0.796; Table 1).

The number of distinct months of reproductive onset did not differ from the random expectation for 151 of 153 trees (98.7 percent of the sample; Figure 2), suggesting that individuals do not consistently initiate flowering in certain months of the year. Accordingly, we found no significant differences in probability of inflorescence production among trees for 10 out of 11 months relative to the reference month of July (Table S3). Though the estimated probability that an individual would produce a new inflorescence in December was 48.9 percent lower than in July (0.103 versus. 0.053; Table S3), the lack of significant differences for all other months points to an aseasonal pattern of inflorescence production.

3.2 Population-level phenological behavior

Our sample captured three population-level reproductive cycles, with supra-annual reproductive peaks separated by periods of approximately 30 to 36 months. The proportion of trees bearing mature fruits (mean ± SD =5.1 ± 6.7 percent, range = 0 – 27.0 percent) surged around 18 to 24 months following peaks in inflorescence initiation, with individuals bearing mature fruits present throughout the majority of the study (85 percent of the monitoring period; Figure 3). The proportion of individuals bearing developing inflorescences remained consistently above 13 percent of the population, and well over 20 percent for most of the study period (mean = 36 percent; Figure 3, Table S2).

3.3 ENSO anomalies and population-level phenology

Our model (Equation 3) explained most of the variation in the proportion of trees initiating new inflorescences through time (R² = .86, df = 21, F = 32.75, p < .001). Under average ENSO conditions, developing inflorescence frequency was negatively correlated with developing-infructescence frequency (Figure 4a; β₃ = −0.052, SE = 0.012, p < .001), with increases of 12 percent in the proportion of trees with developing fruits (equal to 1 SD) associated with a 5.2 percent decrease in the proportion of trees bearing developing inflorescences. In contrast, we detected a positive relationship with the Multivariate ENSO Index (MEI) for average levels of developing-infructescence frequency (Figure 4b; β₂ = 0.056, SE = 0.013, p < .001), indicating a link between the warm phase of ENSO, typically associated with El Niño conditions, and a greater frequency of trees with developing inflorescences. The effects of the MEI on the frequency of trees initiating reproduction depended significantly on the frequency of trees bearing green fruits (Figure 4c; β₄ = −0.040, SE = 0.011, p = .001). During periods of low green fruit frequency (1SD below average, ~13 percent), the model predicted that a change from −1 to 1 in the MEI (moderate la Niña to moderate el Niño conditions) would increase the proportion of trees bearing developing inflorescences by 19 percent, compared to a change of only 3 percent in periods of high developing-infructescence frequency (1SD above average, ~37 percent).

4.1 Individual-level patterns

Reproductive asynchrony could emerge if individual-level differences (in environmental stress, access to resources, genetics, etc.) affect the rate of exposure or the sensitivity of individual plants to relevant environmental cues (Ollerton & Lack, 1992). However, this mechanism predicts variation in flowering time only within seasons in which threshold levels of a cue occur. For example, species flowering in response to spring warming may vary in flowering time only within the spring, while those cued by threshold levels of solar irradiance might vary in flowering time only within seasons of sufficient day length. As a result, variation in exposure to external cues fails to explain the degree of reproductive asynchrony observed in many tropical species. In O. bataua, individuals in proximity and under similar microsite conditions routinely initiate flowering out of phase with one another. Moreover, we found no evidence of seasonality in flowering onset among individuals in our focal population. Together, these observations show that external cues are unlikely drivers of reproduction in O. bataua.

Instead, individual-level differences can influence reproduction through their effect on resource budgets. Access to sunlight influences photosynthetic rates and carbohydrate production and enhances various components of plant reproduction, including reproductive frequency and crop sizes (Poorter et al., 2019). Therefore, reproductive rates may be higher among individuals with greater access to sunlight. Accordingly, we found that crown emergence from the canopy was associated with a 34.7 percent increase in inflorescence production controlling for differences in plant size, suggesting that photosynthetic rates may drive phenological variation among individuals in this species. Additionally, tree height only had significant effects among individuals with non-emergent crowns (Figure 1a). These results suggest that tree height might influence rates of reproduction through increased access to sunlight rather than through allometric effects on resource allocation in O. bataua, though it is possible that size-dependent resource allocation could affect other reproductive variables, such as crop sizes or fruit abortion rates. Although our methodology establishes a link between sunlight and phenological behavior in O. bataua, we used a conservative method to estimate crown emergence. Direct measurements of the amount of solar radiation in treecrowns would likely reveal even greater effects of access to sunlight on inflorescence production rates.

If resource levels cue flowering among individuals, active resource sinks could reduce the probability that a tree will develop new inflorescences (Obeso, 2002). Accordingly, the presence of developing infructescences in an individual led to an estimated reduction of 20.4 percent in the monthly probability of inflorescence production. Moreover, increases in the proportion of trees bearing developing infructescences suppressed the population-level increases in inflorescence production associated with the warm phase of ENSO (below), suggesting that developing infructescences act as resource sinks that influence individual-level phenological behavior. Because the presence of developing fruit is a direct consequence of previous inflorescence production, it is possible that the decrease in the probability of producing new inflorescences reflects resource expenditures associated with recent inflorescence production. However, the significant increase in reproductive biomass associated with the transition of flowering to fruiting, as well as the high nutritional content and long maturation time of its fruits, suggests that infructescence development likely demands a greater amount of resources than inflorescence development, making fruit production and ripening stronger resource sinks than inflorescence production.

Our results support the hypothesis that resource status might mediate phenological behavior in O. bataua by providing endogenous flowering cues (Lastdrager et al., 2014; Ruan, 2014; Wahl et al., 2013). However, further analyses, such as experimental manipulation of inorganic resources, direct measurements of non-structural carbohydrates, nitrogen, and phosphorous concentrations before and after reproduction (e.g., Miyazaki et al., 2009), or developing-fruit pruning treatments, are necessary to conclusively establish the role that resource levels and resource sinks play on reproductive onset in our system.

4.2 Population-level patterns

O. bataua is pollinated by a diverse array of insect species (Núñez-Avellaneda & Rojas-Robles, 2008), and many vertebrate taxa consume its fruits and lipid-rich seeds (Henderson, 1995). We observed a constant presence of flowering individuals and a protracted presence (85% of the study period) of trees with mature fruit. Consequently, although the presence of fruiting trees was not constant, we propose that O. bataua's hyperdominance (ter Steege et al., 2013), massive flower and fruit crops, protracted presence of flowering and fruiting trees (with pronounced peaks lasting ~1.5 years), and the wide array of frugivores and pollinators it sustains (Mahoney et al., 2018; Narasimhan, unpublished Data), makes it a keystone plant resource for many rain forest organisms of the Chocó bioregion (Terborgh, 1986, Diaz-Martin et al., 2014).

While reproduction had no apparent seasonality at the individual level, the frequency of trees producing new inflorescences in the population showed supra-annual peaks occurring at 20- to 36-month intervals (Figure 3). Although some masting tropical taxa use supra-annual climate anomalies to cue reproduction in mass fruiting and flowering events, they are characterized by high levels of synchrony among individuals (Sakai et al., 2006). In contrast, the proportion of reproducing trees in our focal population never dropped below 12 percent, with about a third of the population, on average, bearing developing inflorescences during any given month. Therefore, it is unlikely that the supra-annual phenological patterns reported in this study stem from supra-annual climatic cues.

In contrast, climate variation can affect reproductive patterns through its effects on resource supply (Allen et al., 2017; Pau et al., 2013; Wright & Calderón, 2006). We found that the proportion of trees producing new inflorescences was influenced by ENSO, with anomalously warm conditions (El Niño) associated with higher inflorescence initiation in our study population. Additionally, the proportion of trees with developing fruits was negatively associated with that of trees initiating new inflorescences, with higher developing-infructescence frequencies suppressing the flowering response of the population to ENSO-induced climate anomalies. These results show that inter-annual climate variation can influence temporal patterns of reproduction at the population-level in asynchronous species and that the magnitude of these effects is mediated by the prevalence of individual-level resource sinks in the population—in this case, developing infructescences. Though ENSO anomalies seemed significantly associated with changes in temperature, rainfall, and cloud cover in our study site (Table S1), we did not directly measure local climate variables and cannot currently discern the precise climatic effects of ENSO anomalies nor the variables responsible for O. bataua's population-level responses. Therefore, direct measurements of local climate, coupled with experimental warming (e.g., Nakamura, Muller, Tayanagi, Nakaji, & Hiura, 2010), irrigation (e.g., Wright & Calderón, 2006), or high-intensity light treatments (e.g., Graham et al., 2003), could help identify the climate variables that generate population-level reproductive responses in O. bataua.

Although O. bataua's population-level phenology responded to inter-annual variation in climate, other species are likely to respond to climate fluctuations occurring at different temporal scales. O. bataua produces resource-demanding reproductive structures that develop over multiple years (Rojas-Robles & Stiles, 2009), and shorter-lived climate fluctuations, such as those occurring seasonally, are unlikely to affect resource budgets enough to induce detectable responses in the population. In contrast, many asynchronous species produce flower and fruit crops with lower energetic demands that can develop within a single season (e.g., Milton, Windsor, Morrison, & Estribi, 1982), or multiple times per year (e.g., Wright & Calderón, 2018). Consequently, climate fluctuations occurring over shorter temporal scales might provide resource pulses of sufficient magnitude to influence the frequency of individuals initiating reproduction. This mechanism could explain why the populations of many species show annual reproductive peaks even though the timing of reproduction within the year varies widely among individuals (e.g., Wright et al., 2019).