Fire history and weather interact to determine extent and synchrony of mast-seeding in rhizomatous scrub oaks of Florida

Abstract

In disturbance-prone ecosystems, fitness consequences of plant reproductive strategies are often determined by the relative timing of seed production and disturbance events, but the role of disturbances as proximate drivers of seed production has been overlooked. We use long-term data on seed production in Quercus chapmanii, Q. geminata and Q. inopina, rhizomatous oaks found in south central Florida's oak scrub, to investigate the role of fire history and its interaction with weather in shaping acorn production and its synchrony. Acorn production increased with the time since last fire, combined with additive or interactive effects of spring precipitation (+) or drought (–). Furthermore, multiple matrix regression models revealed that ramet pairs with shared fire history were more synchronous in seed production than ones that burned in different years. Long-term trends suggest that increasingly drier spring weather, in interaction with fire frequency, may drive a decline of seed production. Such declines could affect the community of acorn-reliant vertebrates in the Florida scrub, including endangered Florida scrub-jays (Aphelocoma coerulescens). These results illustrate that fire can function as a proximate driver of seed production in mast-seeding species, highlighting the increasingly recognized importance of interactions among reproductive strategies and disturbance regimes in structuring plant populations and communities.

This article is part of the theme issue ‘The ecology and evolution of synchronized seed production in plants’.

1. Introduction

Mast-seeding—spatially synchronized production of temporally variable seed crops in plant populations—is a ubiquitous reproductive strategy strongly linked to environmental variability and disturbance regimes [1–4]. Fluctuating abiotic conditions, both weak (e.g. weather) and strong (e.g. fire), play a dual role in masting as they are thought to drive annual variation and synchrony in seed production [3,5,6], but also set the stage for successful seedling recruitment, e.g. following a fire or gap formation in forests, thus shaping fitness consequences [7–10]. Despite these important effects on reproductive success, the interactive effects of abiotic extremes, disturbance events, and weather on extent and synchrony of seed production are poorly understood.

While the link between mast-seeding and variable abiotic conditions has long been established, new insights about the underlying mechanisms have recently emerged. The early notion of resource tracking by seed production has been replaced by two contrasting roles of weather in driving seed production, either as a cue or as a mechanistic driver of resource and flowering dynamics [3]. Multiple studies on Chionochloa grasses and beech trees (Fagus spp.) provide evidence that the relative difference between mean daily summer temperatures in the two preceding years (ΔT) functions as the cue for seed production [11–13]. Alternatively, environmental vetoes, weather conditions that prevent or reduce population-wide seed production such as rain during the pollination period or drought before flowering or during seed maturation, are thought to drive seed production and its synchrony among plants, particularly in wind-pollinated species [14,15]. Under a strategy termed flower masting, investment into flowering efforts is determined by weather in the previous summer, followed by effective pollination [6,11,16]. By contrast, under fruit maturation masting, annual variation in weather before and during flowering and fructification interacts with internal resource dynamics to determine pollination success, female flower abortion and subsequent seed crop size [5,17–19]. Both cues and mechanistic interactions can result in large-scale spatial synchrony through the spatial autocorrelation of weather conditions, often driven by teleconnections and climatic dipoles [10,11,20–23]. For example, acorn production in two California oaks, Quercus lobata and Q. douglasii, varies over the entire species range, over distances as far as 745 km, in close concordance with the spatial synchrony of weather conditions that affect pollination success [20]. Even as our understanding of the mechanistic links between annual abiotic conditions and masting has improved tremendously, the role of abiotic extremes and disturbances in long-term seed production patterns is still underexplored (see [24] in this issue).

Independent of weather, fires create biotic and abiotic conditions that strongly affect seed production or the recruitment of seedlings following large seed crops. Beyond the obvious consumption of above-ground vegetation, wildfires open competitive space for novel recruitment and reintroduce accessible nutrients to the soil [25], potentially impacting the seed production of surviving trees. For example, in mast-seeding bur oaks (Quercus macrocarpa), prescribed grass fires in the surrounding savannah resulted in decreased seed production in the year of the fire, followed by a year of increased seed production in burned trees relative to those unburned [26]. These results suggest that local increases in available soil nutrients, in combination with fire-year effects on individual resources dynamics, may translate directly into seed production.

The timing of mast years relative to fires can also have strong effects on plant fitness, in part depending on whether the fire consumes the plants or simply the vegetation below the crown. Natural and experimental fires following mast years of the masting grass Triodia pungens resulted in high recruitment rates of seedlings, with smoke functioning as the germination cue. In contrast, for plants that were experimentally prevented from mast-seeding in the post-fire year, recruitment from the existing seed bank was almost 200-fold lower [27]. Similarly, northern red oaks (Quercus rubra) in North American hardwood forests benefit from mast years following ground fires, with large increases in seedling densities compared to other, non-masting tree species in the same burned area [28]. On a larger spatio-temporal scale, Peters et al. [8] found that in white spruce (Picea glauca), stands originating from fires that coincided with mast years had significantly higher densities than stands impacted by fires in years of low seed crops. In addition to strategies that store and accumulate reproductive potential over years of disturbance, such as serotiny, fire can select for sprouting and post-fire seeding [29]. The interplay between disturbance regime and masting can therefore be a central driver of plant fitness [24].

Considering the fitness benefits of bumper crops produced following disturbances, selection may favour seed production synchronized with abiotic predictors of common disturbances. Several recent studies provide evidence for such an adaptation in fire-prone forests. For example, drought in the previous summer is the most important cue for seed production in P. glauca [30,31], so that climatic teleconnections that drive large-scale temporal variation in precipitation and temperature synchronize the occurrence of drought, the likelihood of subsequent wildfire, and mast-seeding [10]. A similar pattern in southern populations of European beech (Fagus sylvatica) suggests that this ‘disturbance-predictive’ form of masting may be more common than previously thought [9,32]. To our knowledge, however, no study to date has investigated the extent to which disturbances interact with annual abiotic conditions to drive seed production and its synchrony.

This study aims to test the hypothesis that fire history and spring weather interact to shape the temporal and spatial patterns of seed production in fire-adapted, rhizomatous oaks. Specifically, we investigated links among fire frequency and severity, fluctuating weather conditions, and the extent and synchrony of seed production in three species, Quercus chapmanii, Q. geminata, Q. inopina, the dominant shrubs (less than 2.5 m tall) of oak scrub vegetation formation endemic to the Lake Wales Ridge of central peninsular Florida [33]. The area experiences intermediate fire-return intervals of 5–20 years with high-severity fires that consume the majority of vegetation, following which pre-burn composition and cover quickly recover owing to resprouting of dominant shrubs [34,35]. The short intervals prevent the formation of so-called ‘oak domes’, areas covered in taller shrubs (2–10 m in height) in which not all crowns are killed during fires [36].

Previous work from our study area investigated weather and fire effects on acorn production separately, but did not investigate their interaction [37,38]. Following fires, acorn production recovered quickly, suggesting that flower initiation followed within months of sprouting [38]. However, the effect on long-term seed production patterns remains unclear. Based on 27 years of seed production in unburned areas, Abrahamson & Layne [36] identified precipitation in the previous summer as the most important negative predictor of population-level seed production in Q. chapmanii and Q. geminata, while precipitation 2 years prior was positively correlated with seed production in Q. inopina. However, that study did not investigate the interaction of weather and fire and, owing to multicollinearity and statistical independence issues in the stepwise regression models, some findings about abiotic drivers of seed production were difficult to interpret [37].

We pursued two levels of analysis: (i) a temporal analysis focused on the abiotic correlates of annual seed production and (ii) a spatial analysis focused on differences among oak ramets (distinct group of clustered stems) over the 31-year study period (1988–2018). In the temporal analysis, we used climatic window analyses to determine the phenology of key weather parameters that could predict seed production and its interaction with fire history in each species. In the spatial analyses, we determined the relative role of fire history, elevational differences and spatial proximity in driving reproductive synchrony among ramet pairs by conducting dissimilarity analyses using multiple matrix regression models. We predicted that fire history (years since the last fire) would determine whether a ramet produced seeds as resprouting advanced, while weather conditions would determine the extent of seed production, which often depends on conditions during the flower initiation and pollination periods. Furthermore, we predicted that reproductive synchrony of ramet pairs would correlate positively with shared fire history but decline with distance and elevational difference.

2. Material and methods

(a) . Study site and species

The study was conducted at Archbold Biological Station (27° 11' N, 81° 21' W), located near Lake Placid on the southern end of the Lake Wales Ridge in south central Florida, USA. The climate is humid subtropical with hot, wet summers and dry, mild winters. The habitat consists of seasonal ponds, bayhead, scrubby flatwoods, oak scrub and southern ridge sand hills vegetation associations, covering an elevational gradient (35–65 m.a.s.l.), from well-drained to excessively well-drained sandy soils [39]. The three xeromorphic, evergreen species of rhizomatous oaks, Chapman oak Q. chapmanii, sand live oak Q. geminata, and Archbold oak Q. inopina, are part of a low (less than 2.5 m tall) shrubby association interspersed with scrub palmetto (Sabal etonia) and saw palmetto (Serenoa repens). At lower elevations, especially around seasonal ponds, the oaks are joined by scattered south Florida slash pines (Pinus elliotti var. densa). At higher and better drained elevations, the focal oaks are joined by sand pines (P. clausa var. clausa), as well as by scrub hickory (Carya floridana), myrtle oak (Quercus myrtifolia) and turkey oak (Quercus laevis) [33,39]. Of the focal species, Q. chapmanii and Q. geminata belong to the subgenus Leucobalanus or white oaks, which produce acorns in the fall of the same year in which the flowers are pollinated (also called 1-year oaks), therefore requiring two years to recover seed production after sprouting, from bud initiation to mature seed [37]. By contrast, Q. inopina in the subgenus Erythrobalanus (red oaks or 2-year oaks), whose acorns do not ripen until about 1.5 years following fertilization, take nearly 3 years after bud initiation once sprouted [37]. In its natural state, the oak-dominated scrub ecosystem we report on here is well documented to be a fire-maintained formation, characterized by an expected fire return interval of 5–20 years and by highly resilient plant communities that quickly return to pre-fire composition and structure after burning [39–41]. Fires reported on in this paper were a combination of prescribed, natural (lightning ignited) and accidental (railroad sparks) [41].

3. Data

(a) . Seed production

Starting in 1988, we conducted annual acorn counts using 194 permanent posts throughout the approximately 3 km² study area (electronic supplementary material, figure S1). The posts were located throughout a 50 m-grid with a few exceptions owing to topography (mean inter-post distance between all posts: 1186 m; range 36–3054 m; electronic supplementary material, figure S1). At each post, the closest ramet of each species was located within 25 m in each quadrant of the cardinal directions, marked with a permanent tag, the number of stems counted, the distance and bearing to the post measured, and maximum height recorded. Ramets can consist of multiple acorn-bearing stems, which emerge from a resprouting genet (electronic supplementary material, figure S2). We counted all acorns in the ramet and used the log-transformed (ln (count +1)) number of acorns per stem (SAC) based on all stems (whether they carried acorns or not) as the measure of annual seed production for analyses unless stated otherwise. While this measure does not account for individual stem age or size, it controls for the number of stems per ramet, which has a much stronger effect on number of acorns per ramet. For the analysis of masting behaviour, we only considered reproductive ramets that produced at least a single acorn at some point during the study period, thus excluding vegetative ramets. Because natural stem turnover can be quite rapid in rhizomatous oaks (e.g. approx. 10 years in Q. inopina), we moved tags to (emerging) stems in close vicinity following high-severity fires, making the assumption that they belong to the same oak genet, as individual rhizomes can reach up to 30 m in diameter [34,35,37].

(b) . Fire history, habitat association, density and weather data

Following all fire events, date and severity (based on vegetation damage; electronic supplementary material, table S1) were recorded for each post and assumed to apply equally to all oaks associated with the post [41]. For analyses, we used year of fire and years since fire as measures of fire history and only considered high-severity, crown-consuming fires (severity 3; see electronic supplementary material, table S1 for definitions), which constituted the vast majority of events (more than 90% of post-fire years; electronic supplementary material, figure S3). Similarly, we used a post-based map to derive habitat association and elevation for each oak. To calculate pair-wise distance matrices for ramets, however, we projected the GPS location of each tagged ramet using the recorded distance and bearing to the post location. We extracted daily minimum and maximum of temperature, and precipitation estimates from a weather station located directly adjacent to the study tract, at the headquarters of the Archbold Biological Station (available at https://www.archbold-station.org/html/datapub/data/dataovr.html). Furthermore, the daily Keetch-Byram Drought Index (KBDI, hereafter ‘drought index’), was recorded. The index, which ranges from 0 to 800 and increases for each day without rain and drops with precipitation events, aims to capture the dryness of soil and duff layers, thus targeting conditions that affect the occurrence and spread of wildfires [41].

4. Statistical analysis

(a) . Metrics of masting and shared fire history

To characterize and compare the temporal and spatial patterns of seed production in the three species, we calculated a series of commonly used masting metrics across all species on the population-wide level [42,43]. We estimated interannual variability of acorns per stem (SAC) using the coefficient of variation of the annual population mean (CV). We also calculated the lag-1 and lag-2 temporal autocorrelation (AR1, AR2), and synchrony (s) estimated by the mean cross-correlation of SAC time series among pairs of ramets. All analyses were conducted in R v. 3.4.4 [44]. To determine whether seed production declined over the study period, as suggested by inspection of the time series (figure 1), we constructed linear mixed models of SAC for each species using the R package ‘glmmTMB’ v.0.2.2.0 [45]. The models contained the fixed effect year and a lag-1 temporal autocorrelation term of SAC values (ar1 covariance structure), fitted for each oak ramet separately, as seed production is negatively autocorrelated (table 1), a feature commonly observed in mast-seeding species.

Figure 1.

Fire history (a) and acorn production (b) at Archbold Biological Station from 1988 to 2018. (a) Proportion of 194 posts burned per year (grey bars) and average number of years since last fire for all posts (black line). (b) Population-level mean number of acorns per stem (SAC) for Q. chapmanii (n = 1159), Q. geminata (n = 1661) and Q. inopina (n = 1849). (Online version in colour.)

Table 1.

Summary metrics of population-level seed production from 1988 to 2018 for Quercus chapmanii, Q. geminata and Q. inopina at Archbold Biological Station. Mean (±s.e.) number of acorns per stem across ramets, temporal variability (CV) of the annual population mean, synchrony and temporal autocorrelation (lag-1: ACF1, lag-2: ACF2), as well mean proportion of seed producing trees (ramets) and its temporal variability (CV). Sample sizes: Qc: n = 1,159, Qg: n = 1,661 and Qi: n = 1,849.

	acorns per stem (SAC)					proportion of trees producing seed (PST)
	mean ± s.e.	CV	synchronya	ACF1	ACF2	mean ± s.e.	CV	synchronyb	ACF1	ACF2
Q. chapmanii	3.4 ± 0.5	86.6	0.089	−0.26	0.12	0.48 ± 0.03	33.0	0.980	−0.21	0.16
Q. geminata	2.6 ± 0.5	90.9	0.111	−0.16	0.36	0.24 ± 0.02	50.3	0.763	−0.14	0.34
Q. inopina	1.1 ± 0.1	60.1	0.187	0.32	0.10	0.29 ± 0.02	42.6	0.767	0.26	−0.21

^aMean crosscorrelation among seed production time series of ramet pairs.

^bCheckerboard score: lower score indicates higher synchrony, see §4a for details.

In addition, we calculated the annual proportion of seed producing trees (PST) to estimate how many individuals participated in the reproductive effort. Interestingly, this measure explained most of the variation in annual population- and landscape-level seed production, because the number of seeds produced per individual ramet is generally low (figure 1; electronic supplementary material figure S4). Therefore, we conducted a second set of analyses with population-level PST and a binary variable for annual seed production (y/n) as the dependent variables. We used the checkerboard score (hereafter ‘C-score’) as implemented in the R package EcoSimR v.0.1.0 to estimate the level of temporal co-occurrence of seed production for each pair of conspecific ramets [46,47]. This index developed in community ecology describes whether co-occurrence of two species (here replaced with seed production of individual ramets) is aggregated (low score) or mutually exclusive (high score) across a series of samples (here years). It is calculated as the proportion of non-co-occurrences relative to all occurrences [47]. Similarly, to determine whether a pair of ramets shared a similar fire history, we calculated the C-score of high severity fires (level 3) over the 30-year time-series. Lower-severity fires were so uncommon that modelling their effect on seed production patterns was not feasible (electronic supplementary material, figure S3). Finally, we calculated the mean number of years since fire (YSF) across all posts as a measure of the potential cumulative fire impact of frequent burns on the landscape scale.

(b) . Predictors of annual seed production: weather and fire history

To determine which weather parameter and relative time window best predicted annual seed production, we conducted a climatic window analysis using the R package ‘ClimWin’ v. 1.2.0 [48,49]. The analysis uses model selection to determine which parameter and which relative time window best predict the biological trait of interest, here the log-transformed acorns per stem SAC [49]. Model selection is based on linear mixed models (LMM), implemented in the R package ‘lme4’ v.1.1–17 [50], that contain the random effect ‘OAK ID’ to account for repeated sampling of ramets. For the two 1-year species (Q. chapmanii, Q. geminata), we considered weekly means of weather parameters for the 45 weeks prior to 1st September, while we considered the previous 150 weeks for Q. inopina because previous work showed support for weather effects as far as 3 years before [37]. September 1st was chosen because it constitutes the onset of the acorn caching period by Florida scrub-jays, an indirect indicator of acorn ripeness [34,35]. Because preliminary analyses revealed that acorn production increased strongly with the YSF, we included the interaction between the weather predictor and YSF in all models. We report the model selection results among the top models for each weather parameter (electronic supplementary material, tables S2–S4). Furthermore, we present climate window heatmaps and other extended analyses in electronic supplementary material, appendix A1 and figures S6–S8.

To account for potential effects of temporal autocorrelation and zero-inflation (50–70% of data are zero counts), we used the weather data in the time-frame identified by climate window analysis for further analyses. Specifically, we constructed linear mixed models of SAC with the best weather predictor, YSF, and their interaction term as fixed effects, and included a ramet-specific temporal autocorrelation term. We also constructed a set of zero-inflated models, which included either no fixed effect or the same fixed effects as the conditional model, to determine whether the data contained more zeroes than expected by the Gaussian error term. Therefore, two models are coupled in this analysis; a binomial model that investigates zeros versus non-zeros (zero-inflation model) and a Gaussian model that considers the SAC values (conditional model). To determine which model structure best fit the data, we compared AIC values (electronic supplementary material, tables S2–S4). For Q. inopina, we found that the lag-2 model with zero-inflation fit the data best, but that none of the fixed effects were significant in the conditional model, likely because of the small number of acorns produced by individuals. Therefore, we fit a binomial GLMM of the binary outcome (seeds? yes/no) with the same fixed effects and the autocorrelation term. The models were implemented in the R package ‘glmmTMB’ v.0.2.2.0 [45].

(c) . Drivers of synchrony among ramet pairs

To investigate correlates of seed production synchrony (or lack thereof) among pairs of ramets, we constructed multiple matrix regression models MRM in the R package ‘ecodist’ v.2.0.1 [51,52]. For each species, we constructed distance matrices of correlation coefficients of SAC time series, proximity, elevation difference and fire history, as estimated by the C-score of high-severity (electronic supplementary material, table S1) fire events. The same models were also constructed using the C-score for the binary seed production (y/n) as dependent variable. For each model, we report the effect size and associated p-value, as well as the R² and p-value of the model determined from 1000 permutations of the data [21,51].

5. Results

(a) . Interspecific comparison of population-level seed production

The three focal species, Q. chapmanii (Qc), Q. geminata (Qg) and Q. inopina (Qi), differed in the metrics of temporal seed production patterns commonly used to describe masting. Over the 31-year study period (1988–2018), average seed production per ramet in Qc was nearly 1.5 times that of Qg and three times that of Qi (table 1). Similarly, lag-1 negative autocorrelation of population-level seed production was nearly twice as strong in Qc compared to Qg and Qi (lag-2), which indicates that years of high seed production are followed by poor ones, or vice versa (table 1). Population-level synchrony, estimated by the mean cross-correlation of seed production among all pairs of individuals, followed an inverse pattern of seed production Qi > Qg > Qc (table 1). Combined, synchrony and autocorrelation determine the temporal population-level variability of seed production (CV), which was highest in Qg, followed by Qc and Qi (table 1).

The species also differed in the proportion of trees (ramets) producing seeds PST, the annual variation of which explains most of the temporal variation in seed production (electronic supplementary material, figure S4). Of the monitored ramets, 90.7% of Qc ramets produced acorns at least once in 30 years, while only 62.5% of Qg and 73.8% of Qi ramets ever bore acorns and were thus included in further analyses. The majority of the interannual variation in population-wide acorn crops, both in terms of absolute numbers counted across all ramets as well as per stem (SAC), was explained by the proportion of trees that produced seeds in a given year (electronic supplementary material, figure S4; absolute numbers: ANOVA: Qc: F_1,28 = 425.9, p < 0.001, R² = 0.94; Qg: F_1,28 = 1,050.8, p < 0.001, R² = 0.97; Qi: F_1,28 = 1,718.7, p < 0.001, R² = 0.98).

Despite the observed differences in masting behaviour of the three study species, the time-series for two of the three species pairs were more synchronized than expected by chance (table 2). Qc and Qg correlated significantly in both their population-level seed production time series (Pearson's product-moment correlation; r = 0.78, t = 6.6, d.f. = 28, p < 0.001; table 2) and the proportion of seed-producing trees (r = 0.81, t = 7.2, d.f. = 28, p < 0.001). Qc and Qi only correlated significantly in the proportion of seed-producing trees (r = 0.37, t = 2.1, d.f. = 28, p = 0.04), but not in the number of acorns produced (r = 0.30, t = 1.7, d.f. = 28, p = 0.10). Qg and Qi did not correlate significantly in either the proportion of seed-producing trees (r = 0.32, t = 1.8, d.f. = 28, p = 0.08) or the actual seed counts (r = 0.26, t = 1.47, d.f. = 28, p = 0.16).

Table 2.

Interspecific synchrony of population-level seed production. Correlation of time-series of (log-transformed) mean number of acorns (unshaded) and the proportion of seed-producing trees (shaded) 1988–2017. Significant correlation coefficients indicated in italics. See text for detailed statistics.

In summary, Q. chapmanii produced acorns more frequently, in larger numbers, and with stronger negative autocorrelation than Q. geminata and Q. inopina. However, owing to higher synchrony among ramets, Q. geminata showed higher temporal variability of population-level seed production. Q. chapmanii and Q. geminata showed strong interspecific synchrony, both in terms of SAC and PST, while Q. inopina showed significant but low-level synchrony in PST with Q. chapmanii.

(b) . Drivers of temporal variation in seed production

(i) . Temporal trends of seed production

Seed production of all three species declined over the study period (electronic supplementary material, figure S3). The strongest negative correlation of SAC with year was observed in Qg (LMM: B = −0.0155 ± 0.0012 (±s.e.), z = −12.04, p < 0.0001), followed by Qc (B = −0.0109 ± 0.0011, z = −9.61, p < 0.0001) and Qi (B = −0.0059 ± 0.0011, z = −5.46, p < 0.0001). This pattern was also reflected in the proportion of seed-producing trees, which declined from an average of 42% across all species in the late 1980s/early 1990s to below 30% in the 2010s (electronic supplementary material, figure S5).

(ii) . Weather, fire history and annual variation in seed production

Seed production in all three species recovered quickly following fires (figure 2 and table 4). In both Qc and Qg, PST surpassed the long-term mean in the 2nd year following the fire, while SAC took longer to recover (figure 2a,b). In Qi, the species that develops acorns over 1.5 years, the long-term mean of PST and SAC was reached again in the 3rd year (figure 2c).

Figure 2.

The effect of fire history on seed production in Florida scrub oaks. The proportion of seed-producing trees (left) and stem acorn count (right) for (a) Q. chapmanii, (b) Q. geminata and (c) Q. inopina as a function of years since high-severity fires. Raw data shown; see table 4 for statistical analysis. Dashed lines indicate mean values across all years for ramets with 10 years of data and burning at least once during 1988–2018.

Table 4.

Abiotic predictors of acorn production, top models of climate window analyses. Standardized estimates β, standard error, z- and p-values for zero-inflation (seed production: y/n?) and conditional models (number of seeds per stem) for linear mixed models with temporal autocorrelation term and random effect ‘OAK ID’ for repeated sampling of ramets. See table 3 for time window and model selection. Significant variables are highlighted in italics.

	β	s.e.	z	p
(a) Q. chapmanii
zero-inflation
years since fire YSF	0.0006	0.0065	0.1	0.9290
mean drought index	0.9815	0.0587	16.2	<0.0001
YSF × drought index	−0.0038	0.0064	−5.9	<0.0001
conditional model
years since fire YSF	0.0259	0.0020	13.0	<0.0001
mean drought index	−0.1320	0.0138	−9.6	<0.0001
YSF × drought index	−0.0044	0.0016	−2.8	0.0057
(b) Q. geminata
zero-inflation
years since fire YSF	−0.0731	0.0098	−7.5	<0.0001
mean precipitation	−0.4943	0.0626	−7.9	<0.0001
YSF× precipitation	−0.0213	0.0116	−1.8	0.0652
conditional model
years since fire YSF	0.0155	0.0059	2.6	0.0083
mean precipitation	0.1977	0.0316	6.3	<0.0001
YSF× precipitation	−0.0047	0.0054	−0.9	0.3826
(c) Q. inopina
zero-inflation
years since fire YSF	−0.4883	0.0417	−11.7	0.0083
mean precipitation	−0.0534	0.1213	−0.4	0.6666
YSF × precipitation	−0.1868	0.0396	−4.7	<0.0001
conditional model
years since fire YSF	0.0005	0.0044	0.1	0.9170
mean precipitation	0.0341	0.0221	1.5	0.1232
YSF× precipitation	−0.0028	0.0039	−0.7	0.4588

The climate window analyses revealed that, when taking fire history into account, spring precipitation, or the lack thereof, was the strongest abiotic predictor of acorn production (figure 3a–c). In Qc, the top model showed a negative effect of the mean drought index in weeks 28–16 before 1st September (calendar weeks 7–19) on annual SAC (figure 3a; electronic supplementary material, figure S6). In Qg, SAC was best predicted by the mean precipitation 29–23 weeks before (calendar weeks 6–12), while in Qi this relationship was restricted to the window of weeks 27–26 (calendar weeks 8–9; table 3; electronic supplementary material, figures S8 and S9).

Figure 3.

The effect of fire history and weather on seed production in (a) Q. chapmanii, (b) Q. geminata and (c) Q. inopina at Archbold Biological Station. Model predictions of marginal effects from GLMMs of log-transformed acorns per stem. Time windows for weather parameters are presented in table 3.

Table 3.

Climate window analysis of acorn production. Model selection result for linear mixed models of (log-transformed) seed production per stem as a function of years since fire (YSF) and the following abiotic variables: mean maximum temperature (T_max), precipitation and drought index. Climate window in weeks relative to September 1st of a given year, dAIC relative to null model. Strongest predictor is highlighted in italics.

model/time window	Q. chapmanii		Q. geminata		Q. inopina
	weeks	dAIC	weeks	dAIC	weeks	dAIC
temperature (Tmax)	31–16	−1166.5	21–20	−642.2	123–122	−520.7
precipitation	31–19	−1115.6	29–23	−854.1	27–26	−532.7
drought index	28–16	−1274.4	23–18	−778.7	134–132	−421.2

In all three species, fire history (YSF) and abiotic conditions had additive or interactive effects on acorn production (table 4). The models that accounted for zero-inflation and temporal autocorrelation of individual-level seed production revealed that in Qc, the probability of reproductive failure (excessive zeros, i.e. zero-inflation) as well as the extent of seed production is driven by the interaction of fire history and drought. Spring drought increases the number of trees that fail to produce seeds at all, but the effect is offset by the increasing likelihood of seed production as time passes since the last fire (table 4a). This pattern is also observed in the extent of seed production, so that trees that have burned recently produce fewer seeds in response to drought (figure 3a). By contrast, seed production of Qg is driven by the additive effects of fire history and spring precipitation, both of which increase the likelihood and extent of seed production, but their interaction term is not significant (table 4b and figure 3b). In Qi, which generally produces few seeds per individual, only the likelihood of seed production was predicted by fire history and its interaction with spring precipitation, so that rainy springs reduced the impact of fire history (figure 3c). This was the main driver of seed production patterns (electronic supplementary material, figure S4), as the extent of individual-level seed production itself was not significantly correlated with weather or fire history (table 4c).

Fire frequency (figure 1a) and the dominant abiotic predictors of seed production varied over time with few temporal trends that correlated with the decline in acorn production. Total precipitation during the relevant time window for Qg (weeks 6–12) declined significantly over the study period (electronic supplementary material, figure S7B; ANOVA F_1,32 = 4.7, p = 0.037), while the other two predictors showed non-significant trends in the direction that would decrease acorn crops (electronic supplementary material, figure S7; Qc; mean drought index weeks 7–19; F_1,32 = 3.0, p = 0.091; Qi: total precipitation weeks 8–9: F_1,32 = 1.3, p = 0.255). The average years since fire per post, a measure of cumulative fire impact, which would be reduced under more frequent fires, did not increase over the study period 1988–2018 (figure 1; F_1,28 = 0.58, p = 0.454).

(c) . Drivers of intraspecific synchrony in seed production

Ramet pairs with shared fire history were more likely to produce seeds synchronously, while the trees that were most similar in the extent of seed production were closer together. Multiple matrix regression analysis for all three species indicated that the checkerboard score for co-occurrence of seed production showed a strong positive correlation with the checkerboard score for co-occurring fires (table 5a–c). This relationship was strongest in Qi, followed by Qg and Qc (table 5). In Qi, seed production synchrony also declined with distance (table 5c). We also detected a small positive effect of elevational difference on seed production synchrony in Qg and Qi.

Table 5.

Drivers of synchrony in acorn production. Results of multiple regression on similarity matrices with acorn count or co-occurring seed production (checkerboard score) as dependent variable and proximity (log-transformed), elevational distance and fire history as predicting variables. For acorn count, distance and elevation, higher values thus indicate higher similarity, for fire history and co-occurring seed production, lower values indicate higher aggregation. Sample sizes of oak ramets with at least 10 years of data are indicated in brackets. Significant predictor variables are highlighted in italics. Model R² and p-values derived from 1000 permutations of model.

species (n)	response	variables	coefficient B	p-value (variable)	R2 (model)	p-value (model)
(a) Q. chapmanii (645)	co-occurring seed production	elevation	0.0001	0.491	0.0025	0.001
		proximity	0.1238	0.354
		fire history	1.4891	0.001
	acorn count	elevation	<−0.0001	0.142	0.0239	0.001
		proximity	−0.0424	0.001
		fire history	0.0036	0.066
(b) Q. geminata (539)	co-occurring seed production	elevation	0.0004	0.001	0.008	0.001
		proximity	0.0773	0.578
		fire history	1.6899	0.001
	acorn count	elevation	<−0.0001	0.57	0.0085	0.001
		proximity	−0.0281	0.001
		fire history	−0.0015	0.562
(c) Q. inopina (785)	co-occurring seed production	elevation	0.0003	0.001	0.0109	0.001
		proximity	−0.1069	0.038
		fire history	2.2544	0.001
	acorn count	elevation	0.0001	0.001	0.0028	0.001
		proximity	−0.1921	0.001
		fire history	0.0026	0.405

Pair-wise correlation coefficients of acorn counts within species declined strongly with distance. This relationship was most pronounced in the poor seed producer, Qi, followed by Qc and Qg (table 5).

6. Discussion

The results of our long-term study on the three dominant rhizomatous scrub-oak species illustrate how fire history and spring weather determined the extent and synchrony of acorn production at Archbold Biological Station. After high-severity fires, the likelihood and extent of seed production dropped dramatically, then increased over time (figure 2). In Q. chapmanii and Q. inopina, this relationship with time since fire was affected by spring weather, the former by drought (−), the latter by precipitation (+, figure 3). In Q. geminata, the likelihood and extent of seed production also increased with years since last fire and with precipitation, but, unlike in the other two species, the interaction between the two drivers was not significant (table 4). Furthermore, in all three species, ramets that shared a similar fire history were more synchronous in their seed production than those with dissimilar fire history (table 5). While both the probability and extent of seed production have declined over the study period, the cumulative impact of fire, as measured by mean number of years since the last fire for 194 sampling locations, fluctuated over the whole study period and remained stable over recent years (figure 1). By contrast, the abiotic conditions (drought index, spring rainfall) during the relevant time periods did show trends that could reduce acorn crops. The change in seed production over time was only significant for total spring precipitation in the time window relevant to Q. geminata, the species in which acorn production declined most dramatically (electronic supplementary material, figure S3). As precipitation patterns keep shifting as a function of climate change, the interactive effects of fire and drought may thus exacerbate the decline of acorn production in central Florida's oak scrubs.

As rhizomatous, fire-adapted oaks, all three species returned to average pre-fire seed production levels per stem within 3–5 years following high-severity fires as ramets quickly grow and flower within months (figure 2; [37,38,52]). While stem-level reproductive effort may peak relatively early during post-fire regeneration as our results indicated, the accumulation of stems in ramets over time may stabilize overall acorn availability for the next few years following this peak [37].

Consistent with previous work, the climate window analysis showed that precipitation or the lack thereof (drought) during the flowering period in spring was the dominant driver of annual variation in seed production in the two 1-year oaks, Q. chapmanii and Q. geminata (table 3; [37]). In contrast to previous findings, however, we also found that precipitation in the year of acorn maturation best predicted seed production even in the 2-year oak Q. inopina, despite considering weather as far as 3 years before acorns ripened. Abrahamson & Layne [36], who used 24 years of data from a nearby location, found that February precipitation in the year before fructification (i.e. 1.5 years before ripening) best predicted Q. inopina acorn crops. However, that study sampled oaks in plots that had been unburned for decades as a result of purposeful fire-exclusion. We found that years since fire was the dominant predictor of acorn crops, whose annual variation was then mediated by precipitation (figure 3c). It is, therefore, possible that the larger, unburned Q. inopina ramets in the Abrahamson & Layne [36] study responded differently to precipitation from the recently burned ramets in our study. Importantly, both analyses found that precipitation, rather than temperature, affected seed production of all three species. While total annual precipitation has not changed over the study period (data not shown), the timing has shifted to result in drier springs, reducing available water during critical periods for acorn production (electronic supplementary material, figure S6B). Importantly, climate change affects both weather and fire frequency and severity. Thus, the interactive effects between disturbance and abiotic conditions during recovery leave these oaks and the species that depend on their acorns particularly vulnerable to climate change.

While population-wide synchrony of seed production was comparatively low for all three species, the multiple matrix regression analysis showed that ramet pairs with similar fire history were more synchronous in their seed production than trees that had burned in different years. This may largely be a physiological effect of the time that is needed for post-fire sprouting and recovery of flower and acorn production under conditions of elevated nutrient availability in the soils [53]. Abrahamson & Layne [36] discuss that both white and red oaks initiate floral buds on ramets in the year of the fire. For the white oaks (Q chapmanii and Q. geminata), a total of 2 years are required for post-fire seed production, while the red oak (Q. inopina) requires 3 years until acorn production sets in, as reflected in figure 2. We argue that these physiologically determined seed production recovery times are the main factor synchronizing seed production of the three species. This is corroborated by the fact that Q. inopina, which requires one more year for recovery of acorn production, showed stronger synchrony than the other two species.

Such effects of fire history on reproduction can align the flowering efforts of individuals, thus providing the economy of scale that is considered a fitness benefit of mast flowering and seeding [2,54]. Interestingly, this effect was only found when analysing the timing of seed production by using the checkerboard score, but not when considering the number of acorns per stem (table 5). Rather, the correlation of acorn counts among tree pairs was best predicted by proximity, as is commonly found in oaks (e.g. [21]). This finding suggests that the timing and extent of seed production may be decoupled to some degree, so that other individual or site-specific factors such as rhizome size, soil properties, water access or conspecific density could affect the number of acorns a stem can produce.

Rapid recovery of acorn production when resprouting following fires may provide fitness benefits, as fires can open new areas for colonization nearby. While population dynamics and ground cover of rhizomatous oaks during disturbance-free periods are mainly driven by vegetative growth and cloning [55,56], fires can open competitive space for dispersed seeds in areas where non-sprouters dominate or were killed. Reduced vegetation cover and increased nutrient content in the soil may result in better seedling establishment conditions than before the disturbance [25,57]. However, while low-severity fires may occasionally increase germination of acorns cached before [57], seeds near the surface, and in shallow caches will generally be killed [58]. In contrast, when seeds arrive soon after stand-clearing disturbance, this can result in successful recruitment of seedling cohorts, resulting in spatial signatures that can sometimes be detected decades after the event [8,24]. At the study site, the recovery of the vegetation cover to pre-fire states is rapid, however, and long-term changes in species composition were only observed under fire exclusion [34–36].

Seed predation and dispersal by vertebrates can be affected by fires as well. Granivorous rodents, which mainly function as seed predators of acorns and other large seeds (e.g. [59,60]), avoid recently burned areas and thus reduce acorn pilferage rates [57]. Scatter-hoarding corvids, however, often prefer recently disturbed areas, including fire scars, for seed caching [61–63]. In fact, territories of the federally endangered Florida scrub-jay in our study area include substantial proportions of recently burned areas, which quickly become of high value as the vegetation recovers [64]. These jays harvest and individually cache thousands of acorns each fall, and their preferred cache sites are open sand free of leaf litter [65]. Fires create numerous such clear spots, which invite acorn caching and thereby generate germination opportunities for unrecovered acorns. In areas where suppression of natural, lightning-ignited fires was practiced for several decades, Florida scrub-jay vital rates plummeted and their populations declined or were extirpated [35,64].

Our results lend support to the growing recognition of the pivotal roles played by fire in mediating ecosystem dynamics and the evolution of plant life-history strategies [25]. In terms of the synchrony of active reproductive potential with suitable post-fire conditions, such as reduced competition and high contents of plant available nutrients, these strategies vary across different fire regimes [29]. Resprouters maintain community presence in fire regimes of higher severity that would not support traits for enhancing aboveground survival, such as enhanced bark thickness, with fire return intervals that are too short for the accumulation of serotinous canopy seed banks that are released through fires [29]. For Florida scrub-oaks, the combined post-fire strategy of resprouting and seeding thus maintains the potential of successful local re-establishment, while providing opportunities of long-distance dispersal and colonization, mediated by scatter-hoarding corvids.

Data accessibility

The data are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.ht76hdrg7 [65].

Authors' contributions

M.B.P., R.B. and J.W.F. conceived the research, R.B., A.T. and S.P. coordinated data collection, M.B.P. analysed the data and wrote the first draft of the manuscript. All authors contributed to writing and gave final approval for publication.

Competing interests

We declare that we have no competing interests.

Funding

This work has been supported by the U.S. National Science Foundation (NSF), grant nos. BSR-8705443, BSR-8996276, IBN-0077469, IOS-0346557, DEB-0855879 and DEB97-07622.