Oak masting under climate change: when spatially heterogeneous shifts in flowering phenology reshape pollen limitation

Summary

• Masting, the irregular, synchronised production of seeds within tree populations, drives forest regeneration and ecosystem dynamics. Its response to climate change may depend on phenological shifts in key reproductive stages, yet this link remains poorly understood.

• Using 30 yr of daily airborne pollen data from 20 temperate oak populations (Quercus petraea and Q. robur), we examined the relationship between shifts in pollen phenology and pollen limitation (estimated from annual pollen concentration). We analysed how these changes were driven by site‐specific weather trends during the forcing and diffusion periods.

• Pollen phenology advanced overall, while pollen limitation decreased. On average, pollen concentration nearly doubled in 30 yr (+93%) yet with striking contrasts among populations. During the study period, late‐flowering populations experienced significant warming during the preseason period, leading to earlier phenology but stable pollen concentration. Conversely, early‐flowering populations saw little change in preseason temperatures or phenology but experienced sharp warming during the diffusion period, resulting in a strong increase in pollen concentration.

• These findings demonstrate that heterogeneous local climate trends induce divergent pollen phenology and pollen limitation trajectories, reshaping reproductive conditions differently among populations. Consequently, oak masting and forest regeneration success may respond differently across populations under continued climate change.

Conceptual framework illustrating the pollen variables and the hypotheses linking trajectories of pollen phenology and masting.

Introduction

Masting, the highly variable and synchronised production of seeds across years within a population (Janzen, 1976; Bogdziewicz et al., 2025), is a key reproductive strategy that deeply influences forest ecosystem dynamics (Ostfeld & Keesing, 2000; Ostfeld et al., 2006; Widick et al., 2025). Episodic heavy seed events shape the population dynamics of seed consumers (Yang et al., 2010; Venner et al., 2011; Gamelon et al., 2013; Zwolak et al., 2016, 2018) and, through cascading effects, determine forest regeneration success (Frey et al., 2007; Gratzer et al., 2022; Pesendorfer et al., 2024). Because the mechanisms driving masting remain only partly understood, its trajectory under climate change, and that of forest regeneration, remains uncertain (Hacket‐Pain & Bogdziewicz, 2021; LaMontagne et al., 2021; Bogdziewicz, 2022; Bogdziewicz et al., 2025).

Weather conditions influencing masting are diverse, acting across seasons and at different stages of reproduction (Wright et al., 1999; Koenig & Knops, 2013, 2014; Gallego Zamorano et al., 2016; Pearse et al., 2016; Chen et al., 2018; Journé et al., 2025). Depending on species and population, masting has been associated with variations in temperature, precipitation or drought during these critical stages (Pérez‐Ramos et al., 2010; Pearse et al., 2016; Bogdziewicz et al., 2025). Because phenology determines the timing of exposure to such drivers, phenological shifts can strongly reshape how masting dynamics respond to climate change. Yet, to date, no study has assessed how shifts in reproductive phenology co‐occur with, and potentially modify, changes in the weather conditions that regulate key phases of masting.

This knowledge gap stems from the scarcity of multisite, multidecadal datasets that jointly monitor fruiting dynamics and the phenology of key reproductive stages. To circumvent this, we indirectly address this gap by linking recent climate‐driven shifts in flowering phenology to concomitant changes in pollen limitation, a primary driver of oak masting. This is made possible through the analysis of long time series of airborne pollen waves (both onset and concentration). Airborne pollen concentration was shown to be a robust proxy for pollen limitation in temperate oaks, as it is strongly correlated with fruit set (Fleurot et al., 2024). Furthermore, the day of pollen wave onset provides a reliable proxy for flowering phenology (García‐Mozo et al., 2007; Jato et al., 2007; Fernández‐Martínez et al., 2012). Analysing these two components of pollen dynamics therefore offers a powerful, indirect way to explore how recent shifts in flowering phenology may reshape pollen limitation and, ultimately, oak masting.

We explored this link in temperate oaks (Quercus petraea and Q. robur), two widespread species in Europe whose mast‐seeding dynamics strongly structure forest ecosystems. Interannual variation in pollen limitation, linked to the availability of pollen for reproduction, is a primary driver of interannual variation in seed production (Rapp et al., 2013; Pearse et al., 2015, 2016; Schermer et al., 2020). Pollen limitation can drive strong interannual variability in pollination success, while also facilitating seeding synchrony among individual trees (Isagi et al., 1997; García‐Mozo et al., 2007; Fernández‐Martínez et al., 2012; Koenig et al., 2015). It results from a combination of mechanisms, including resource allocation to flowering, the degree of phenological synchrony among individuals within a population and the weather conditions during floral bud development, pollen maturation and release, which together drive the pollination efficiency (Crone & Rapp, 2014; Koenig et al., 2015; Schermer et al., 2019). Interestingly, airborne pollen concentration effectively captures the net outcome of all three mechanisms, making it a comprehensive proxy for evaluating interannual shifts in pollination success (Fleurot et al., 2024).

Flowering phenology, by indirectly determining the weather conditions associated with pollination, constitutes a pivotal life‐history trait. For instance, in temperate oaks, early flowering in spring often exposes pollen maturation and dispersal to unfavourable weather, resulting in frequent pollination failures. Conversely, flowering later would reduce environmental vetoes and promote more cyclic, predictable seed production (Schermer et al., 2020). Despite its importance, the vulnerability of pollen limitation to climate change has received little attention. Anticipating how oak masting will be affected by the ongoing warming requires understanding the current links between weather, phenology and pollen limitation across diverse climates and tracking how they have already changed over recent decades.

Under climate change, the interplay between shifts in flowering phenology and in pollen limitation might be complex and heterogeneous across populations (Richardson et al., 2005; Kotlarski et al., 2012; Vitasse et al., 2018a,b). Oak flowering phenology, tightly linked to the well‐documented timing of leaf budburst (Pesson & Louveaux, 1984; Chesnoiu et al., 2009), is highly sensitive to temperatures during late winter and early spring, the forcing period, when accumulated heat drives bud development after endodormancy release (Dantec et al., 2014; Gauzere et al., 2017; Baumgarten et al., 2021). Although warming generally advances flowering (Menzel et al., 2006; Adams‐Groom et al., 2022; Vogel, 2022), responses are expected to vary geographically, mirroring leaf phenology shifts along altitudinal and latitudinal gradients (Dantec et al., 2014; Vitasse et al., 2018b; Alcázar et al., 2024). Because climate warming is spatially uneven (Guan et al., 2021), population differences in flowering shifts may primarily reflect contrasting temperature trends during the forcing phase. Alternatively, divergent phenological responses could arise from other regulatory mechanisms. In warm regions, where flowering is expected to be already early (Menzel et al., 2006), reduced winter chilling and photoperiodic constraints may limit further advances (Fu et al., 2015), whereas in colder regions, sufficient chilling and weaker photoperiodic constraint could permit stronger forcing‐temperature‐driven advances.

The phenological responses could, in turn, modify the timing of exposure to environmental cues and, consequently, the average local weather conditions experienced during pollination, leading to contrasting temporal trends in pollen limitation (Fig. 1). We hypothesise that the trajectory of pollen limitation likely falls between two extremes: (1) flowering phenology advanced in response to warming, leading to broadly steady weather conditions prevailing during pollination and hence resulting in a stable trajectory in pollen limitation dynamics; or (2) flowering phenology remained stable on average, pollen maturation and dispersal would occur under increasingly warmer and more favourable conditions: This could increase average pollen availability, driving a long‐term trajectory towards reduced pollen limitation.

Fig. 1.

Conceptual framework illustrating the pollen variables and the hypotheses linking trajectories of pollen phenology and masting. Based on annual daily series of airborne pollen collected across 20 sites over 30 yr for two temperate oak species (Quercus petraea and Q. robur), we explored the diversity of trajectories in both pollen wave onset (a proxy of pollen phenology) and pollen concentration (a proxy for pollen limitation). (a) Illustrates the pollen variables used in this study (see Material and Methods). (b, c) Illustrate two extreme scenarios within the context of recent climate change: In the first scenario (b), preseason temperatures (during the forcing period) increase, which is accompanied by a phenological shift towards earlier onset dates. Due to this phenological shift, temperatures during the pollen wave remain unchanged on average, thus resulting in no significant temporal shifts in annual pollen concentration, pollen limitation and masting dynamics. The second scenario (c) represents the exact opposite situation: Pollen phenology remains unchanged on average (either because of stable preseason temperatures or because trees have reached a physiological limit to advancing their phenology). Consequently, temperatures during the pollen wave increase, driving an increase in pollen concentration over time. This leads to reduced pollen limitation, potentially decreasing masting intensity through a reduction in interannual variation in seeding and a loss of synchrony among trees. Data from the 20 monitored sites reveal a strong heterogeneity in responses: Sites that historically exhibited a late phenology behave according to Scenario 1 (b), whereas sites that were already structurally early‐flowering align closely with Scenario 2 (c).

Using three decades of daily airborne pollen data from 20 oak populations in mainland France, we assessed trends in pollen phenology and pollen limitation under ongoing climate change. We hypothesised (1) an overall advance in pollen phenology, mirroring shifts in leaf phenology driven by forcing temperatures (Ahas et al., 2002; Morin et al., 2010), with geographic differences similar to those observed in leaf unfolding (Vitasse et al., 2018b); and (2) site‐specific changes in airborne pollen limitation according to local weather conditions during pollen maturation and release.

Materials and Methods

Study System

We focused on the two European temperate oak species, Q. petraea (Matt.) Liebl. and Q. robur L., two monoecious, wind‐pollinated, self‐incompatible species that are phylogenetically close and occasionally hybridise (Kremer et al., 2002). Both species are commonly found across Europe, with the exception of the Mediterranean rim (Eaton et al., 2016; Caudullo et al., 2017; Badeau et al., 2022). Flowering in these species follows a 1‐yr cycle with male and female floral buds being initiated during the previous summer; in France, pollination typically occurs from late April to mid‐May, varying with latitude and altitude (Pesson & Louveaux, 1984; Firmat et al., 2017; Grundström et al., 2019; Badeau et al., 2022).

Airborne pollen data and derived pollen variables

We used daily airborne pollen counts, from the database acquired by the Réseau National de Surveillance Aérobiologique (RNSA) covering 1987–2020. This network consists of Hirst‐type volumetric traps (Hirst, 1952; Thibaudon et al., 2013) settled at 79 sites across mainland France. These traps continuously draw air at a constant flow rate, enabling standardised quantification of daily airborne pollen amounts available for pollination (Fleurot et al., 2024). Because species‐level attribution is not possible based on pollen morphology, we restricted analyses to sites where Q. petraea and Q. robur together accounted for at least 80% of oak forest cover within a 50‐km radius (Schermer et al., 2019; IGN, 2020; Fleurot et al., 2024). The relative local abundance of these two species had no significant impact on pollen dynamics (Fleurot et al., 2024), which is consistent with their overlapping flowering phenology at the local scale (Bacilieri et al., 1996; Eaton et al., 2016). We then restricted the dataset to the 20 sites that had at least 20 yr of daily observations between 1987 and 2020 (Supporting Information Table S1).

From each annual daily airborne pollen series and at each site, four synthetic pollen variables were computed (Fleurot et al., 2024): (1) pollen amount, that is, the annual sum of daily airborne pollen amounts; (2) pollen wave duration, that is, the number of days elapsed between the 10th and 90th percentiles (Q10 and Q90) of the annual cumulative airborne pollen curve (proxy for pollination period length); (3) pollen concentration, that is, the average daily airborne pollen concentration during the pollen wave. This metric is calculated over the specific window defined by the Q10 and Q90 quantiles of the seasonal wave to exclude long tail distributions; (4) pollen wave onset as proxy of pollen phenology, that is, the day‐of‐year (DOY) when the cumulative pollen curve reaches Q10; we also computed Q5, Q25 and Q50 as sensitivity checks (see Supporting Information). Pollen concentration serves as a robust proxy for pollen limitation: Previous empirical work has demonstrated that airborne pollen concentration accounts for a large proportion of variance in the fruit set (R ² = 0.40), even when quantified at a distance from forest sites, and successfully captures the various components of pollen limitation (Fleurot et al., 2024). Airborne pollen concentration is inversely related to pollen limitation. For analyses, pollen concentration and pollen amount were log‐transformed using a natural logarithm.

Identification of two critical weather‐sensitive periods for pollen dynamics

A two‐step procedure was employed to identify the critical weather‐sensitive periods for both the onset of the pollen wave (Q10) and the pollen concentration. (1) Identification of candidate windows: Daily mean temperatures and cumulative precipitation were extracted from the SAFRAN database (Durand et al., 1993; Vidal et al., 2009) using the 8 × 8 km grid cell closest to each pollen site. We applied a sliding moving window approach to identify candidate windows where weather variables showed significant correlations with Q10 and pollen concentration, respectively (Journé et al., 2024; Oberklammer et al., 2026). (2) Model Selection: We identified the most parsimonious statistical models for Q10 and pollen concentration separately by evaluating all possible combinations of the candidate weather windows (see Methods S1 for detailed procedures and results).

Weather‐sensitive periods for Q10: The sliding windows were based on calendar dates and identified a single 55‐d window beginning on February 23, during which the average temperature has a strong influence on Q10 (Fig. S1a,c); this is referred to below as the ‘preseason temperature’. This period corresponds to the preanthesis forcing period, during which temperature drives bud development and determines the onset of pollination (Fu et al., 2015).

Weather‐sensitive periods for pollen concentration: Pollen concentration likely depends on weather conditions at several time periods starting from the summer before pollination when floral induction occurs, followed by floral organogenesis, up to the period of pollen release. We therefore used the sliding moving window approach on two distinct time periods to investigate its climatic drivers. We first investigated separately the effect of mean daily temperatures and summed precipitation around the pollen wave (Figs S1b,d, S2). As the diffusion window was short and tied to the pollen wave onset, we used Q10 as day 0 to account for spatial and interannual variation in phenology. Second, we ran the analysis based on calendar dates over a far larger time period starting from the summer before pollination through the end of winter (late February) (Fig. S3). In total, 11 candidate weather‐sensitive periods have been identified (Table S2), but the most parsimonious linear mixed model predicting log‐transformed pollen concentration only included the effect of the temperatures around the pollen wave onset (Q10), hereafter called ‘diffusion temperatures’, computed over a 30 d window that started 16 d before Q10 (see Table S3; Methods S1 for details).

Once the climatic drivers were identified for both pollen concentration and pollen wave onset, we computed the mean temperatures for each site‐year over these two specific timeframes.

Linking trajectories of pollen dynamics to masting patterns

Due to the lack of long‐term oak acorn production data (which would require a 30‐yr time series matching our pollen dataset), we were unable to directly cross‐analyse the temporal trajectories of pollen dynamics with those of oak masting. To circumvent this limitation and explore the connection between these trajectories, we developed an indirect approach. We leveraged a key finding of our study: a site's initial pollen wave onset (i.e. its estimated phenological baseline in 1987) serves as a reliable predictor of its subsequent long‐term temporal shift in both pollen phenology and pollen concentration (see Results).

Consequently, we utilized a network of 14 oak populations (hereafter ‘masting sites’) currently monitored for reproduction across contrasted climatic regions, which exhibit distinct masting patterns (Fleurot et al., 2023). For each site, we estimated its yearly pollen wave onset (Fig. S4) and reconstructed its initial pollen phenology (Fig. S5) while characterising masting patterns (Table S4). We then tested the prediction that contemporary masting patterns – specifically individual‐ and population‐level interannual variation in seed production (${\mathrm{CV}}_i$ and ${\mathrm{CV}}_p$), as well as synchrony of seed production between trees within populations ($\mathrm{Sy}$) – should be correlated with the initial pollen phenology reconstructed for these specific forest sites (Fig. S6; see Methods S2 for full methodological and analytical details).

Statistical analyses

To quantify the global temporal trend (1987–2020) for the four key variables (preseason temperature, diffusion temperature, pollen wave onset and pollen concentration; Fig. S7), we used linear mixed models with ‘year’ as a fixed effect and ‘site’ as a random effect on both intercept and slope. The slope value was interpreted as the global temporal trend. To check the robustness of our findings, we repeated the analyses using three additional phenology quantiles (Q5, Q25 or Q50) and further investigated global temporal trends for the two components of pollen concentration separately (i.e. the pollen amount and pollen wave duration).

The site‐level heterogeneity in temporal trends of the four variables (preseason temperature, diffusion temperature, pollen wave onset and pollen concentration) was investigated using a three‐step approach. First, site‐specific temporal trends were estimated as the slopes of linear regressions relating each variable to year. Because annual rates of change can be difficult to interpret, we expressed temporal shifts as the cumulative change over the entire monitoring period (1987–2020), which provides a more meaningful measure of the magnitude of long‐term change. Second, we tested whether these shifts were related to initial pollen wave onset at each site. Instead of using the first observed phenology, we used the phenology estimated for 1987 (i.e. at the onset of monitoring, hereafter ${\varphi }_{1987}$) from linear models predicting pollen wave onset as a function of year at each site. This choice accounts for strong interannual fluctuations in pollen wave onset at each site and slight differences in monitoring start years, which could introduce noise in the observed initial phenology. To ensure the robustness of the results, the analysis was repeated using the average phenology observed during the first 5 yr of survey for each site, which yielded qualitatively similar results (Fig. S8). Third, we explored potential mechanisms underlying site‐level temporal shifts in pollen wave onset and concentration. Specifically, simple linear regressions were used to assess (1) the effect of preseason temperature shifts on phenology shifts, (2) the effect of phenology shifts on diffusion temperature shifts and (3) the effect of diffusion temperature shifts on pollen concentration shifts.

All analyses were performed with the R free software (v.4.5.0, http://cran.r‐project.org). Packages lmerTest (Kuznetsova et al., 2017), merTools (Knowles et al., 2020), dplyr (Wickham et al., 2014), boot (Canty & Ripley, 1999), lubridate (Spinu et al., 2010), tidyverse (Wickham, 2016), ggplot2 (Wickham et al., 2007), ggspatial (Dunnington, 2017), car (Fox et al., 2001), performance (Lüdecke et al., 2019), eurostat (Lahti et al., 2015), rnaturalearth (Massicotte & South, 2017) and wesanderson (Ram & Wickham, 2014) were used to perform analysis and/or create the figures. R ² for mixed models was computed using the function ‘r2_nakagawa’ from the package performance (Lüdecke et al., 2019).

Results

Dual weather control of oak pollination

Pollen wave onset in temperate oaks is tightly linked to temperatures during preseason period, which govern bud development once endodormancy is released. Warmer late‐winter to early‐spring conditions consistently advance pollen wave onset, with a mean shift of over 5 d per additional degree at the global scale (Figs 2a and S1a,c; slope = −5.14 ± 0.20, t = −25.22, P < 2 × 10⁻¹⁶). In addition, warmer conditions around the onset of pollination (diffusion period) substantially enhanced pollen concentration (Figs 2b and S1b,d; slope = 0.08 ± 0.006, t = 13.00, P < 2 × 10⁻¹⁶): Cold diffusion temperatures (< 9°C) were consistently associated with low pollen concentrations. Conversely, to reach maximum pollen concentrations, it is necessary, but not sufficient, for average diffusion temperatures to be high (above 11°C). Together, these results highlight the dual climatic controls of preseason and diffusion temperatures on both the timing and efficiency of oak pollination.

Fig. 2.

Warmer conditions lead to earlier pollen wave onset and higher pollen concentration in Q. petraea and Q. robur. (a) Pollen wave onset is strongly determined by preseason temperature, highlighting the predominant role of environmental forcing: Warm late winters accelerate bud development and advance the timing of the pollen wave. (b) Log‐transformed pollen concentration increases sharply with diffusion temperature, reflecting its influence on pollen maturation, release and aerial diffusion. Dots represent annual site‐level values for the 1987–2020 period. Solid lines represent the linear mixed models, that is, the global relations at France scale, with the shaded area showing 95% confidence intervals. The coefficient of determination (R ²) corresponds to the conditional R ² of the mixed models. Site‐specific relationships are shown as dotted lines (one line per site). DOY, day‐of‐year.

Large‐scale temporal shifts in weather and pollen variables

Over the past three decades, oak pollen wave onset has advanced significantly at the France scale, pollen wave onset being 5.73 d earlier on average (Fig. 3b; Table 1). This shift, consistent across multiple phenological quantiles (Fig. S9), mirrors the rise in preseason temperatures (+0.73°C; Fig. 3a; Table 1). Diffusion temperatures also rose significantly (by 1.08°C; Fig. 3c; Table 1), providing increasingly favourable conditions for effective pollen maturation and diffusion. Consequently, pollen concentrations increased substantially, nearly doubling on average across all sites over the 30‐yr period (+16% on log_e‐transformed data relative to 1987, corresponding to a +93% increase in absolute pollen concentration; Fig. S10; Table S5).

Fig. 3.

Temporal trends in climatic conditions, pollen wave onset and pollen concentration over the past three decades. (a) Preseason and (c) diffusion temperatures have risen over the past three decades at the global scale, coinciding with the advance of (b) pollen wave onset and the increase in (d) pollen concentration in Q. petraea and Q. robur. Dots represent annual site‐level values. Solid lines show the global temporal trends (coloured areas: 95% confidence interval), whereas dotted lines indicate local temporal trends. Detailed model results are provided in Table 1. DOY, day‐of‐year.

Table 1.

Thirty‐year global temporal‐trends in pollen wave onset, pollen concentration, pre‐season and diffusion temperatures.

	Unit	Mean	SE	Linear mixed models results
				Intercept		Slope
				Est	SE	Est	SE	t‐value	P‐value
Pollen wave onset	DOY	109.54	0.36	112.69	1.31	−0.17	0.04	−4.40	< 0.001
Pollen concentration	nb grains d−1 (loge)	4.46	0.04	4.12	0.14	0.02	0.01	3.16	0.005
Preseason temperature	°C	8.12	0.05	7.71	0.20	0.02	0.01	3.99	< 0.001
Diffusion temperature	°C	10.86	0.06	10.21	0.17	0.04	0.01	5.22	< 0.001

Each model used the variable listed in the first column as the response variable and the number of years elapsed since 1987 as the predictor. Columns three and four report the mean and SE for each variable across all years and sites, and the subsequent columns present the results of linear mixed models with site as a random effect on both intercept and slope. Significant P‐values are shown in bold. DOY, day‐of‐year.

Site‐level temporal trends in weather and pollen variables

We evaluated whether site‐level temporal shifts were structured along the gradient of initial pollen wave onset (${\varphi }_{1987}$). The intensity of the phenological temporal shift was correlated with ${\varphi }_{1987}$ (Fig. 4b; Table 2), with late‐flowering sites (around the 119^th day in 1987) advancing their pollen wave onset by 10 d on average (Fig. S1b), whereas early‐flowering sites (around the 105^th day in 1987) exhibited only minor shifts (1 d on average) or remained stable (Fig. S7a).

Fig. 4.

Distinct 30‐yr trajectories of pollen dynamics of temperate oaks (Q. petraea and Q. robur) across sites sorted by their initial pollen wave onset (${\varphi }_{1987}$). This figure illustrates the diverging temperature and phenological trajectories of sites based on their historical baseline. Panels show 30‐yr temporal shifts in (a) preseason temperature, (b) pollen wave onset, (c) diffusion temperature and (d) pollen concentration (expressed as % change derived from log_e‐transformed data) plotted against the initial pollen wave onset (${\varphi }_{1987}$). Sites with historically early phenology (low ${\varphi }_{1987}$) showed (b) little further phenological advance but (c) marked increases in diffusion temperatures and (d) pollen concentration. Conversely, late‐flowering sites (high ${\varphi }_{1987}$) exhibited (b) strong phenological advances associated with (a) rising preseason temperatures, but (c) little shift in diffusion temperature or (d) pollen concentration. Dots represent estimated average initial phenology (estimated pollen wave onset in 1987, see Method) and temporal shift at the site level; vertical and horizontal straight segments indicate their SE. Solid lines represent fitted linear models with 95% confidence intervals (coloured area), whereas dashed horizontal lines represent the lack of temporal shift. Results are qualitatively similar when using the mean pollen wave onset measured over the first 5 yr of monitoring for each site instead of ${\varphi }_{1987}$ (see Fig. S8). DOY, day‐of‐year.

Table 2.

Analysis of the factors determining shifts in pollen wave onset and pollen concentration.

		Linear models results						R 2
		Intercept		Slope
		Est	SE	Est	SE	t‐value	P‐value
Fig. 4	A	−5.88	2.68	0.06	0.02	2.52	0.021	0.26
	B	59.09	13.42	−0.58	0.12	−4.89	< 0.001	0.57
	C	14.31	3.06	−0.12	0.03	−4.29	< 0.001	0.54
	D	298.11	92.44	−2.49	0.82	−3.05	0.007	0.34
Fig. 5	A	−1.81	0.98	−5.31	0.91	−5.81	< 0.001	0.65
	B	1.85	0.34	0.12	0.04	2.61	0.018	0.27
	C	1.91	7.14	13.47	4.98	2.71	0.014	0.29

We report the results of the linear models associated with each panel of Figs 4 and 5. Models associated with Fig. 4 all integrated the initial pollen wave onset of each site (${\varphi }_{1987}$) as a predictive variable and the site‐level temporal shifts for the temperatures, pollen wave onset and concentration as response variables (see Fig. 4). Models linked to Fig. 5 investigated links between the site‐specific temporal shifts in temperatures and pollen variables. Significant P‐values are shown in bold.

However, these phenological temporal shifts were not constrained by initial pollen wave onset per se but were instead tightly coupled to site‐specific changes in temperature during preseason period (Fig. 5a; Table 2). Indeed, early‐flowering sites experienced little to no warming during this period, whereas late‐flowering sites warmed by up to 2°C (Fig. 4a; Table 2).

Fig. 5.

Pathways linking local temperature warming to shifts in pollen processes in temperate oaks (Q. petraea and Q. robur). This figure explicitly tests the mechanistic links between climate warming and biological responses. Panels show relationships between temporal shifts in pollen‐related variables and in temperature for each site over the past three decades. (a) Pollen wave onset advanced in sites where preseason temperatures warming was strong. (b) Diffusion temperatures showed little change in sites with strong phenological advances but rose sharply in sites where pollen wave onset remained unchanged. (c) Pollen concentrations increased in sites that experienced strong warming during the diffusion period (expressed as % change derived from log_e‐transformed data). Dots represent temporal shift at the site level while vertical and horizontal segments around them indicate SE. Solid lines represent fitted linear models with 95% confidence intervals (coloured area), whereas dashed lines indicate no temporal shift at the site level.

An inverse pattern emerged for pollen concentration (Fig. 4d; Table 2). Early‐flowering sites exhibited strong increases in pollen concentration since 1987, the reference year (up to +56.7% based on log_e‐transformed data, which corresponds to an 83% increase in absolute pollen concentration, Fig. S7c), whereas late‐flowering sites showed little or no change (Fig. S7d). This response tracked the temporal shifts observed at each site for diffusion temperature, which rose sharply at early‐flowering sites (by +2°C to 3.5°C) but remained relatively stable at late‐flowering sites (Fig. 4c; Table 2). Accordingly, site‐level temporal shifts in pollen concentration appear to be driven by changes in diffusion temperatures (Fig. 5c; Table 2), which are themselves linked to changes in phenology (Fig. 5b; Table 2).

Linking initial pollen wave onset to contemporary masting patterns

By leveraging the fact that a site's initial pollen wave onset (${\varphi }_{1987}$) serves as a reliable predictor of its subsequent long‐term temporal shift in both pollen phenology (Fig. 4b; Table 2) and pollen concentration (Fig. 4d; Table 2), we tested whether initial pollen wave onset is correlated to contemporary masting dynamics (Fig. S6). Specifically, we expected masting sites with an early initial pollen wave onset, which are predicted to have experienced a long‐term increase in pollen concentration and a concomitant reduction in pollen limitation, to exhibit lower masting intensity than at historically late‐flowering sites. In accordance with this prediction, we found that the mean degree of individual‐level variation in fruit production ($\overline{{\mathrm{CV}}_i}$) was significantly and positively correlated with the estimated initial pollen wave onset (see Fig. S6a). Meanwhile, population‐level variation (${\mathrm{CV}}_p$) (Fig. S6b) and mean synchrony between trees ($\overline{\mathrm{Sy}}$) (Fig. S6c) exhibited a similar positive trend, although it did not reach statistical significance (see Methods S2 for full statistical outputs).

Discussion

Masting in temperate oaks is a hallmark of pollen limitation, driven by strong interannual fluctuations in airborne pollen concentration (Pearse et al., 2015; Pesendorfer et al., 2016; Schermer et al., 2019; Fleurot et al., 2024). Over the past three decades, we detected a significant overall advance in pollen phenology along with rising temperatures before, and around, the pollination period. Crucially, this climate shift was accompanied by a nearly twofold increase in pollen concentration (+93% on average over the 30‐yr study period). These concurrent changes suggest a reduction in the frequency of pollination failures, consistent with the global trend of increased oak seed production in France (Caignard et al., 2017).

Our study confirms that late‐winter and early‐spring temperatures (preseason temperature) capture the thermal forcing controlling pollen phenology, aligning with extensive literature on leaf‐out and flowering phenology in temperate trees (Dantec et al., 2014; Gauzere et al., 2017). Crucially, our sliding window analysis and model selection approach (see Methods S1) demonstrate that temperatures immediately surrounding the onset of the pollen wave are the main meteorological driver of pollen concentration. In agreement with recent findings (Bogdziewicz et al., 2026), we did not detect any significant explanatory power of alternative weather variables, and especially no effect of previous summer temperature at the time of floral initiation. This lack of alternative environmental signals highlights the overriding importance of late‐winter and early‐spring temperature trajectories in shaping both pollen limitation and oak masting dynamics.

Our study also revealed pronounced disparities among sites in the magnitude of the shifts in both pollen phenology and pollen concentration, suggesting that these two ecological responses are tightly intertwined. Although these differences were correlated with the sites' initial pollen wave onset, our results do not support the hypothesis that early‐flowering populations were intrinsically constrained in their phenological response. Instead, the heterogeneity among sites appears explained by localised differences in climate trends during the forcing period. Late‐flowering populations, which experienced the strongest preseason warming during the study period, showed marked phenological advances (10 d earlier on average over three decades) and stable pollen limitation. By contrast, early‐flowering populations, exposed to little preseason warming, exhibited limited phenological changes but substantial warming during pollen diffusion (+4°C since 1987), resulting in a sharp increase in pollen concentration (up to 83% in three decades). Ultimately, the contrasting trajectories of phenology and pollen concentration among sites are inextricably linked: Local differences in climate warming during the forcing period determine the temporal window during which pollen is released, subsequently determining the microclimatic conditions and temperature levels experienced by populations during pollen diffusion.

Our results highlight the central role of flowering phenology in mediating the effects of climate change on pollen concentration and pollen limitation. Ultimately, changes in flowering phenology are likely to shape, at least in part, the trajectory of oak masting. The variation in masting patterns recently reported among oak populations (Fleurot et al., 2023) might partly reflect the heterogeneous trends in pollen phenology and concentration trends identified here. To indirectly test this hypothesis, we estimated the initial pollen phenology (in 1987) of 14 oak populations for which seed production has been monitored over 12 yr. Consistent with our hypothesis, populations characterized by an initially early flowering phenology, where our results predict the strongest reduction in pollen limitation since the 1990s, show lower interannual variation in seed production at the tree level than populations with initially later phenology. Similarly, both interindividual reproductive synchrony and the magnitude of interannual variation in seed production at the population level tend to be lower in these early populations. These potentially recent contrasting shifts in pollen dynamics and masting patterns could have cascading effects on seed predator dynamics and forest regeneration processes (Bogdziewicz et al., 2020, 2021, 2023b). In particular, such spatial heterogeneity in pollen dynamics and pollen limitation shifts might help explain recently observed regional declines or variations in viable seed production across European tree species (Foest et al., 2026).

Because masting trajectories may depend on how weather conditions will change gradually around the pollen wave, it is crucial to anticipate the timing of this wave and, consequently, how floral phenology will respond in the coming decades. In temperate oaks, flowering is closely linked to leaf budburst and primarily driven by environmental forcing, with heat accumulation from late winter to early spring controlling bud development (Roberts et al., 2015; Diggle & Mulder, 2019; Guo et al., 2023). Over the past 30 yr, rising forcing temperatures have advanced flowering, especially at late‐flowering sites. Yet, multiple factors shape floral phenology, and species‐specific responses to warming can differ (Ziska et al., 2019; López‐Orozco et al., 2021; Manangan et al., 2021). Global warming may limit the chilling exposure required for dormancy release, counteracting forcing effects, while photoperiod can set a lower limit on flowering dates (Zohner & Renner, 2015; Singh et al., 2017). Controlled experiments and mechanistic models of pollen‐wave dynamics, based on leaf phenology frameworks, are needed to untangle these effects. Extending this approach to other wind‐pollinated species is essential to identify the mechanisms shaping flowering phenology and predict their consequences for pollen limitation and masting dynamics across forests.

Beyond pollination, the trajectory of masting will also depend on how phenology shifts across other key reproductive stages. In that sense, floral induction, which partly determines the quantity of flowers produced, represents a major driver of masting in many tree species (Satake & Kelly, 2021). Floral induction can be highly sensitive to temperature within a narrow time window, and exposure to temperature is expected to change depending critically on whether the phenology of floral induction remains stable or shifts under climate change. For instance, the European beech (Fagus sylvatica), whose budburst phenology is linked to photoperiod (Vitasse & Basler, 2013; Basler & Korner, 2014; Meng et al., 2021), may also exhibit remarkably stable induction timing. Such stability could explain why floral induction in this species is particularly sensitive to temperature fluctuations occurring during a very specific calendar period, around the summer solstice (Journé et al., 2024). Such phenological stability may also underlie both the strong large‐scale synchronisation of fruiting observed across Europe (Bogdziewicz et al., 2023a) and the recent changes in beech masting dynamics, accompanied by a sharp decline in viable seed production (up to 80%) driven by reduced synchrony, loss of pollination efficiency and increased seed predation (Bogdziewicz et al., 2020, 2021, 2023b). In other species, flowering induction phenology may be far more variable among populations, individuals and years, owing to weaker photoperiodic control and greater temperature sensitivity (Foest et al., 2026). This variability makes the underlying mechanisms harder to disentangle, yet they are likely to play an equally important role in shaping the future trajectories of masting under climate change.

Postpollination phenology may also exert a strong influence on reproductive success, particularly in species or populations highly sensitive to environmental stress during fruit development. Shifts in the timing of flower fertilisation, fruit maturation and fruit fall under climate change may alter the exposure of developing fruits to summer droughts, the seed biomass or lead to changes in the microclimatic conditions of seed germination and establishment, ultimately affecting recruitment and forest regeneration (Badano & Sánchez‐Montes De Oca, 2022; Vicente & Benito Garzón, 2024).

Our study provides the first empirical evidence that climate‐driven shifts in flowering phenology co‐occur with changes in pollen limitation, revealing a novel mechanism that can drive contrasting masting trajectories across oak populations. This work calls for research at the interface of phenology and masting to establish robust predictions for the future of masting and forest regeneration under continued climate change. Advancing understanding will require mechanistic models that couple key phenological processes with quantitative effects of weather on pollen limitation, reproductive resource allocation and seed development and germination, moving beyond correlative masting–climate analyses. Our work demonstrates that reproductive phenology, including flowering timing and pollen concentration, is itself changing under climate warming and should be explicitly incorporated into process‐based models to improve predictions of future reproductive dynamics and their ecological consequences.

Competing interests

None declared.

Author contributions

EF, LK, SV and M‐CB‐V conceived and led the study. AS and the RNSA provided the pollen dataset. EF and LK assembled the dataset. SV and JL supervised data analyses. EF and LK analysed the data. EF, LK, M‐CB‐V and SV wrote the first draft. All authors revised the manuscript. EF and LK contributed equally to this work and M‐CB‐V and SV contributed equally to this work.

Disclaimer

The New Phytologist Foundation remains neutral with regard to jurisdictional claims in maps and in any institutional affiliations.

Supporting information

Fig. S1 Identification of the time period during which temperature best explains pollen wave onset (preseason period) and pollen concentration (pollen diffusion period).

Fig. S2 Identification of the best suited windows to account for the effect of summed precipitation on pollen concentration during pollen diffusion period.

Fig. S3 Identification of the best suited windows to account for the effect of temperature (a) and summed precipitation (b) on pollen concentration during flower initiation.

Fig. S4 Comparison between predicted and observed pollen wave onsets for 16 pollen sites with a short pollen survey.

Fig. S5 Between sites variability of initial pollen wave onset across masting's sites.

Fig. S6 Link between masting statistics and initial pollen wave onset.

Fig. S7 Time series of pollen processes across sites.

Fig. S8 Mean pollen wave onset confirms the importance of pollen wave onset in mediating the effect of climate change.

Fig. S9 Temporal trends observed for pollen wave onset are consistent with changes in other quantiles of pollen distribution.

Fig. S10 The amount of pollen emitted over the past 30 yr has increased, but not the duration of the pollen wave.

Methods S1 Sliding moving window approach to identify key weather‐sensitive periods explaining pollen wave onset and pollen concentration.

Methods S2 Linking initial pollen phenology to masting patterns.

Table S1 Comparison of climatic conditions among pollen‐monitoring sites.

Table S2 Coordinates of the best candidate drivers of pollen concentration during two key time periods.

Table S3 Comparison of the performances of models explaining pollen concentration.

Table S4 Spatial coordinates, masting metrics and initial pollen wave onset across temperate oak site.

Table S5 Quantification of the average value and of the strength of the temporal trend observed at France scale for pollen wave duration, pollen amount and three phenological dates of pollen period.

Please note: Wiley is not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.

Data availability

The authors confirm that the data and the code supporting the results are available online on the Zenodo platform at doi: 10.5281/zenodo.21302294.