Seasonal climate drives population growth but not costs of reproduction of a perennial wildflower

Abstract

Costs of reproduction are predicted to shift under climate change, but the extent to which weaker or stronger costs influence population responses to interannual climate variation is unknown. We asked how seasonal climate, manipulated rainfall, and costs of reproduction influence vital rates and population growth in a long‐lived herbaceous perennial plant, Primula hendersonii, across an ongoing rainfall manipulation experiment in oak savanna of northwestern North America. Simulated drought reduced population growth rates, and vital rates (e.g., probability of flowering, individual growth) responded individualistically to variation in winter, spring, and summer temperatures, although not to variation in seasonal precipitation. However, only warmer spring temperatures were associated with a decline in population growth rates. Although we observed a weak negative effect of past reproduction on growth and future reproduction for large individuals, these costs of reproduction ultimately did not influence population growth. Further, observational and manipulative experiments to detect costs of reproduction suggest subtle differences in cost expression. We show that direct climate drivers had a stronger effect on population growth than indirect changes in costs of reproduction and may be more important for understanding population persistence under climate change.

INTRODUCTION

Accurately predicting population growth under environmental change requires a clear understanding of the climatic drivers and demographic trade‐offs influencing individual performance and population vital rates (e.g., survival, growth, and reproduction that contribute to population growth rates) (Morris et al., 2020; Villellas et al., 2015). However, both the magnitude and direction of demographic trade‐offs are expected to depend on environmental conditions (Reznick, 1985); for example, stronger life‐history trade‐offs are predicted under decreased resource availability (van Noordwijk & de Jong, 1986). Thus, to make more accurate predictions of population responses to climate change, and here we focus on plants, requires understanding how the climate drivers that influence vital rates and population growth (Ehrlén & Morris, 2015; Morris et al., 2020) interact with life‐history trade‐offs to jointly influence plant population response to environmental change (Compagnoni et al., 2016; Jongejans et al., 2010).

Climate directly affects plant population dynamics through effects on underlying vital rates (Andrzejak et al., 2025; Christiansen et al., 2023; Ehrlén et al., 2016; Lindell et al., 2022; van de Pol et al., 2016). Yet, making the explicit link between climate drivers and population growth rates is challenging. For example, the importance of dormant season and lagged climate is becoming increasingly clear (Anderson, 2023; Evers et al., 2021, 2023; Schnablová et al., 2021; Scott et al., 2022), but these factors are uncommonly examined in demographic studies. Second, different climate variables can influence different vital rates, even in opposing directions, leading to responses that are specific to each vital rate (Boggs & Inouye, 2012; Dalgleish et al., 2011; Doak & Morris, 2010; Lindell et al., 2022; Louthan et al., 2021; Williams et al., 2015), and require population models to scale individual vital rate responses to population growth rates (Ehrlén & Morris, 2015). Finally, much of what is known relies on correlative relationships between climate and vital rates across time (Compagnoni et al., 2021), even as these correlative relationships can covary with other factors, making predictions under future climates difficult. Climate manipulation experiments can help identify and confirm the causal influence of climate on vital rates and population growth. Despite the importance of identifying causal relationships, these climate manipulation experiments may have low biological realism (Carlyle et al., 2011; Korell et al., 2020), which highlights the need for pairing long‐term observational demographic data with climate manipulations.

In addition to direct effects of climate on vital rates, climate may also influence demographic trade‐offs between vital rates, leading to indirect effects on population growth. The cost of reproduction hypothesis predicts that reproduction carries a fitness cost, such that allocating finite resources to current reproduction reduces resources available to allocate to future reproduction, growth, or survival (Reznick, 1985; Stearns, 1989; Williams, 1966). Fitness costs are thus predicted to be detected demographically through negative relationships between current reproduction and future vital rates (Compagnoni et al., 2016; Ehrlén & Morris, 2015; Miller et al., 2012; Obeso, 2002). In iteroparous herbaceous plants, which reproduce multiple times throughout their lifetime, current reproduction has been shown to reduce subsequent reproduction (Hamann et al., 2021; Horvitz et al., 2010; Sherman et al., 2019; von Euler et al., 2012), survival (Aragón et al., 2009), growth (Jacquemyn et al., 2010; Miller et al., 2012; Williams et al., 2015), or a combination of these vital rate responses (Kellett & Shefferson, 2018; Sletvold & Ågren, 2011, 2015a, 2015b; Sun et al., 2022). However, how climate drivers influence costs is less clear. Costs are in part determined by the total resources available and an individual's reproductive effort, here defined as the proportion of the total resource budget that is allocated to reproductive processes (Obeso, 2002). If climate affects resource availability and resource allocation to reproduction, then climate change may indirectly influence costs of reproduction in addition to directly influencing reproduction (Iler et al., 2019; Lambrecht et al., 2007; Sletvold & Ågren, 2015b; Williams et al., 2015). Importantly, because environmentally driven demographic trade‐offs may scale to changes in population growth rates (Laughlin et al., 2020; Siefert & Laughlin, 2023), especially trade‐offs that are predicted by life‐history theory to impact fitness (Dorken et al., 2025), these indirect effects may buffer or exacerbate population responses to climate (Hamann et al., 2021; Jongejans et al., 2010). For example, if the lifetime fitness cost of reproducing increases under more stressful conditions, population decline could be exacerbated (Hamann et al., 2021; Jacquemyn et al., 2010).

Quantifying the effect of climate on costs of reproduction requires quantifying costs, which is notoriously difficult (Dorken et al., 2025; Obeso, 2002). Individual variation in resource acquisition can overshadow the detection of costs; for example, if individuals have poor future performance independent of previous reproduction (van Noordwijk & de Jong, 1986; Villellas & García, 2018). Further, spurious correlations can occur if an environmental driver has a positive effect on growth but a negative effect on reproduction but does not influence resource allocation to different vital rates (Knops et al., 2007). Multiple methods have been used interchangeably to detect and measure costs of reproduction, but whether manipulative and correlational studies result in similar observations of costs remains unclear (Dorken et al., 2025; Fox & Peter, 1991). Costs of reproduction can be measured experimentally by manipulating reproductive effort, namely through flower clipping (lowering reproductive effort) or pollen supplementation (increasing reproductive effort). In observational demographic data, costs can be detected by comparing the relative effects of initiating reproduction on future vital rates of nonreproductive plants (low reproductive effort) and reproductive plants (high reproductive effort) (Dorken et al., 2025; Miller et al., 2012; Obeso, 2002). However, the expression of costs may differ if energy allocation differs between initiating reproduction and producing seeds, and reproductive effort is measured in different ways across study types. Experimental and observational studies also differ in duration, with manipulative experiments usually lasting only 1–2 years, and observational studies requiring longer term data (Dorken et al., 2025; Miller et al., 2012; Obeso, 2002). Finally, costs may not be evident at low environmental stress if resources are abundant, leading to small or undetectable trade‐offs. Thus, manipulating environmental conditions can aid in cost detection, particularly through climate manipulations (Hamann et al., 2021), comparing populations with different habitat types or stress (Sletvold & Ågren, 2015a, 2015b), or examining the effects of reproduction across long‐term natural variation in climate (Miller et al., 2012; Williams et al., 2015).

Here, we paired experiments with observational demographic data to examine the influence of seasonal climate and manipulated rainfall on costs of reproduction, vital rates, and population growth in an herbaceous perennial plant, Primula hendersonii, in Garry oak savanna in the Pacific Northwest of North America. In this system, water limits ecosystem productivity (Smith & Williams, 2023), and future climate projections predict changes in seasonal precipitation intensity. To disentangle direct effects of climate variation on vital rates from indirect effects through costs of reproduction, we paired a 1‐year flower clipping experiment that manipulated reproductive effort with observational demographic and climate data across seven years (2017–2023), all collected within a rainfall manipulation experiment. Using this combined approach, we evaluated (1) how individuals express demographic costs of reproduction across two experiment types; (2) how interannual variation in seasonal climate variables and simulated rainfall change (i.e., manipulated water availability) influence costs of reproduction; (3) which seasonal climate variables influence vital rates and how those then translate to effects on population growth rate; and finally (4) the extent to which costs of reproduction influence population growth rate across interannual variation in climate. We compared results from the short‐term manipulative experiment of costs of reproduction with those from observed longer term natural variation in climate and reproduction. We hypothesized that increased reproductive effort would decrease future reproduction and growth; reproductive costs would be higher under drier and hotter conditions; and simulated drought and hotter temperatures would contribute to lower population growth rates.

METHODS

Study system and species

This study took place at the Cowichan Garry Oak Preserve in Duncan, BC, Canada. Garry oak ecosystems are sub‐Mediterranean oak savannas spanning from southern British Columbia to northern California. Average annual precipitation at the site is approximately ~1100 mm/year (460 mm in winter, 210 mm in spring, 81 mm in summer, and 325 mm in fall for 1980–2010 climate normals; Wang et al., 2016). Under a high emission scenario (Representative Concentration Pathway 8.5), annual precipitation is predicted to increase by 11% by 2080, primarily in rainfall intensity in winter (14% increase), spring (7% increase), and fall (19% increase), while summer precipitation is projected to decline by 26% (Cowichan Valley Regional District, 2017). Annual average temperatures are predicted to increase by 3°C by 2050 and 4.5°C by 2080 (Cowichan Valley Regional District, 2017).

Primula hendersonii (A. Gray) A.R. Mast & Reveal is a long‐lived early spring wildflower common in oak savannas of the Pacific Northwest of North America (Hall & Mathew, 1997). The aboveground growing period is restricted to February through June, and plants perennate underground from the onset of summer drought through mid‐winter. Basal rosette growth and flower production within a season is determinate and draws resources from perennating structures. Plants spend 2–5 growing seasons as vegetative plants before flowering. Flower primordia are preformed prior to leaf out; inflorescence buds emerge with leaves in late winter, and plants put out a single flowering stalk with 1–17 hermaphroditic flowers that open in April to May. Flowers are described as requiring buzz pollination for seed set (Douglas et al., 1999; Hall & Mathew, 1997), though one source suggests that unpollinated flowers can self‐pollinate (Gielens et al., 2014). In a seed addition experiment, seeds only germinated within 1 year of planting, so we assume the population does not have a seed bank (Loesberg, unpublished data). Seeds germinate in late winter and spend one growing season with only cotyledons, emerge in their second year with one true leaf, and emerge in their third year with more than one leaf, which we consider an adult rosette. Plants also reproduce asexually through bulblets, which form underground and emerge as one‐leaf plants in their first year, although asexual reproduction was not quantified in this study. Vertebrate herbivores (deer and small mammals) consume 20%–60% of stalks of flowers and fruits across the field site (Loesberg, unpublished data). Insects (Lepidoptera larvae) bore into individual capsules and consume seeds.

Rainfall manipulation experiment

We used a rainfall manipulation experiment established in 2015 to simulate changes in precipitation intensity following the International Drought Experiment protocol (Knapp et al., 2017). Treatments included ambient rainfall, simulated drought, and simulated increased precipitation (n = 5 per treatment, 2 × 2 m plots). Drought treatments simulated 100‐year drought conditions using 3 × 3 m wood structures with corrugated clear plastic intercepting 50% of incoming rainfall, allowing for a buffer around the focal 2 × 2 m study area. Irrigated treatments simulated higher rainfall intensity by irrigating the focal 2 × 2 m study areas via polyethylene emitter tubing with the intercepted 50% of ambient rainfall from drought shelters. For further protocols and design, see Knapp et al. (2017) and Smith and Williams (2023). In response to three consecutive years (2018–2020) with low spring rainfall, resulting in little difference in water availability between rainfall treatments, in 2021–2023 we added an additional 60 L of water two or three times per week in April–June (15–22.5 mm/m² per week) to ensure that the irrigated treatment received more water than the control and drought treatments. We added a third round of watering in weeks where air temperatures were above 20°C to account for water loss from increased evapotranspiration under warmer temperatures. Air temperature and ambient precipitation were measured hourly on site with a HOBO U30 weather station. Instances of missing daily climate data were imputed using data from the nearest weather station (North Cowichan, from Environment and Climate Change Canada Historical Climate Data at https://climate.weather.gc.ca/index_e.html) using a linear regression model between climate variables from each site (r ² = 0.87 for precipitation; r ² = 0.99 for temperature).

In the 15 rainfall experiment plots, P. hendersonii naturally occurs in five control plots, four drought plots, and four irrigated plots. Beginning in 2017, we marked adult P. hendersonii plants in 1 × 2 m² plots within the rainfall experiment focal study area with permanent aluminum tags. Annually in early spring from 2017 to 2023, we collected demographic measurements of rosette size (total leaf area estimated as the area of an ellipse using the length and width of the rosette), survival, flowering status, and number of flowers for marked adult individuals. For plots with greater than 100 adults, we only marked new adult plants in 1‐m² subsets of the plots. We counted the number of seeds produced by each tagged plant that flowered and made seeds (64–157 plants per year) in 2021–2023. Vegetative dormancy was rare, with 49 of 1212 total plants reemerging as adult plants after not emerging for one growing season across all years of the experiment, so dormancy was not examined in analyses.

We estimated one‐leaf plant (the juvenile stage) survival and probability of becoming an adult rosette by counting the total number of one‐leaf plants annually from 2017 to 2023, and in addition, marked a subset of one‐leaf plants in 2021–2023 with toothpicks and relocated them the following year. We quantified seed germination rates in a pot experiment and a seed addition experiment (Loesberg, unpublished data). To estimate the transition from seedlings to one‐leaf plants, we counted the number of seedlings in 2023 and the number of one‐leaf plants in 2024 in two 10 × 10 cm² areas in each plot.

Clipping experiment

To determine the effect of reproductive effort on future vital rates (growth, survival, and reproduction) within the rainfall experiment, we used a simulated flower herbivory experiment to manipulate reproductive effort for P. hendersonii in 2021 and examined the effect on vital rates in 2022. These treatments have natural analogs for this population, where deer and rabbits clip flower stalks and insects consume individual flowers and capsules. In April 2021, when greater than 75% of flowers on each plant had opened, we assigned flowering plants in each plot randomly to one of three clipping treatments: control treatments (high reproductive effort) received no clipping or herbivory for the season; half clipped (medium reproductive effort) treatments had half of their flowers clipped; all clipped (low reproductive effort) treatments had all flowers clipped. Wire mesh cages were placed over individuals that still had flowers to prevent further natural mammal florivory. Sample sizes varied due to varying numbers of flowering individuals in each rainfall experiment plot, ranging from one to five plants in each clipping treatment per plot, for a total of 147 plants. At the end of the growing season, we counted the number of seeds produced by each plant. In April 2022, we relocated and determined the survival, growth, flowering status, and flower production of each individual.

We used linear mixed‐effects models to evaluate the effects of reproductive effort in 2021 across rainfall treatments on growth, probability of flowering, survival, and number of flowers produced in 2022. All models included size in 2021 as a fixed effect and plot as a random intercept effect to account for nonindependence of observations. We used successful seed set in 2021 as a continuous measure of reproductive effort to capture the range of reproductive effort since flower production varied among individuals within a clipping treatment due to size. To test the significance of fixed effects (size in 2021, seed set in 2021, rainfall treatment, all two‐way interactions, and a three‐way interaction between all variables), we used likelihood ratio tests on nested models. Growth was modeled with a linear mixed‐effects model with a Gaussian distribution, probability of flowering and survival were modeled with generalized linear mixed‐effects models (GLMMs) with binomial distributions, and number of flowers was modeled with GLMMs using a Poisson distribution.

Climate variables

We used principal components analysis (PCA) to select seasonal climate variables to include in vital rate model selection (Appendix S1: Section S1). The PCA included seasonal fall (September–November), winter (December–February), spring (March–May), and summer (June–August) climate variables (total precipitation, maximum temperature, minimum temperature, and mean temperature) for 2013–2022. We included mean, maximum, and minimum temperature in the PCA to identify the temperature variable that contributed most to variation in the PCA because these three summary statistics were often correlated and because we anticipated that this species could respond to temperature in various directions (e.g., either minimum or maximum spring temperature could be important for an early season species' performance in a relatively cold region). We chose to include seasonal values of climate rather than annual or monthly because seasonal values directly relate to season‐specific climate projections for the Cowichan Valley and for precipitation received in the rainfall manipulation experiments. We selected the two seasonal variables with the highest absolute loadings from PC axes 1 and 2, for a total of four variables, to include in vital rate model selection (Appendix S1: Table S1). Climate variables were standardized to a mean of zero and a SD of one, where one SD was 1–1.5°C for temperature variables or 75 mm for precipitation variables from the mean (Appendix S1: Figure S3). If the two highest loadings included temperature from the same season (e.g., summer maximum and summer mean temperature), we selected mean temperature. Selected variables from the PCA were mean winter, spring, and summer temperature and spring total precipitation. Distributions of climate variables can be found in Appendix S1: Figure S3.

Climate‐dependent vital rates

To evaluate costs of reproduction across time, climate, and rainfall treatment, we modeled vital rates with mixed‐effects models for the long‐term demographic data from 2017 to 2023. The four vital rates examined were size in t + 1 (growth), survival in t + 1, probability of flowering in t + 1, and flower production in t + 1, each as a function of size in t. Vital rates were modeled with the same error distributions as described in the clipping experiment. To select models that provided the best fit to the data, full models that included plant size in t, climate variable in t, flowering status in t (flowering or vegetative), rainfall treatment, and two‐ and three‐way interactions were compared with all possible model subsets using Akaike information criterion (AIC; 105 candidate models per vital rate). Including flowering status in t as a fixed effect in candidate models tested the effect of current flowering and reproductive effort on future vital rates. To simplify models and to make examining interactions among predictor variables tractable, we included only one climate variable in each vital rate candidate model. Preliminary analyses showed that size in t was a better predictor than size in t + 1 for vital rates in t + 1, indicating support for incorporating climate from the previous year. Early spring forbs, including the Primula genera, that overwinter are predicted to preform leaves and inflorescences prior to emergence in the growing season (Schnablová et al., 2021). Thus, we only included climate variables in year t and assumed that vital rates in t + 1 for this species are primarily influenced by the previous year's climate. All model subsets included size in t as a fixed effect and plot and year as random intercept effects. We selected the candidate model with the lowest difference in AIC between models (ΔAIC), but if there was more than one model with a ΔAIC of less than 2, we chose the model with the fewest number of parameters. All models were fit in R (version 4.2.2, R Development Core Team, 2022); mixed‐effect models were fit using the lme4 package (version 1.1–35.1; Bates et al., 2015).

Integral projection model

We used the best supported vital rate models outlined in the previous section to parameterize climate, reproduction, rainfall treatment, and size‐dependent integral projection models (IPMs) to integrate the effect of seasonal climate, reproductive effort, and rainfall treatment on the deterministic population growth rate, λ. We used a continuously size‐structured population of adult individuals with greater than one‐leaf and two discrete stages: seedlings and one‐leaf plants. The continuous IPM kernel describes the size distribution of adult plants in t + 1, across the possible range of plant sizes, based on functions for survival and growth of adult plants in t and the discrete probability of one‐leaf plants entering the adult population. Fecundity is estimated as the number of seedlings in t + 1 produced by the adult population in time t. IPM equations and discrete transition probabilities can be found in Appendix S1: Section S3. We built separate IPMs for each combination of parameters specific to the best supported vital rate models and estimated the deterministic population growth rate, λ, as the dominant eigenvalue of each IPM kernel. To evaluate the influence of costs of reproduction on λ, we compared λ from IPMs built from functions that incorporate a cost of reproduction, where it was supported at the level of the vital rate, to when reproduction carried no cost. We estimated 95% CIs for λ with parametric bootstrapping by resampling coefficients of each vital rate function 1000 times and recalculated λ for each bootstrapped replicate.

RESULTS

Clipping experiment

Reproductive effort, quantified as seed set in t, had a marginally significant effect on growth from 2021 to 2022, where seed production reduced plant size the following year (p = 0.073; Table 1; Figure 1). Plants that flowered but made zero seeds, or low reproductive effort (where plants initiated reproduction but did not produce seeds), were on average 12% larger than plants with medium reproductive effort (few seeds produced, or the 10th percentile of seed production) and 21% larger than plants with high reproductive effort (high seed production, or the 90th percentile of seed production; Figure 1). This effect did not interact with plant size in t (p = 0.376; Table 1). Reproductive effort in 2021 did not have a significant direct effect on the probability of flowering or flower or seed production in the subsequent year (2022), although few candidate models converged for the flower production models. Survival models had low instances of convergence due to high survival rates (131 of 141 plants survived). Overall, the probability of flowering and flower production increased with plant size. The effect of reproductive effort was not influenced by rainfall treatments for any vital rate (Table 1).

TABLE 1.

Results from likelihood ratio tests for variable selection in linear mixed‐effects models (growth) and generalized linear mixed‐effects models (GLMMs; probability of flowering, number of flowers, and number of seeds) for Primula hendersonii in the flower clipping experiment.

Variable removed	Growth		Probability of flowering		No. flowers
	χ2	p‐value	χ2	p‐value	χ2	p‐value
Plant size × total seeds in 2021 × rainfall treatment	0.277	0.871	2.465	0.292	0.842	0.656
Total seeds in 2021 × rainfall treatment	3.255	0.196	0.286	0.867	0.501	0.779
Plant size × total seeds in 2021	0.785	0.376	0.501	0.479	…	…
Plant size × rainfall treatment	0.283	0.868	3.845	0.146	0.023	0.989
Total seeds in 2021	3.221	0.073	3.914	0.141	1.002	0.606
Rainfall treatment	0.101	0.951	1.711	0.191	0.448	0.503

Note: All models included plot as a random intercept effect and plant size as a fixed effect. Variables listed in the first column refer to the variable removed from the nested models in stepwise selection, ordered from most complex to least; statistics are results from likelihood ratio tests (sample size = 141 individuals). The degrees of freedom for each likelihood ratio test are 1. Ellipses indicate models that did not converge, × indicates interaction terms, and bolded values indicate marginally significant values.

FIGURE 1.

Seed production reduces plant growth in the flower clipping experiment from 2021 to 2022. Lines indicate predicted growth under different amounts of seed production in 2021. The dark purple line indicates growth for 90th percentile seed production, medium purple line indicates 10th percentile seed production, light pink line indicates zero seeds produced. Points are individual plants and size of point indicates how many seeds were produced by the individual. The gray 1:1 line shows stasis (no change in plant size from 2021 to 2022). The log transformation is the natural logarithm.

Long‐term vital rates

Across the long‐term demographic study, the best supported model for individual growth included summer mean temperature in time t, flowering in time t, and rainfall treatment (Table 2). The effect of flowering was size‐dependent. Only large plants paid a cost of reproduction, albeit small: large plants (90th size percentile) that flowered in t were 1.6% smaller in t + 1 than those that were vegetative in t. However, small plants (10th size percentile) that flowered in t were 33% larger in t + 1 than vegetative plants (Figure 2a). Plants in irrigated treatments were, on average, 5% and 15% larger in t + 1 than plants in control and drought treatments, respectively. Summer temperature had a positive association with growth: plants were larger after years with warmer summers. Importantly, summer temperature interacted with rainfall treatment, such that cooler temperatures were associated with larger growth differences between irrigated and drought plots, and growth did not differ across treatments in warmer summers (Figure 2a).

TABLE 2.

Best fit models from model selection for climate and size‐dependent vital rates in the long‐term demographic data.

Vital rate	Climate variable in best fit model (scaled)	Fixed effects in best fit model	Parameter estimates
Growth	Summer mean temperature Summer mean temperature × rainfall treatment	log(rosette area) in t + 1 ~ log(rosette area) + p(flower) + log(rosette area) × p(flower) + rainfall treatment + summer mean temp. + rainfall treatment × summer mean temp.	Drought: y F  = 2.13 + 0.32z + 0.45x y V  = 1.54 + 0.32z + 0.58x
			Control: y F  = 2.21 + 0.20z + 0.45x y V  = 1.63 + 0.20z + 0.58x
			Irrigated: y F  = 2.26 + 0.18z + 0.45x y V  = 1.67+ 0.18z + 0.58x
Probability of flowering	Spring mean temperature	p(flower) in t + 1 ~ log(rosette area) + p(flower) + log(rosette area) × p(flower) + spring mean temp.	β(x) F  = logit(−4.44 + 0.92x −0.36z) β(x) V  = logit(−7.36 + 1.54x −0.36z)
Survival	Spring mean temperature	p(survival) in t + 1 ~ log(rosette area) + spring mean temp.	θ = logit(0.61 + 0.59x − 0.36z)
Number of flowers	Winter mean temperature	Number of flowers in t + 1 ~ log(rosette area) + winter mean temp.	φ = logit(0.54 + 0.20x − 0.08z)

Note: All models include plot and year as random intercept effects, and fixed effects are from time t; × indicates interaction terms in the fixed‐effects column. For all functions, plant size (log rosette area in square centimeters) in time t is indicated by “x,” temperature in year t is indicated by “z” (spring temperature in probability of flowering and survival functions, summer temperature for growth functions, and winter temperature for the number of flowers function), plant size in year t + 1 is “y,” probability of flowering in t + 1 is β, survival in year t + 1 is θ, and number of flowers in time t + 1 is φ. “F” and “V” subscripts on response variables indicate functions for estimates of flowering and vegetative individuals, respectively. Logit is the logistic function, where logit(k) = e ^k /(1 + e ^k ). Mixed‐effects models were fit using the lme4 package in R (version 1.1‐35.1; Bates et al., 2015).

FIGURE 2.

Size‐dependent vital rate models showing the influence of (A) summer mean temperature (t), rainfall treatment, and past reproduction (t) on growth, (B) spring mean temperature (t) on survival, (C) past flowering (t), and spring mean temperature (t) on the probability of flowering in t + 1, and (D) winter mean temperature (t) on flower production in t + 1. Warmer seasonal temperatures were associated with lower survival, probability of flowering, and flower production, but higher growth. For growth (A) and probability of flowering (C), solid curves show predicted relationships of plants that flowered in t, dotted curves show plants that were vegetative in t. Color of curve refers to scaled seasonal mean temperature (in °C, dark blue is cool or 1 SD lower than the 2013–2022 average, light blue is average, and orange is warm or 1 SD higher than the average). The gray 1:1 line in panel (A) shows stasis (no change in plant size from t to t + 1). The log transformation is the natural logarithm.

Probability of flowering in t + 1 was best predicted by size in t, spring mean temperature in time t, and flowering in time t (Table 2). Similar to the cost of reproduction detected for growth, the effect of flowering in t was size‐dependent and only large plants faced a cost of reproduction. Large plants that flowered in time t had a lower probability of flowering again in time t + 1 than vegetative plants, and small plants that flowered had a higher probability of flowering again. Threshold flowering size (the size at which 50% of plants flower) in t + 1 for low reproductive effort plants was 4% smaller than for high reproductive effort plants. Warmer springs were associated with lower flowering in t + 1 (Figure 2c). As an indicator of the effect of spring temperature on flowering, the threshold flowering size under cool springs was 37% smaller than warm springs (−1 and 1 SD from mean spring temperature, respectively). Rainfall treatments did not influence the probability of flowering, survival, or flower production. Survival was dependent on size in time t and spring temperature in time t (Table 2), and lower survival was associated with warmer springs (Figure 2b). Flower production in t + 1 was influenced by size and winter temperature in time t (Table 2), where warmer winter temperatures resulted in lower flower production (Figure 2d). See Appendix S1: Table S2 for AIC rankings of top candidate models for each vital rate.

Population growth

Population growth rates (deterministic λ) were highest in irrigated plots and lowest in drought plots across interannual variation in all seasonal climate variables (Figure 3), and past seasonal temperatures had varying effects on λ. Specifically, warmer spring temperature and cooler summer temperature were associated with declines in λ, and winter temperatures did not influence λ (Figure 3). The associated negative effects of cool summer temperatures on λ were exacerbated in drought treatments. Within drought treatments, IPMs built from demographic estimates of high reproductive effort showed slightly lower λ than IPMs built from demographic estimates of low reproductive effort, especially under cool summers. However, 95% CIs overlap (Figure 3), suggesting that even though costs were evident in growth and probability of flowering for large plants, they did not significantly influence λ under natural variation in seasonal climate or rainfall manipulation.

FIGURE 3.

Population growth rate (λ) in rainfall treatments across varying (A) spring, (B) summer, and (C) winter temperature. For each row, the indicated temperature variable varies while the others are held at the mean. The x‐axis is shown in units of SD, a zero x‐axis values refer to the mean temperature from 2013 to 2023, and absolute 1 and 2 values indicate temperatures that are 1 or 2 SDs from the mean. Bold lines show mean λ with 95% CIs obtained from 1000 parametric bootstrap samples from vital rate function coefficients. Color indicates rainfall treatment. Solid lines refer to models built from demographic functions of plants that flowered in time t. Dotted lines refer to models built from demographic functions of plants that were vegetative in time t. The gray horizontal line indicates a λ value of one, indicating stasis, λ values under one indicate population decline.

DISCUSSION

Costs of reproduction are predicted to depend on climate, but whether they influence how populations respond to interannual climate variation is unclear. We detected weak size‐dependent costs of reproduction in an observational demographic study in a long‐lived perennial. However, the observed costs did not influence population growth rates, nor did they meaningfully impact how population growth changed across variation in climate or rainfall manipulation. Climate variables influenced vital rates and population growth rates individualistically, and only hotter spring temperatures were associated with reductions in population growth rates. Thus, climate primarily drove population growth but not costs of reproduction in P. hendersonii.

Despite past research examining the influence of climate drivers on vital rates and population growth rates, the role of dormant and past season climates is rarely incorporated in demographic analyses (van de Pol et al., 2016). We found that all vital rates were climate‐dependent and responded individualistically to climate variables of the past seasons, in line with previous studies (Dalgleish et al., 2011; Iler et al., 2019; Williams et al., 2015). Notably, we found that for P. hendersonii, flower production and growth were influenced by dormant season (winter and summer) temperatures, but winter temperature did not influence population growth rates. Any explanations we have for how dormant season temperature influences P. hendersonii are conjecture, but there is evidence that dormant season climate can be more important for vital rates than growing season climate (Evers et al., 2021). One hypothesis is that the beneficial impact that hotter summers have on P. hendersonii may be mediated through the effects of summer temperature on other late season species, notably invasive perennial grasses that produce thatch that may negatively impact native species in subsequent growing seasons (Thomson, 2005; Williams & Crone, 2006). Although this hypothesis has not been evaluated in this system, warmer summers could reduce grass productivity through lower soil moisture (Smith & Williams, 2023), thereby decreasing the amount of thatch that grasses produce. On the other hand, cooler winters were associated with higher flower production; this could be due to chilling requirements or vernalization of the species or a benefit of cool weather when plants emerge (Anderson, 2023; Chouard, 1960). Collectively, these results support that dormant season climate is important for vital rates and population growth rates, and should not be excluded when examining the influence of climate on population dynamics.

Although precipitation has been shown to limit ecosystem productivity within the rainfall manipulation experiment at this study site, especially in the spring growing season (Smith & Williams, 2023), we did not detect effects of natural rainfall in any season on any vital rates. However, the rainfall manipulation treatments did influence both individual plant growth and population growth, mainly within the drought treatments. The magnitude of water manipulation ultimately relied on natural variation of precipitation, with the exception of the irrigated treatments in spring 2021–2023. This may have obscured our detection of effects of water availability in this study, where weak effects of manipulated rainfall were ultimately influenced by small differences in treatments due to unexpectedly low ambient rainfall. This highlights the need for both observational and manipulative climate studies to detect subtle impacts of climate. In a previous experiment in a Pacific Northwest savanna, experimental warming with open top chambers reduced population growth rates of annuals within the growing season, but treatments simulating small changes in spring precipitation did not influence population growth (Reed et al., 2021). However, there were key differences between our study and the previous study: the previous experiment was conducted much further south in the ecosystem's range using annual species under only 20% addition and 40% reduction in rainfall (Reed et al., 2021). We primarily saw that spring temperature was associated with larger shifts in population growth rates than winter temperature and summer temperature, likely because spring temperature influenced the vital rates (survival and probability of flowering) with high elasticities, or high proportional influences on the population growth rate (Appendix S1: Figure S6, Franco & Silvertown, 2004; Pfister, 1998). Climate projections for our study site indicate season‐specific changes in precipitation but warmer conditions across all seasons (Cowichan Valley Regional District, 2017), and our results imply that while increases in spring temperature would negatively affect P. hendersonii, warmer summers and wetter springs may increase population growth rates.

Observational and manipulative studies can lead to differences in the detection and strength of costs of reproduction (Dorken et al., 2025; Fox & Peter, 1991), and thus influence conclusions about the strength of costs for particular species. In support of this, we detected a size‐dependent cost of reproduction in the observational demographic experiment, and little evidence for a cost of reproduction in the clipping experiment. A key difference in the two methods is the way that reproductive effort is measured, that is, seed production in the manipulative clipping experiment as opposed to initiating reproduction in the observational data (Dorken et al., 2025; Obeso, 2002). Primula hendersonii has determinate growth and emerges in the winter with buds already formed, suggesting that stored resources are already allocated to reproduction prior to flowering, and making seeds is not the most expensive stage of reproducing. However, this may be specific to overwintering species with organ preformation, where resource storage from previous growing seasons influences current reproduction (Ehrlén & Van Groenendael, 2001; Schnablová et al., 2021). Due to small sample sizes in the clipping experiment resulting in models of some vital rates failing to converge, it is possible that we were not able to detect other existing costs of seed production. Alternatively, this weak influence of manipulating reproductive effort through flower clipping could be an adaptation to frequent floral and capsule herbivory by mammals and insects in this population (J. Loesberg, personal observation). In populations with a high risk of floral herbivory, low resource investment in flowers and fruits may buffer the negative effects of herbivory if flowers and fruits are lost (Ehrlén, 1991), which may have led to small differences in reproductive effort across clipping treatments if plants had already allocated resources prior to flowering. Thus, identifying when resources and effort are allocated to reproduction may be key to detecting costs, as well as considering the life histories of study species (Sun et al., 2022).

Costs of reproduction that are ameliorated or exacerbated by climate change can influence future population persistence and dynamics (Hamann et al., 2021). Overall, there was little evidence for costs in P. hendersonii: only large plants paid a cost for reproducing, but small individuals experienced positive effects of reproduction. Positive correlations are not uncommon in studies that have looked for costs of reproduction (Lehtilä & Syrjänen, 1995; Obeso, 2002; Stearns, 1989). Under abundant resources, costs can remain undetected; reproduction and future vital rates may be positively correlated, or costs can appear in complex ways (e.g., costs that differ with plant size; Miller et al., 2012; Obeso, 2002; van Noordwijk & de Jong, 1986). Theory predicts that costs of reproduction increase with plant size, where small plants allocate a lower proportion of total resources to reproducing than large plants to increase lifetime fitness (Hemborg & Karlsson, 1998; Obeso, 2002; Reekie & Bazzaz, 1992; Worley & Harder, 1996), even as other studies have found that small plants pay higher costs of reproduction than large plants (Miller et al., 2012). Regardless, the lack of change in population growth rates across reproductive effort may be driven by individuals that are able to flower at small sizes or young ages, suggesting that these “lucky individuals” contribute strongly to population growth rates (Snyder & Ellner, 2018). This has been detected in ungulates, where variation in individual quality masks costs of reproduction (Hamel et al., 2009; Weladji et al., 2008). Our combined rainfall manipulation experiment and clipping experiment aimed to control for the effects of individual heterogeneity and resource availability, but the year of the clipping experiment had low spring precipitation and particularly warm summer temperatures (Appendix S1: Figure S3). Failing to detect costs of reproduction across water manipulation treatment may have been driven by little difference between water availability in control and drought treatments.

Here we show that climate influences both vital rates and population growth rates, but not costs of reproduction of a long‐lived iteroparous wildflower. We add to the growing number of studies that show that costs of reproduction can be detected in long‐term observational data. We highlight that clipping experiments may be less practical, as sample sizes are limited to only reproducing individuals, are time intensive, and are difficult to maintain over a long time period. Future studies should incorporate further indirect responses of climate change, for example how neighboring plant species, which are also responding to changes in climate, influence target species responses.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Appendix S1.

Loesberg, Jenna A. , and Williams Jennifer L.. 2025. “Seasonal Climate Drives Population Growth but Not Costs of Reproduction of a Perennial Wildflower.” Ecology 106(12): e70240. 10.1002/ecy.70240

DATA AVAILABILITY STATEMENT

Data and code (Loesberg & Williams, 2025) are available in Figshare at https://doi.org/10.6084/m9.figshare.30038485.v2.