Responses to abiotic and biotic factors
All data analysis was conducted using R version 3.5.3 (R Core Team 2019). To quantify plot-level abiotic conditions we ran a Principal Components Analysis (PCA) including canopy closure, litter cover, soil pH, and phosphorous, ammonium, nitrate, and potassium content (chosen as the most important soil nutrients for plant growth). The first PC axis (PC1) explained 54.2% of the variation in measured abiotic conditions and was mainly loaded with canopy closure, litter cover and soil macronutrients (phosphorous and nitrate). PC2 explained 18.5% of the abiotic variation and was mainly associated with soil pH (see Figure S3).
To assess how demographic rates correlate with abiotic conditions and plant-plant interactions, we modelled emergence, survival, seed production, and population growth rate separately per species using mixed-effects models. Emergence was modelled as the number of seeds that emerged, accounting for the number of seeds that were sown into each replicate, and survival was modelled as a binary outcome of whether a focal plant survived to produce at least one seed (viable or inviable). Due to low emergence fractions and hence insufficient focal plants to model survival and seed production, Podolepis lessonii was only assessed for emergence responses, whereas the remaining eight species were modelled for all demographic rates.
Prior to modelling, we log-transformed total neighbour abundance to improve linearity with vital rates and standardised all continuous explanatory variables to a mean of 0 and standard deviation of 1. Block and plot (nested in block) were included as random effects in all models. An observation-level random effect (OLRE) of sub-plot ID was nested within plot for emergence models to assist with potential overdispersion for all species, except for the G. rosea model which was not over-dispersed and including the OLRE led to convergence issues.
Emergence and survival were modelled using binomial errors and logit-link functions with the “glmer” function from the ‘lme4’ package (Bates et al. 2015). Seed production was modelled with negative binomial error distributions and log link functions using the “glmmTMB” function from the ‘glmmTMB’ package (Brooks et al.2017). Population growth rate was log-transformed and modelled with a normal error distribution using the “lmer” function from the ‘lme4’ package.
To allow for quadratic responses to the environment, we initially modelled each vital rate with main and quadratic terms for PC1 (soil fertility, canopy closure and litter cover), PC2 (soil pH) and neighbour abundance. Where quadratic terms were significant (see Table S1), they were retained in all subsequent models. Emergence was modelled in response to PC1 and PC2 only (because it was measured before the watering and neighbour treatments were implemented). Survival, seed production, and population growth were modelled in response to PC1, PC2, watering treatment (dry, ambient, or wet) and neighbour abundance (for survival and seed production) or presence (for population growth). The presence of Cuscuta campestris , a parasitic invasive annual plant, was included as a covariate as it appeared to impact host plant performance. To allow for interactions between factors, we included all pairwise interactions between watering treatment, PC1 and neighbour abundance in a full model, and then removed non-significant interaction terms to create the final models from which results are reported.
Results figures were plotted in base R or using the “ggplot” function in the package ‘ggplot2’ (Wickham 2016) and tables were built using the “kable” function in kableExtra (Zhu 2021). We calculated marginal and conditional pseudo-R2 values using the “r.squaredGLMM” function in the ‘MuMIn’ package (Bartoń 2022) to estimate the proportion of variance explained by fixed effects and combined fixed and random effects (Nakagawa & Schielzeth 2013). All data and R code to reproduce our results are available at https://github.com/acatling/Perenjori_watering_exp.