Data analysis
For each species and container, we calculated the allocation to generative reproduction (RA) as the number of flowering ramets divided by the number of all ramets (over all individuals present in each container). For tussock species, we also calculated the RA for each founding individual plant in a container. The RA of species and individuals was calculated as the mean RA over the years (i.e. 2012, 2013, 2014 and 2015). To express whether the reproductive allocation of species and individuals was stable or highly variable over the years, we calculated the stability of RA as its standard deviation divided by its mean. Lower stability of RA under the interannual drought regime indicates increasing interannual variation in RA, as predicted by the resource switching hypothesis. Lower stability under the permanent drought regime indicates that species invest more in flowering after accumulating resources over years, as predicted by the resource budget hypothesis.
To estimate interspecific RA synchrony we used correlation among species RA over years and calculated it for each container as RA synchrony =\(\sum_{i}^{n}{\left[r\left(RA_{i},\overline{RA_{j}}\right)\right]/n},\ \)where the value r is the correlation coefficient between \(RA_{i}\) of species i and the mean RA of the remaining species j in the container \(\overline{RA_{j}}\) and n is the number of species in the community. The synchrony index reaches a maximum of 1 when species are perfectly synchronised in RA and a minimum of -1 when species are perfectly asynchronised. It equals 0 if species RA are temporally independent. Accordingly, we also calculated intraspecific RA synchrony, where \(RA_{i}\) is the RA of species individual i and\(\overline{RA_{j}}\) is the mean RA of the remaining individuals of the given species j in the container, and n is the number of individuals of the given species in the container.
The effects of hydrological regime and dominant species removal on species mean RA, its stability and temporal shift were analysed using Bayesian mixed linear models. We also tested the effects of these treatments on RA synchrony among all and subordinate species (see Supporting information S3 for methodological details). Treatments were included as fixed variables and blocks as a random variable in the models. We described the effects of the model variables from the parameter estimates as the 89% highest density interval (HDI) (because the 95% HDI is unstable when the effective sample size is less than 10,000), the probability of direction (PD), and the region of practical equivalence (ROPE) (Makowski et al. 2019). To visualise the pairwise Pearson correlations of RA between species under different treatments, we used the chord diagram. All data analysis was performed using R Statistical Software (version 4.3.3; R Core Team, 2024).