Fig. 1: Study sites, treatments and conceptual scheme depicting the predictions.

a Study sites. b Treatment combinations designed to manipulate the quantity of rain entering the ecosystem (50–200%) and distribution of rainfall frequency (10–300%) relative to average ambient conditions of each site (μ = 1, k = 1, in the centre of the table). Gradients of rainfall conditions, from severe drought to frequent rainfall, are represented by red and blue colours. k is the dispersion parameter (a measure of evenness in the frequency distribution of rainfall), and μ is the mean parameter (a measure of mean daily rainfall). These parameters were calculated per each site based on rainfall patterns of the last five years before the beginning of each experiment (see “Methods”). Stronger impacts of rainfall events are predicted when their components depart from current scenarios. c We predict that larger predators are more sensitive in environments experiencing infrequent and low rainfall amount (low k, low μ), characterized as extreme drought conditions, than smaller organisms. Under these conditions we expect communities dominated by smaller organisms (detritivores and filter feeders), which tend to decrease in standing stock under more favorable conditions (ambient) via top-down control. In contrast, more frequent rainfall and high rainfall amount (high k, high μ), characterized as heavy rainfall, could impact lower trophic levels through hydrodynamic perturbations (e.g., overflow of nutrients and basal resources). Bigger predators could be more resistant to flooding, but their biomass could decrease slightly via bottom-up effects. d Changes in standing stock, in turn, could alter predator–prey mass ratios (pyramid shape). Drought, ambient and heavy rainfall conditions could cause bottom-heavy, invariant and top-heavy biomass pyramids, respectively.