Propagation of uncertainties in mesocosm experiments on ocean acidification
Observations from different mesocosms exposed to the same treatment typically show variability that hinders the detection of potential treatments effects. To unearth relevant sources of variability, I developed and performed a model-based data analysis that simulates uncertainty propagation. I described how the observed divergence in the outcomes can be due to the amplification of differences in experimentally unresolved ecological factors within same treatment replicates. Three independent ocean acidification experiments on the response of phytoplankton to high CO2 concentrations in aquatic environments were used as tests cases. I first simulated the dynamics of the mean phytoplankton biomass in each treatment and detected acidification effects on the timing and intensity of the bloom in spite of the so far negative results obtained by statistical inference tools. By using the mean dynamics as reference for the uncertainty quantification, I showed that differences among replicates in parameters related to initial i) plankton community composition and ii) nutrient concentration can generate higher biomass variability than the response that can be attributed to the effect of elevated levels of CO2. I calculated confidence intervals for parameters and initial conditions. They can serve as estimation of the mesocosms tolerance thresholds below which uncertainties do not escalate into high outcomes variability. This information can improve the detection of treatment effects in next generation experimental designs and contributes to the ongoing discussion on the interpretation of controversial results in mesocosm experiments.