Summary of location and depth (m) of samples, total sequences before cleaning (Raw Reads) and after cleaning (Final Reads), observed richness (S_obs) computed as the total number of Operational Taxonomic Units (OTUs) clustered at 97% identity, and the percentage of singletons.

Total richness (S) was estimated using the Chao1 lower bound estimator [39] and using the Sichel distribution fitted to the count frequency data by the Bayesian method of [33] and selected from four alternative candidate models using the Deviance Information Criterion (DIC). Using the Sichel distribution, point estimates and 95% credible intervals (CIs) for S were obtained from the mean and (2.5%, 97.5%) quantiles of the posterior distribution sampled 150000 times by Markov Chain Monte Carlo (after a burn-in period of 100000 samples, see 33). The Required Sequencing Effort (RSE) to sequence 90% of the total richness was predicted by hierarchical simulation (see Materials and methods) and is quoted in terms of the number of final reads and as a multiple of the present sequencing effort. Point estimates and 95% prediction intervals (PIs) for RSE were calculated as the mean and (2.5%, 97.5%) quantiles from a set of 80 simulations using the Sichel distribution.