Shuffled-year series give about 80% of the Blowes-style turnover rate and most of its marine excess in 450 BioTIME 2.0 studies; directional turnover is 0.055 Jtu per decade in old and new studies

Blowes et al. (2019) used BioTIME v1. They reported fast species turnover (about 0.28 Jtu per decade), faster turnover in marine than in terrestrial assemblages, and the most extreme changes in the tropical marine band. We did these tests again in BioTIME 2.0 (April 2025). We compared 221 studies that were in BioTIME v1 with 229 studies that were new in BioTIME 2.0 (out of sample). Assemblages were study x ~100 km2 cells with at least 5 sampled years and sample-based rarefaction (24,991 assemblages). The unit of inference was the study, with a study bootstrap. The Blowes-style rate (slope of Jtu to the first year, first year = 0) reproduced: 0.30 per decade (v1 studies) and 0.25 (new studies). But when we shuffled the year order in each series, the rate was 0.25 and 0.20, about 80% of the observed rate. Simulations confirm that series with no trend give large Blowes-style rates. The directional part (observed minus shuffled) was 0.055 [0.039, 0.072] in v1 studies and 0.056 [0.041, 0.070] in new studies. An all-pairs lag slope gave 0.079 and 0.082. In v1 studies, marine minus terrestrial was +0.236 [0.134, 0.343] for the Blowes-style rate, but +0.217 of this came from the shuffled series. For the directional part it was +0.019 [-0.017, 0.054] (v1) and +0.023 [-0.016, 0.064] (new). With 3 or more years, a small directional marine excess was possible (+0.035 [0.004, 0.067] pooled). We did not find a tropical marine hotspot in any metric (10 and 8 tropical marine studies). Mean log-richness change was zero in both sets. Thus the directional turnover rate is reproducible out of sample, but it is about one fifth of the Blowes-style rate. The marine excess comes mainly from higher year-to-year dissimilarity, which can be real fluctuation or incomplete detection. Limitations: few tropical studies, the two sets differ in taxa, and our grid and study-mean statistics differ from the original Bayesian models.

BioTIME · temporal turnover · Jaccard dissimilarity · replication · out-of-sample test · marine versus terrestrial · latitude · permutation null · biodiversity change · rarefaction