The end-Triassic extinction (~200 million years ago) marks the fourth greatest extinction event in Earth history.  Here, we use relative abundance distributions (RADS) to test whether decreasing diversity was gradual or sudden, and thus to test gradual versus catastrophic causal mechanisms of extinction.  We examine 5 plant communities containing 2965 macrofossil plant specimens from Kap Stewart Group strata (Greenland) leading up to the Triassic/Jurassic (Tr/J) boundary. For each assemblage, we determined which of four ecological models (geometric, lognormal, Zipf and zero sum multinomial) provided the best model RAD given the expected and observed samples. We then found the range of alternatives with log-likelihoods within 2.0 of the best hypothesis (i.e., 2 units of support).  Implicit to each of these models is the number of taxa with frequencies of (say) one in a million or greater.  Thus, the support bars for RADs also put support bars on diversity hypotheses that allow us to test more general hypothese about changes in diversity over time. 

The most likely number of species with frequencies of ³10^-6 shows a general decline over the five beds.  Correspondingly, we can strongly reject a hypothesis of consistent diversity given a hypothesis of continuous decrease.  However, the decrease through the first 20 meters is negligible; thus, we can reject the hypothesis of continuously decreasing diversity in favor of one with static diversity and then a sharp decrease over the last 4-11 meters of rock. This coincides with evidence for sharp environmental change in the final 10 meters of rock (defined by a globally registered isotopic excursion). These results support predictions of modern metapopulation theory and suggest that major ecological changes in Triassic vegetation, and perhaps ecosystem instability, were in place before plant species-level extinction, global temperatures and greenhouse gasses reached their peak at the Tr/J boundary.