Rarefaction vs extrapolation for temporal species richness comparison

Answer

Coverage-based rarefaction (Chao & Jost framework) is the better-supported choice for temporal richness comparisons when sampling effort varies across years, but the corpus does not show that it is more accurate in every case, and it measures a different thing from size-based rarefaction. The framework papers say the two standardisations give the same qualitative richness ordering for species richness while the coverage-based one is more efficient, needing smaller samples to detect where accumulation curves cross 6d41d693. A record summarising later work reports that the two can lead to different conclusions about temporal trends because size-based rarefaction reflects changes in community size and coverage-based rarefaction reflects changes in composition f7cc0c5c. In practice, a 27-year US stream monitoring analysis used both: sample-size rarefaction to 300 individuals for site-level richness and coverage-based rarefaction at 70-80% coverage for regional richness, and found density down 11% but genus richness up 12.2% 0fac60eb.

Summary

Two statements about agreement between the methods are not matched. The framework paper states that, for species richness, the two methods always give the same qualitative ordering 6d41d693, while the record on Shimadzu 2018 states that they can lead to different conclusions about temporal trends f7cc0c5c. These are not comparable: the first is a statement about ordering two assemblages at a given sampling depth, the second about the direction of change through time, and the corpus does not contain a dataset analysed both ways that would settle it. The same record also reports a counterpoint that coverage-based rarefaction does not respond predictably when data have strong internal structure such as highly dominant species f7cc0c5c. A field comparison of four community assessment methods found coverage-based curves confirmed that differences between methods were due to the methods and not to sample completeness 153719d7. Extrapolated species richness is trustworthy only up to about twice the reference sample size 6d41d693.

Evidence

• Chao et al. 2014 (Ecological Monographs) - unifies size-based and coverage-based rarefaction and extrapolation for Hill numbers; same qualitative ordering for richness, coverage-based more efficient; only coverage-standardised diversity obeys the replication principle; richness extrapolation reliable to about double the reference sample. 6d41d693
• Chao & Jost 2012, with Shimadzu 2018 and a 2025 preprint (as summarised in one record) - coverage-based rarefaction less biased under unequal effort; the two methods quantify different aspects of change and can give different trends; coverage-based rarefaction unpredictable with strong internal structure. Secondary summary. f7cc0c5c
• Rumschlag et al. 2023 (Science Advances) - 27 years of US stream data; site-level sample-size rarefaction to 300 individuals, regional coverage-based rarefaction at 70-80% coverage; density down 11%, richness up 12.2%, insect density down 23.3% and insect richness down 6.8%. 0fac60eb
• Cox et al. 2017 - four assessment methods compared with both curve types; coverage-based curves showed method differences were not due to completeness; spatial method comparison, not a time series. 153719d7

Methodology

All four verified facts are used. I separated claims about ordering assemblages, claims about trends through time, and applied case studies, and I did not pool the percentages from the stream dataset with anything else because no other record reports a trend. The Shimadzu and 2025 statements are second-hand through one record and were not checked against their originals here.

Limitations

• No record runs one monitoring dataset through both methods and reports the trend each gives, so accuracy for temporal comparison is not measured directly 0fac60eb.
• The claim that the methods can give different trends rests on a secondary summary of Shimadzu 2018 f7cc0c5c.
• The Cox study compares sampling methods in space, not time points 153719d7.
• The stream analysis uses the two methods at different scales, so it cannot be read as a head-to-head 0fac60eb.