Translating bumblebee dialect…

How has RMBL research changed over time?

Several views of the basin’s research across eras. The Corpus context section traces scale — how the published record grew, how well-covered it is by full text, how the community expanded. The Disciplines and Methodological approaches panels measure how the breadth and evenness of research shifted. The Community compositionchart maps the generational layers of authors active in each era. Together they sketch how the basin’s research has changed over the past 75+ years.

Research only All sources

Corpus context

Six trends framing the diversity story above: how the corpus grew, how well-covered it is by full text, how the research community changed, and how internally connected the literature became. The items-per-era chart respects the research lens (publications + datasets); the other five are inherently publication-bound (no “documents version” of average co-authors), so they show publications regardless of lens.

Items per era

1,435

Total publications + datasets dated within each era.

Full-text coverage

79%

Share of publications with substantive full text — frames the diversity reliability.

References per paper

54.1

Average extracted references per publication — connection to the broader literature.

Unique researchers

1,627

Distinct authors publishing in the era — community growth signal.

Co-authors per paper

4.2

Average authors per publication — collaboration signal.

Internal citations

10%

Share of references that point to other RMBL publications/documents/datasets — the corpus citing itself.

The next two panels both report two complementary diversity measures, in “effective number of categories” units — interpretable as “as if there were Nequally-weighted categories.” A single number can’t capture both the breadth of a long tail and the evenness of the dominant categories, so the two together tell the full story.

Shannon counts every category proportional to its share — sensitive to the long tail. Answers “how many categories are in play, broadly?”
Inverse Simpson emphasizes the dominant categories — the long tail barely contributes. Answers “how evenly distributed are the common categories?”

When the lines diverge, you learn where change is happening. Shannon rising faster than Inverse Simpson = more small categories appearing in the tail. Inverse Simpson rising faster = the dominant categories are becoming more evenly distributed without much change in breadth.

Disciplines

Research disciplines represented in extracted concepts (population ecology, hydrology, evolution, biogeochemistry, …). The discipline lens for the diversity question.

Not enough reliable eras to summarize a trend.

Composition

Share of mentions in each disciplines per era. Numbers under each bar are total mentions for that era. Categories are grouped by domain: earth sciences (cool palette) at the bottom of each bar, life sciences (warm palette) above, cross-cutting categories on top.

Methodological approaches

Protocol categories — sampling, measurement, experimental, computational, observational, analytical, laboratory — capturing how the research was conducted.

Broad diversity (Shannon)

5.9

vs. 5.0 in 1996–2000 ↑ +0.9

Top-category evenness (Inverse Simpson)

5.3

vs. 4.6 in 1996–2000 ↑ +0.7

Community composition by cohort

Pure measurement, no inference. Each bar is the research community active in that era, segmented by the era of each author’s first publication in the corpus. Reads as the community’s generational layers: deep cohorts at the bottom carry institutional memory; the lighter top layer is the era’s new arrivals.

1,253

new researchers first published in the 2021–25 (77% of 1,627 active)

374 continued from earlier cohorts — the long tail of researchers whose first publications go back as far as Pre-1950 contribute to institutional memory.

Generational layers

Total active researchers per era. Numbers under each bar are the active community size. Author identity is keyed on (family + given) name pairs — small inflation possible from spelling variations; the trend is robust.

Caveat: Eras before the 1990s have thin extraction coverage (the full-text PDF coverage gap discussed in the broader plan), so their effective-N estimates are based on very few mentions and should be read as suggestive rather than definitive. Sparse eras are dimmed in the charts; hover any point or bar segment for the underlying mention count. The reliable signal is the 1990s onward.