Phase 3 — Iberian regional replication

Phase 3 asks: does the same thermal-exposure → extirpation signal still emerge when the curated continental dataset is replaced with open GBIF occurrence records for the Iberian Peninsula?

This is a Replicability check in FORRT vocabulary — same code (the validated Python port from Phase 2), different conditions (regional zoom to a southern-Europe subregion, different occurrence data source).

Why Iberia¶

Many Bombus species reach their southern range limit on the Iberian Peninsula. If Soroye’s mechanism is real — that the frequency of historically extreme temperatures, not mean warming, drives extirpation — then Iberia is exactly where one would expect a stronger signal than the continental mean. The Phase 3 result either supports that prediction or it doesn’t.

What was changed¶

Headline result¶

sc_TEI_delta = +0.4792, 95 % CI [0.266, 0.693]   (mixed-effects, VB)
sc_TEI_delta = +0.2943, p = 0.007                (plain logit)
n = 528 species × cell observations across 31 Bombus species, 99 sampled cells

The claim replicates regionally and with a stronger effect than the continental mean. The Iberian coefficient is approximately 3× larger in magnitude than Phase 2’s +0.15 continental average.

Notable differences from Phase 2:

• sc_TEI_delta is larger (+0.48 vs +0.15).

• sc_TEI_bs is also larger (+0.61 vs +0.21) — Iberia is concentrated near the warm end of Bombus thermal envelopes.

• The hot-edge × thermal-change interaction (sc_TEI_bs:sc_TEI_delta) is positive but not significant on the smaller Iberian sample. This is most likely a sample-size / range-restriction effect rather than a real difference in the underlying mechanism.

The nanopubs¶

Replication Study¶

View Phase 3 Study on Science Live →

Replication Outcome — Validated, High confidence¶

View Phase 3 Outcome on Science Live →

CiTO link to Soroye 2020¶

View Phase 3 CiTO on Science Live →

Caveats¶

• Iberian sample size is small (528 species × cell observations) so credible intervals on individual coefficients are wider than on the continental Phase 2 dataset. The point estimate (+0.48) should be interpreted with this in mind.

• Variational Bayes posterior may underestimate credible-interval width by approximately 10–20 % relative to full MCMC.

• GBIF Iberian Bombus coverage before approximately 1920 is sparse, which reduces the temporal span of the effective baseline window relative to Soroye’s continental dataset.

What this enables¶

With Iberian replication confirmed, the same validated pipeline becomes a tool — projecting onto future climate, identifying refugia, applying the mechanism to other thermally-sensitive insect taxa. See Discussion for what’s next.