Why the Iberian coefficient is stronger¶
Phase 3’s sc_TEI_delta = +0.48 is roughly three times the magnitude of
Phase 2’s continental mean of +0.15. This isn’t a contradiction — it’s
exactly what Soroye’s mechanism predicts.
The mechanism says extirpation risk depends on how often local temperatures exceed a species-specific historical limit. Two factors amplify the signal on the Iberian Peninsula:
Many Bombus species reach their southern range edge there. A species’ baseline thermal position (
sc_TEI_bs) is closer to its hot limit than at the continental average. Phase 3 corroborates this:sc_TEI_bsis +0.61 on Iberia versus +0.21 in Phase 2.Recent warming over the Iberian Peninsula has been substantial. Combined with the elevated baseline thermal position, even moderate warming pushes more local temperatures above species-specific limits.
The Phase 3 result is therefore a stronger test of the mechanism, not an artefact of regional zoom or different data.
What about the hot-edge × thermal-change interaction?¶
In Phase 2, sc_TEI_bs:sc_TEI_delta is positive and significant — species
already near their warm limit are more sensitive to additional warming.
On the Iberian subset that interaction collapses to +0.03 and becomes
non-significant.
The most parsimonious explanation is range restriction: the Iberian
sample concentrates species at the warm end of their distribution, so
there’s less variation along sc_TEI_bs for the interaction to act on.
Statistically the cells are clustered in the upper-right of the (TEI_bs,
TEI_delta) space, leaving the slope of the interaction poorly identified.
We do not read this as evidence against the mechanism — only as a known limitation of testing an interaction term on a small, range-restricted sample.
What the chain enables — future research¶
A claim that has been independently replicated becomes a tool. Three follow-up directions seem most promising:
1. Future-climate projection¶
Project the validated pipeline onto future climate scenarios — most
naturally Destination Earth Climate Digital Twin
(~5 km resolution, EU FAIR-aligned). The TEI definition extrapolates
naturally: keep the same per-species historical thermal limits, swap in
projected tasmin / tasmax for the future window, predict from the
v0.2.0 mixed-effects model.
This produces a spatially explicit map of where Iberian (or pan-European) Bombus extirpation risk will rise, and where conditions may fall back within thermal limits — i.e. candidate climate refugia. A separate companion repo and a new FORRT chain, citing the v0.2.0 nanopubs as methodological provenance, would be the natural shape of that work.
2. Cross-taxon transfer¶
The same pipeline can be applied to other thermally-sensitive insect taxa
that have GBIF occurrence coverage and historical climate baselines:
solitary bees, butterflies, hoverflies. The Snakefile + Dockerfile +
parameterised OUT_SUBDIR make adding a new taxon a matter of swapping
in the cleaning script for that group.
3. Conservation prioritisation¶
By overlaying the projected risk maps with protected-area boundaries and known refugia, this analysis can flag conservation priority areas where intervention would protect species from thermal-exposure events — the “manage habitats to reduce exposure to the growing frequency of temperatures that are extreme relative to species’ historical tolerances” recommendation Soroye et al. close their paper with.
What this work doesn’t do¶
It does not test the mechanism on individual species — the GLMM treats species as a random effect rather than fitting per-species coefficients. Per-species rankings of vulnerability would need a follow-up study.
It does not evaluate alternative climate predictors (drought indices, diurnal range, extreme-event metrics). Soroye’s TEI is a monthly-resolution, single-statistic predictor; finer climate variables may yield a stronger or sharper signal.
It does not quantify the contribution of land-use change. Soroye’s paper controls for land use and reports the climate effect as independent of it; a parallel land-use replication on Iberia would require regional land-use rasters we have not assembled here.
Reflection on the FORRT chain itself¶
Two practical points worth surfacing for anyone doing similar work:
Paper-rooted vs question-rooted chains start differently. If there’s an upstream paper, start with Quote-with-comment + AIDA. If you’re asking your own question, start with PCC question. Both end at the same Claim → Study → Outcome backbone.
One repo can host two replication branches. When a single Python pipeline is exercised on two datasets (Phase 2 = Soroye’s data, Phase 3 = GBIF Iberia), the cleaner shape is one Claim with two Study + Outcome branches, not two separate repos. The 95 %-shared code stays in one place; the chain on Science Live stays navigable.
Both observations are now folded into the FORRT chain-design heuristics captured in our memory for future projects.