Habitat configuration dominance varies by biodiversity facet in European amphibians

This repository accompanies a continental-scale assessment of habitat fragmentation and climate change on amphibian biodiversity across 31,419 retained 10-km cell–time combinations in the EEA39 domain (39 European Environment Agency countries; ~34–72 °N, 25 °W–45 °E) over 1990–2018. The study uses explainable machine learning (XGBoost with **native missing-value handling**, TreeSHAP) and a canonical variance partition, under spatially blocked five-fold cross-validation, to address three questions for nine biodiversity response variables: whether the primacy of habitat configuration over habitat amount is facet-dependent (the Habitat Amount Hypothesis); whether configuration's biodiversity value couples to water balance rather than temperature; and whether compound fragmentation × climate signals are stable couplings or artefacts of analytic choice that dissolve under multi-filter robustness scrutiny. The archive contains the analysis code and derived data: native-missing-value XGBoost fit on the cells with **observed** DeltaFragmentation (n = 23,883; no median imputation), coverage-standardized richness from iNEXT, a hybrid Moura/Jetz + AmphiBIO trait matrix with an AmphiBIO-only sensitivity, Faith's PD with a Portik et al. (2023) anuran-subset sensitivity, and a data-distribution Friedman's H² interaction screen validated by a 100-permutation spatial-fold null, 10-seed Monte Carlo stability, and an observed-versus-imputed missing-data sensitivity. A per-response RandomForest-vs-XGBoost model-selection routine (`scripts/ml_model/train_rf_xgboost.py`) is included as an **auxiliary** tool used only by the secondary sensitivity analyses; the nine headline response models, all SHAP/HAH results (and the canonical variance partition that corroborates them, `varpart_hah.py`), the fragmentation × climate interaction gate, and the reported performance come exclusively from the native-missing-value XGBoost pipeline (`C14_xgboost_pipeline.py`, `C15a_shap_export.py`).