Review Papers Claim by Claim

 Quick Explanation Copied

 Long Explanation

What MInOSSE Does

MInOSSE (Massively Interpolated Occurrences for Species Spatial Estimation) is a two-step, model-based method for reconstructing fossil species' geographic ranges without environmental predictors: minosse.data builds predictor species' probability maps via Ordinary Kriging (intamap) over presences and simulated pseudo-absences, and minosse.target predicts the target's range using Regression Kriging with XGBoost as the deterministic model, thresholded by nine binarization metrics MInOSSE method paper. The design exploits Bottjer et al.'s taphonomic-control theory: coeval species' occurrences can serve as indirect evidence of the target's presence.

Case-study quality fits are strong: the woolly mammoth (24–14 kya, LGM) yielded RMSE 0.23, 79.04% explained variance, and AUC 0.97; the European roe deer (130 kya–present) yielded RMSE 0.150 and AUC 0.986, correctly splitting the range east/west of the Urals in line with C. pygargus allopatry Case-study validation metrics.

How the Claims Were Tested

Two simulation families with realistic taphonomic bias (bias intensity inversely proportional to range size): Eco-sims (26 species, 100 fossil localities, three site scenarios, grid resolutions 100–300 km plus nearest-neighbour-derived) and ST-sims (100 species, 100–200 localities, four clustering grades). MInOSSE's TSS significantly exceeded both MCP and Alpha-Hull (ANOVA F=93.64, p<<0.001), with the advantage largest when target presences were scarce — precisely where polygons fail. Sensitivity analyses (permutational MANOVA) showed co-occurrence-based predictor selection and grid resolution generally did not change TSS significantly, except under LGM-based site distributions (F=5.015, p=0.002) Simulation comparison statistics. Predicted areas averaged 2.53× the true range across thresholds (1.96× after dropping worst performers), versus 0.89× for MCP and 0.53× for Alpha-Hull — a relevant overestimation caveat.

Blind Spots and What Would Disprove It

The authors themselves flag two load-bearing assumptions: (1) target and predictors must be genuinely coeval — diatopy (spatial mismatch) and diachrony (temporal mismatch) between target and predictor localities inflate prediction error; (2) target–predictor spatial associations must remain stable over the interval, which fails for dynamic paleocommunities (e.g., Ordovician brachiopod niche shifts cited via Stigall 2012). The simulated fossil records also cannot fully mimic real-record complexity, and the Eco-sims' LGM scenario derived both species and sites from the same real fossil data, risking non-independence Author-stated assumptions and caveats. Falsification would follow from showing hull methods match MInOSSE TSS under controlled sparse-record conditions, or that accuracy collapses when predictor associations are deliberately perturbed. The EcoPast R package and simulation scripts are publicly released at github.com/francesco-carotenuto/EcoPast, supporting reproducibility EcoPast R package release.

 BGPT Paper Review

Study Novelty

Methodological rather than mechanistic paper; explains why kriging+XGBoost outperforms hulls (spatial autocorrelation, extrapolation beyond fossil sites) but does not model the underlying biogeographic processes.

Scientific Quality

Rigorous simulation design with sensitivity analyses, threshold replication, and honest assumptions discussion; red flags are predicted-area overestimation (2.53× true range across thresholds) and circularity risk in the LGM-based simulation scenario; case studies validated by AUC but not against independent ground truth.

Study Generality

Applicable wherever environmental predictors are unavailable (deep time, poor surveys) across any taxon with coeval occurrences; demonstrated on mammals and an ammonite, though validated mainly in Eurasian settings.

Study Usefulness

Fills a genuine gap for deep-time macroecology where SDMs fail; directly useful for range-evolution and extinction-risk studies; open R package lowers adoption barrier.

Study Reproducibility

Full R package (EcoPast) with simulation and comparison scripts publicly released; GROBID-processed text confirms parameter settings and minimum data requirements (30 kriging points, 50 RK points) are documented.

Explanatory Depth