Geng Report: On the Impact of Climate Change on Food Security in the Yellow River Basin — Highly Suspected Integrity Irregularities (DOI: 10.13866/j.azr.2026.02.17)

Academic fraud report · Geng Detector

Summary

This report assesses "气候变化对黄河流域粮食安全的影响" (Chen Feng, Bai Ting, Shanxi University of Finance and Economics), published in 干旱区研究 (Arid Zone Research), Vol. 43 No. 2, Feb 2026, DOI 10.13866/j.azr.2026.02.17, also posted on ChinaXiv:202603.00177v1. Overall verdict: highly suspicious. Two findings were confirmed after review: (1) Figure 2 shows a steep basin-wide decline in food-sustainability (2000–2016) that contradicts Figure 3d and the text, which report seven of nine provinces as stable or rising with only Henan declining; (2) the abstract states extreme heat has the largest negative impact on the middle reaches, while conclusion (4) claims the largest positive impact, contradicting Table 7's negative coefficient of -0.885*. Two additional findings carry benign explanations pending raw data: near-mirror-identical +10%/-10% sensitivity curves in Figure 4 (plausibly symmetric under an absolute-value rate formula), and anomalous coefficient/significance combinations in Tables 6–7 (e.g., -0.779* vs -0.013*** for downstream energy consumption across differently specified models). A last-digit distribution test was judged inapplicable to regularly rounded regression coefficients. Sample-size and breakpoint calculations check out. A programmatic Bayesian synthesis yields 91.6% posterior probability (BF = 207.53, 95% CI [59.7%, 98.4%]), though conditional-independence assumptions may overstate joint evidence. Verification requires raw data and original figure files; this is not a formal misconduct finding.

Verdict

Highly suspicious (🟠 / composite rating 🔴). Two substantive internal contradictions were verified verbatim in the source text (Figure 2 vs. Figure 3d/text; abstract vs. conclusion (4) regarding the direction of extreme-heat effects). Two further findings remain flagged but have plausible benign explanations pending raw-data verification. The report does not constitute a formal finding of academic misconduct.

Quantitative synthesis (automated): posterior probability 91.6% (prior 5%), 95% credible interval [59.7%, 98.4%], Bayes factor BF = 207.53 (decisive). Note: evidence items were treated as conditionally independent, which may overstate the joint strength of related findings. 15 leads were excluded after review as benign or non-applicable.

Key findings

Finding 2 (confirmed ✅): Figure 2 shows a pronounced basin-wide decline in the food-sustainability index over 2000–2016, yet Figure 3d and the text state that seven of nine provinces were basically stable, one (Shandong) rising, and only one (Henan) declining. No weighting scheme or data-processing explanation in the paper accounts for this whole-vs-parts directional divergence. Severity: 🟠.
Finding 3 (confirmed ✅): The abstract states extreme heat has a significant *negative* effect at the 5% level, strongest in the middle reaches; Table 7 shows a middle-reaches extreme-heat coefficient of -0.885* (negative). But conclusion (4) claims extreme heat's *positive* impact is largest in the middle reaches. Severity: 🟠.
Finding 1 (needs verification ⚠️): In Figure 4 (p. 416, ±10% sensitivity analysis, panels a/b), the +10% and -10% perturbation curves for all four dimensions are nearly identical in shape and inflection points. Programmatic image forensics (I1) detected multiple pairs of exactly duplicated blocks on that page. A benign explanation exists: the rate formula (Eq. 11) takes absolute values, so a locally linear weighting system could produce near-symmetric responses — but the paper offers no supporting statement; duplicated drawing objects with altered data sources is another possible cause. Severity: 🟠.
Finding 4 (needs verification ⚠️): Table 7 downstream energy consumption coefficient is -0.013*** while Table 6 downstream shows -0.779* for the same variable — a ~60× magnitude difference. Middle-reaches mean temperature terms (-4.140 linear, 4.259 quadratic) are large but non-significant. Benign explanations: the two tables use different model specifications (Table 7 adds extreme heat, extreme cold, frost days), tiny coefficients can be highly significant with very small standard errors, and the middle-reaches subsample (2 provinces) limits degrees of freedom. Severity: 🟡.
Finding 5 (insufficient basis ⚠️): A last-digit distribution test over 63 tabulated values showed uneven distribution and odd/even imbalance (19/63). However, the data are regularly rounded regression coefficients and weights, for which last-digit/Benford-type tests are not applicable; this lead alone supports no conclusion.

Evidence highlights

Abstract vs. conclusion direction contradiction, anchored to Table 7 middle-reaches coefficient -0.885* — verbatim verified; the largest evidence contribution (29.3% of the synthesized posterior) alongside Finding 2 (30.8%).
Figure 2 (whole basin declining) vs. Figure 3d and text (most provinces stable/rising) — directional whole-vs-parts contradiction.
Figure 4 mirrored ±10% sensitivity curves with programmatically detected duplicated image blocks (I1).
Cross-table coefficient inconsistency: -0.779* (Table 6) vs. -0.013*\*\* (Table 7) for downstream energy consumption.
Checks that passed / non-findings: sample sizes self-consistent (9×22=198; excluding 2001–2002, 9×20=180, matching Table 5); breakpoint arithmetic self-consistent (0.74×3.88+10.56≈13.43℃; 1.448/(2×0.979)≈0.74); PRNU weak-similarity results (I5–I13, NCC ≈ 0) and single-page noise variance (I3) consistent with benign pipeline effects; no biological imagery present.

Notes

Paper: "气候变化对黄河流域粮食安全的影响", Chen Feng & Bai Ting, School of Resources and Environment, Shanxi University of Finance and Economics. 干旱区研究, Vol. 43 No. 2, February 2026. DOI: 10.13866/j.azr.2026.02.17. Preprint: ChinaXiv:202603.00177v1 (posted 2026-03-24).
Recommended actions: request raw panel data, CRITIC weighting computation, full regression outputs, and underlying data/plotting source files for Figures 2–4; request author explanation and correction of the abstract/conclusion (4) contradiction; submit correction/verification requests to the journal; escalate to the authors' institution depending on the response.
Limits: AI-assisted analysis for academic discussion only; prior fixed at 5% with conditional-independence assumptions that may inflate the joint evidence; final misconduct determination requires institutional investigation and the authors' right of reply must be respected.