Concerns regarding 'A Convective Initiation Nowcasting Algorithm Based on FY-4B Satellite AGRI and GHI Data' (Atmosphere, DOI: 10.3390/atmos17040380)

Academic fraud report · Geng Detector

Summary

Highly suspicious. No direct evidence of figure fabrication was found, and internal arithmetic in Tables 5 and 7 (e.g., POD=138/(138+28)=83.13%) is internally consistent, indicating that the authors possess the technical competence to construct (or curate) plausible-looking numbers. However, multiple severe inconsistencies undermine the paper's credibility. The case study at 23:00 UTC on 4 August 2023 directly contradicts the authors' own statement that GHI data were unavailable for Sichuan due to Typhoon Doksuri's tasking through early August, strongly implying figure reuse or data splicing from a different period. Table 3's GHI minute series systematically omits minutes ending in 5 and 0, which is physically implausible for a ~1-minute sensor. Section 3.1 reports '95% confidence intervals' but Table 7 prints a '96% CI', an irreconcilable statistical contradiction. Reference 30 omits its author (Farnebäck). The MDPI timeline (21 days submitted-to-published, including a 3-day revise-to-accept and 5-day accept-to-publish) suggests minimal peer review. Confidence is high on the textual/numerical contradictions; image-level fraud cannot be confirmed from the report.

Verdict

🟠 Highly suspicious. The paper exhibits an unusual degree of internal arithmetic self-consistency in its confusion-matrix metrics, but it contains a cluster of methodological timeline contradictions, anomalous sensor scheduling, statistical reporting errors, and an improbably fast MDPI editorial turnaround that together indicate probable data handling irregularities and a deficient review process. No direct image-fabrication evidence is documented in this report.

Key findings

🔴 Timeline contradiction involving Typhoon Doksuri and the case study. Section 2.1 declares that GHI data for Sichuan were unavailable due to Typhoon Doksuri tasking, while Section 3.3 uses a CI event at 23:00 UTC on 4 August 2023 (the last day of that declared outage window) and successfully runs the GHI-based algorithm on it. No reconciliation is provided.
🟠 Implausible GHI minute cadence in Table 3. Within each AGRI 15-minute block, GHI observations systematically avoid minutes ending in 5 and 0 (e.g., 5, 10, 20, 25, 35, 40, 50, 55 are absent). This is inconsistent with the stated ~1-minute sensor resolution and standard satellite scanning mechanics.
🟠 Conflicting confidence intervals in Section 3.1 / Table 7. The methodology text states '95% confidence intervals' via bootstrap (n=1000), but Table 7 labels the FAR interval as a '96% CI: 26.3–39.6%'.
🟡 Missing author in Reference 30. Reference 30 ('Polynomial Expansion for Orientation and Motion Estimation'; Linköping University, 2002) omits the author Gunilla Farnebäck, who is cited by name elsewhere in the paper.
🟡 Implausibly fast MDPI handling. Received 10 March 2026; revised 31 March 2026; accepted 3 April 2026; published 8 April 2026. Revise-to-accept: 3 days; accept-to-publish: 5 days.

Evidence highlights

Article History (Page 1, DOI: 10.3390/atmos17040380): Received 10 March 2026; Revised 31 March 2026; Accepted 3 April 2026; Published 8 April 2026.
Section 2.1 (Page 3): Statement that GHI data were unavailable for convective initiation nowcasting in Sichuan during the Doksuri tasking window (28 July–early August 2023).
Section 3.3 (Pages 14–16): Case study timestamp '23:00 UTC on 4 August 2023' with successful GHI-inclusive algorithm output.
Table 3 (Page 5): GHI minute series shows systematic omission of minutes ending in 5 and 0 across all 15-minute AGRI blocks.
Section 3.1 (Page 11) vs. Table 7 (Page 12): Methodology claims 95% CI; FAR row printed as '33% (96% CI: 26.3–39.6%)'.
Reference 30 (Page 18): Author field empty for the Farnebäck 2002 polynomial expansion paper.
Internal arithmetic check (Tables 5 and 7): POD = 138/(138+28) = 83.13% — mathematically consistent with stated counts.

Notes

The high arithmetic consistency of TP/FN/FP and POD/FAR/CSI raises the possibility of curated or reconstructed data rather than naive fabrication; this makes the timeline and sensor-scheduling anomalies more, not less, concerning.
DOI: 10.3390/atmos17040380 — preserved as published.
Recommended follow-ups (from the source report): obtain the FY-4B GHI raw-data download log and NSMC tasking schedule for 4 August 2023 to resolve the timeline conflict; post concerns on PubPeer regarding Table 3 minute omissions and case-study timing; consider a formal letter to the editorial office covering the 95%/96% CI discrepancy and the incomplete Reference 30.
Limitations of this assessment: image-level duplication analysis (e.g., Western blots, microscopy) is not applicable to this satellite-meteorology study, but no independent verification of the satellite figures themselves was conducted. All verdicts remain subject to official institutional investigation.