On the robustness of Mann-Kendall tests used to forecast critical transitions
On the robustness of Mann-Kendall tests used to forecast critical transitions
Authors: Tristan Gamot, Nils Thibeau—Sutre, Tom J. M. Van Dooren Date: 2026-04-16 Paper ID: openalex:2604.15230
Summary
This paper investigates the statistical validity of the Mann-Kendall trend test when applied to the detection of early warning signals (EWS) for critical transitions. By comparing empirical distributions of the Mann-Kendall statistic against the theoretical Gaussian assumptions typically employed, the authors identify systemic mismatches exacerbated by autocorrelation. The study concludes that current practices suffer from inflated type I error rates, rendering standard Mann-Kendall tests unreliable for critical transition forecasting and suggesting the need for more robust alternatives.
Key Contributions
- Demonstrates that the Gaussian assumption underlying Mann-Kendall trend tests fails in the context of critical transition detection due to induced autocorrelation.
- Provides evidence that mismatches between empirical and theoretical Mann-Kendall distributions lead to significantly inflated type I error rates in early warning signal (EWS) detection.
- Recommends against the usage of standard Mann-Kendall tests for forecasting critical transitions and proposes alternative statistical methodologies.
Open Questions & Future Work
Archivist Review
The paper highlights a critical failure in the standard application of Mann-Kendall tests due to induced autocorrelation in early warning signals. I have approved two open questions that directly address the need for either mitigating these artifacts through weighting or re-establishing a valid theoretical foundation for the statistic. I found no concepts that meet the high bar for permanent vault storage, as the main finding is essentially a cautionary note regarding the misuse of an existing statistical method.
Approved Open Questions
- Gaussian Kernel EWS Robustness: High type I error rates in EWS detection are often driven by autocorrelation artifacts; finding effective mitigation strategies is crucial for reliable forecasting.
- Asymptotic Distribution of Mann-Kendall: Valid theoretical distributions are required to eliminate inflated false-positive rates that currently plague critical transition forecasting when using standard trend tests.
Rejected Candidates
- [open_question] Gaussian kernel window robustness (
gaussian-kernel-rolling-window-robustness) - other: Renamed for consistency and clarity.
Links
Metadata & Links
- url
- https://arxiv.org/abs/2604.15230
- paper_id
- 2604.15230
- paper_source
- openalex
- domain
- time-series
- tags
- time-seriesstatistical-analysisforecasting
- architectures
-
- datasets
-
- concept_slugs
-
- dataset_slugs
-
- skill
- TimeSeriesSkill
- created_at
- 2026-04-19T05:44:35Z