Augmented KS Test
The augmented Kolmogorov–Smirnov (KS) test [@Bagniewski2021] detects abrupt transitions by sliding windows of varying sizes along the time series and comparing the statistical distributions on either side of each candidate point. A transition is flagged when the two distributions are significantly different, subject to additional criteria on minimum sample size, rate-of-change, and standard deviation ratio, reducing false positives from noise alone.
When to use
Best for: Sharp, amplitude-based transitions where the distribution of values changes abruptly.
Limitation: Detects only distributional shifts. If the climate system changes dynamical regime without a large amplitude change, this method will miss it. For dynamically-sensitive detection, see LERM.
Reference
Bagniewski, W., Ghil, M., & Rousseau, D.-D. (2021). Automatic detection of abrupt transitions in paleoclimate records. Chaos, 31, 113129. https://doi.org/10.1063/5.0062543
Tutorial
See the KS Test Validation notebook for a worked example.