The signal in the record
Earth's rotation appears to show a broad rhythm of roughly seven decades, according to an analysis of long-term rotation records. The study says the signal is consistent with a core-related contribution to low-frequency rotational variability, but its fitted model does not point to an imminent negative leap second.
At the center of the question is length of day, or LOD. The study asks whether recent rotational acceleration is consistent with an approximately 70-year core-related modulation and what that could mean for a possible negative leap second. It addresses that question by comparing an LOD record with a CAM-derived equivalent series.
The common comparison covers 1883 through 2022, the period shared by the annual LOD and CAM-derived series. Because the CAM series ends in 2022, all reported multidecadal fits and direct LOD-CAM comparisons use that common interval.
How the comparison was built
The analysis constructed residual LOD by removing the IERS 2010 zonal-tide term, the positive 2.3-millisecond-per-century secular tidal-braking term and available EAM terms.
To create the second series, the analysis obtained CAM-derived equivalent LOD from the COV-OBS.x2 geomagnetic model through WebGeodyn, using frozen-flux and geostrophic assumptions. It was therefore a model-derived comparison series rather than a second direct LOD record.
Periods from 50 to 100 years were tested at 0.1-year increments with a four-parameter least-squares fit. BIC ranked the candidates, while Lomb-Scargle spectra and continuous wavelet diagnostics provided additional checks.
A pattern, but not a fixed clock
The two series converged on a similar timescale. The fitted period was 69.7 years for residual LOD and 71.8 years for CAM-derived equivalent LOD. Their fitted amplitudes were 2.87 and 1.94 milliseconds, with maxima near 1970.45 and 1967.01, respectively.
Other checks landed in the same range: Lomb-Scargle peaks were 67.8 and 70.5 years, and BIC favored a period near 70 years over comparisons near 60 and 90 years. But only about two complete multidecadal cycles are present. The authors therefore describe the result as a 60-to-80-year modulation with an approximate timescale near 70 years, rather than a uniquely fixed stationary oscillation.
The original LOD record showed a marked trend change near 1972. In the paper's angular-acceleration measure, the value was negative 13.1 x 10^-22 rad s^-2 before 1972 and positive 50.6 x 10^-22 rad s^-2 afterward. The modeled tidal and EAM terms did not reproduce that low-frequency reversal.
Across 1883-2022, zero-lag correlation between the residual LOD and CAM-derived series was 0.918. Using a centered 11-year moving average, the correlation was 0.925. The lagged maximum was 0.932 when CAM led LOD by two years, within a broad one-to-three-year maximum.
Those numbers describe alignment, not predictive power. The analysis examined lags from -20 to +20 years in one-year increments, but strong autocorrelation led the authors to treat the correlations as correspondence rather than predictive-significance tests. The result therefore does not establish a unique causal direction or mechanism.
What it says about leap seconds
To estimate the timekeeping effect of the fitted cycle, the analysis integrated its multidecadal LOD component from 1972 through 2022. It yielded an approximately 23.3-second peak-to-peak contribution to continuous UT1, which represents only one component of observed UT1-UTC variability.
In the fitted model's extrapolation, the next positive maximum is near 2040. The multidecadal component alone does not imply sustained near-term day shortening requiring a negative leap second, so the analysis does not support an imminent negative leap second. That conclusion is conditional on the fitted model and is not an operational UTC forecast.
A separate post-1972 calculation found a residual LOD slope of -0.108 milliseconds per year and a fitted-component slope of -0.099 milliseconds per year. Their ratio was about 92%, but the paper treats that as an empirical fraction of the residual slope, not as a physical decomposition among mechanisms.
Why the result remains conditional
The core-related interpretation remains broad. The similar low-frequency signals are consistent with a core-related contribution, but the analysis does not distinguish among electromagnetic, topographic, gravitational and viscous forms of core-mantle coupling.
The main caveat is that only about two complete cycles are resolved, so the period remains approximate. The CAM series is model-derived under frozen-flux and geostrophic assumptions, and the correlation figures are limited by strong autocorrelation. These constraints make the interpretation descriptive and conditional.
The document is a preprint identified as arXiv:2608.25964v1 and dated 26 August 2026. Its primary IERS, GFZ, COV-OBS.x2 and WebGeodyn data are reported as publicly available; processed series and analysis scripts can be requested from the corresponding author.
The open question is whether the apparent multidecadal structure persists with later observations and under alternative residual constructions. Further work would also be needed to separate the possible core-mantle coupling mechanisms and to test any leap-second implication within an operational UT1 forecasting framework.
Paper data and sources
Original title: An Approximately 70-Year Core-Related Modulation of Earth Rotation and Its Implications for the Leap Second
Authors: Zewen Zhang, Yuanwei Wu, Xishun Li et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text