A computer model of the Leonid dust trail forecasts relatively quiet activity in 2031 and 2032, a peak modeled rate of 1,409 in 2033, two strong 2034 encounters at 1,401 and 1,217, and another active year at 651 in 2035. The paper labels these class-B projections and places them in a 0.29 to 3.5 band. The figures are forecasts, not a confirmation of what future observations will record.
The sharpest uncertainty comes in 2034. The early peak is assigned to the 1932 trail and placed at 03:30 UT; the later peak is assigned to 1733 and placed at 23:45 UT. The relative ranking can reverse: the 1932 rate is 2.6 × 10^4 without the empirical density factor and 1.4 × 10^3 with it, while an unmeasurable population index gives a reported range of 4 × 10^2 to 1.4 × 10^3.
A model built to catch timing errors
The study is a modeling exercise aimed at two linked questions: how well a Leonid dust-trail model assigns past encounters, and how much its results shift when the parent body's trajectory is handled differently. The parent comet was integrated with the IAS15 scheme in REBOUND, alongside the eight planets and a Marsden-form non-gravitational acceleration. Simulated grain sizes and ejection times were sampled from their distributions, with each grain standing in for one visual meteoroid with unit statistical weight.
To turn the simulated dust into an activity profile, the calculation stepped through solar longitude in 15-minute intervals and used an adaptive nearest-neighbour density kernel with four neighbours. It produced five epoch-matched forecast datasets, one for each year from 2031 through 2035. Each contained 450,000 grains across 18 trails, covering ejection returns from 1433 to 1998, with 25,000 grains assigned to each trail.
That construction changed the historical attribution in the 2001 and 2002 hindcasts. After the ejection sites were anchored, the 1767 trail was dominant in both years. Its modeled peaks came one minute after observation in 2001 and 40 minutes after observation in 2002, with radial miss distances of 2.9 × 10−4 au and 1.0 × 10−4 au. Before anchoring, the miss distance was 3.5 × 10−3 au. A residual old-trail propagation error remained.
The weak spots are in the details
Another test showed why epoch matching matters. Across 0.12 revolutions, the real trail nodes moved a median 1.15 × 10−2 au, or 23 radial-kernel widths. Frozen-element propagation left the reported node 6.4 × 10−3 au, or 13 radial-kernel widths, in error. The amount of omitted motion varied among trails, which is why the forecast calculation used epoch-matched datasets.
To set the overall amplitude, the author jointly calibrated three anchored storm hindcasts with one constant, Zstorm = 8.6 × 10^3. The individual model-to-observed ratios were 0.89, 1.25 and 0.90, with a geometric mean of 1.00. The calibration constant remains uncertain because different realizations of the simulation vary.
The validation also exposed a false-positive amplitude problem that can disappear when a model is judged only against nominated trails. An audit of all predicted encounters found that the 1932 trail was over-predicted eighteenfold. The model still reproduced the timing and observational existence of the relevant encounters, but the result shows why a complete scorecard matters.
A density factor called fM followed the expected inverse-return-count pattern, with a median ratio of 0.95 within three radial-kernel widths of a trail's core, but it was unconstrained beyond five widths. The two 2034 sections sit at 1.1 and 2.8 widths, respectively, so both are near the core-to-wing boundary rather than direct central hits. Near-miss rate predictions are therefore less secure than near-direct-hit predictions.
The forecast is waiting for a real-world test
The 2034 encounter is the clearest test of the model's remaining assumptions. The early 1932 section and late 1733 section are close enough in the forecast to look comparable, but the choice of empirical density correction and the treatment of the unmeasurable population index can change which one leads. The paper presents the encounter as a test of whether a correction inferred from a far-tail section transfers to a core crossing.
The work is an arXiv preprint, version 1, dated 26 August 2026. Its epoch-matched simulation datasets are archived at Zenodo. The paper reports no specific grant funding, and the author declares no known competing financial interests or personal relationships that could have influenced the work.
Paper data and sources
Original title: A calibrated dust-trail model of the Leonid meteoroid stream and forecasts of the 2031-2035 encounters
Authors: Shinsuke Abe
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text