Earthquakes & the Moon: Testing the Connection
The 30-day live view showed striking patterns: more quakes while the Moon was sinking in the west (+18%) and setting (+14%), fewer with the Moon underfoot (−14%), more at new moon (+22%), fewer at full moon (−18%). Thirty days is about one lunar cycle, a few thousand quakes and one busy swarm. This page asks whether those patterns are real, or what you get from any short, clumpy record.
The test: download ten years of GeoNet earthquakes, work out where the Moon was for every one, divide by how long the Moon spends in each position, and ask how much of any pattern chance alone could produce when quakes arrive in clusters. The five 30-day patterns were written down in advance as the only tests that count; everything else on this page is labelled exploratory.
Loading the analysis…
Ten years of GeoNet earthquakes
Every event in the GeoNet catalogue from 1 January 2016 to 23 September 2026, downloaded in checkpointed windows and checked event by event against a second download. The 30-day dashboard window (24 September to 23 October 2026) is kept out of the replication so the original patterns are tested on data they were not found in.
Earthquakes by lunar position
Each bar is a rate ratio: quakes seen in that Moon position divided by the number expected from how many hours the Moon spent there. 1.00 means no effect. The thin line is the 95% interval from resampling two-week blocks (so one swarm cannot count as hundreds of independent events). The shaded band is the range a meaningless Moon would produce on the same quakes.
Hover or focus a bar for the numbers.
Show the numbers
Earthquakes by lunar phase
The same test for the monthly cycle: new moon, quarters and full moon, split into eight equal slices of the 29.5-day cycle. The filters above apply here too. Spring tides come around new and full moon, so a tidal effect would show as the same sign at both.
Hover or focus a bar for the numbers.
Show the numbers
Tidal potential level (fifths of the time)
Tide rising or falling
Tidal potential is computed at each event’s own latitude and longitude from the Moon’s and Sun’s positions, then converted to ground uplift and areal strain with IERS 2010 Love numbers. These are orientation-free measures of how strongly the Earth is being deformed. They are not stress on any fault: that depends on the fault’s orientation and slip direction, which an ordinary catalogue does not contain.
Hover or focus a bar for the numbers.
Does it depend on how big or how deep?
If the Moon nudged faults that were already close to failing, the effect might show more clearly in some groups (shallow thrust events, or large quakes, as some published work suggests). Each row is a separate test; the p-values are the cluster-aware ones from shifting the Moon timeline. With this many tests, a few small p-values are expected by chance alone.
By minimum magnitude
By depth
Regions and time periods
New Zealand sits on very different tectonic settings. The regions here are rough latitude-longitude boxes, not an official regionalisation, and their edges are arbitrary. Periods check that no pattern appears only in the Kaikōura-dominated early record.
By region
By period
How confident can we be?
This is the section that decides it. Five tests were fixed in advance, copied from the 30-day patterns. Each is run on the ten-year hold-out, with and without declustering, and corrected for running five tests together.
Clustering is the problem: one large quake makes hundreds of aftershocks that all share nearly the same Moon position. Naive tests count each one as independent evidence.
What the literature says
Tidal triggering of earthquakes is a real research topic, and one that stays disputed. The credible work looks at stress on a particular fault in a particular setting, not at the Moon’s position over a country. What is established, and what is not:
Links go to the publishers or Crossref; titles, authors, journals and years were checked against Crossref. Summaries are paraphrased, and I have not reproduced figures from any paper.