[maratumba]

Seismologist here. So far, no reliable predictor of main shocks (largest earthquake in a sequence) has been found. People have been looking for them ever since the invention of the modern seismometer networks in the 60s.

Most of the EM based earthquake prediction is based on really loose reasoning. In summary, it goes like this: "earthquakes can generate EM fields through piezo-electric effect, therefore small movements before big earthquakes should generate small EM fields we can measure". But there is almost never no such thing as "small movements before big earthquakes", which is why reliably predicting them has been impossible so far.

Most likely, this will turn out to be an example of confirmation bias. In the unlikely event that it is not, people will be all over this.

[jofer]

To be fair, there are some well documented EM precursors. The most classic is from the 1989 Loma Prieta earthquake: https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/GL01... Recently, some fairly convincing ionospheric-related (i.e. GPS delay) precursors have also been observed before Tohoku and (less convincingly) several other major earthquakes. (e.g. Heki, 2011, Iwata & Umeno 2016, and several other references I forget. Mostly from the same couple research groups.)

We don't have a full mechanism to explain them (or, rather, there are a lot of competing mechanisms that don't fully explain things). More importantly, precursors don't seem to occur consistently as you noted.

However, they're also not worth automatically dismissing. The idea that they're all a simple case of confirmation bias has been extensively discussed, and while it's not currently possible to refute, it's starting to seem less likely. There's certainly been a lot more attention given to possible precursors and mechanisms in the last 5 years than there were before. It's a serious avenue of research right now. Keep an eye out then next time you're at AGU. I guarantee you you'll see at least a few posters around possible precursors and/or precursor mechanisms.

Forecasting is definitely pseudoscience, but EM-related precursors are a fairly hot (and controversial) research topic at the moment.

[bsder]

> Forecasting is definitely pseudoscience, but EM-related precursors are a fairly hot (and controversial) research topic at the moment.

It's not pseudoscience because it can be disproven--which is generally what happens.

The issue that everybody forgets is that predictions have FOUR outcomes, not two.

You have the one everybody remembers: "I predicted X and X happened".

You have the one some people remember: "I didn't predict X and X didn't happen."

You have the one that people rarely remember: "I predicted X and X didn't happen."

You have the one nobody remembers: "I didn't predict X, but X happened."

The problem is that for rare events, the "predict X and not X" and "didn't predict X but X" have to be REALLY low probability for a measure to be useful.

[gpm]

Didn't predict X but X can be high probability and it still be a useful predictor. Reporting even 10% of earthquakes hours in advance would be useful.

Predict X and not X is a huge issue though, as you suggest.