In 2018, a magnitude 7.1 earthquake rocked Southcentral Alaska, leading to injuries and millions of dollars in damage. While major quakes seem to come out of nowhere, researchers at the University of Alaska Fairbanks hope a new research method can help forecast them in the state months before they happen.
Társilo Girona with the UAF Geophysical Institute says he and other researchers trained an algorithm to analyze 30 years of earthquake data in Alaska and California.
Listen:
This interview has been edited for length and clarity.
Társilo Girona: Basically, we told the algorithm to look at… during this time period there was an earthquake or during this time period, there was not an earthquake. And we are teaching the algorithm to recognize any kind of pattern emerging during the different time periods. And then, as I said before, we use those algorithms trained with previous earthquakes to see whether we can detect similar anomalous behavior in other earthquakes that were not used in the training process. In this case, for the Anchorage earthquake and also for the Ridgecrest sequence happening in California.
Wesley Early: Studying two earthquakes seems to be a small data set. How confident are you that your method of forecasting large earthquakes is accurate?
TG: Well, we can never forget that science is a building process. So the most relevant here is that we are detecting some anomalous behavior in the low magnitude seismicity that hasn’t been reported so far, before major earthquakes. So, that’s a very relevant finding. Then the point is, is this something happening before the occurrence of other large magnitude earthquakes and in other regions of the world? So these are the kind of questions that we still need to address in the future. There are some important things to keep in mind here. We decided to use these two earthquakes basically because we have a great record of data. So in California, we have a very good earthquake catalog, thanks to the tons of research that has been done over the last 34 years. So that means that we needed 30 to 40 years of data in order to be able to train an algorithm like this. I’m highlighting this because in other parts of the world, it might not be possible to apply this methodology right now, because we don’t have such a great amount of data in, and that’s the challenge that we have. That’s another reason why we couldn’t work with more earthquakes at this stage. But as I said at the beginning, the key point here is that with these two case studies, we were able to detect some anomalous behavior in the low magnitude seismicity that hasn’t been reported so far, and that’s what we think is an important thing to keep in mind, because this opens new new doors, new perspectives, to potentially use those statistic anomalies to better forecast earthquakes in the future.
WE: So it sounds like you would need to study areas that have well documented earthquake data in order to sort of apply this method.
TG: That’s correct, because with these machine learning algorithms, we need to train the algorithms with previous data from that specific region. Of course, one could train the algorithms with data from California, for example, or from Alaska, but then applying that same algorithm to a different region, that’s not an easy step. So I always recommend to train the algorithms with the earthquakes of the particular region that we want to explore, that we want to analyze. But again, we don’t have very long data sets for any seismic region of the world, so there are some caveats there, and there are definitely a lot of things to do in the future to keep helping us to detect the emergence of those statistical anomalies.
WE: And one thing that you note is that the earthquake forecasting is kind of like a double-edged sword. You can have false alarms that can generate panic and economic issues unnecessarily, while missed predictions can be kind of deadly. How do you hope to balance those two things?
TG: Well, that can lead to a very long debate. Definitely here in this research, in this project, we have been focusing on the science part, understanding the preparatory phase of a large magnitude earthquake is important for many reasons. One of them is because we can use that for forecasting. But the other reason is that we can also use that to better understand how faults behave, and to better understand how the earth basically works. And that has been the main focus of this project, trying to better understand the preparatory phase of a large magnitude earthquake, exploring how the low magnitude seismicity can somehow inform us of the period of unrest of a specific region. But then, of course, when we go from the science to the application of science, then we need to deal with those societal challenges. And definitely that’s something we will need to figure out. There is not an easy answer right now to this, but as I also said, at some point, science is a building process. We need to move forward to better understand how the earth works, but also to make sure that we can move forward in terms of generating algorithms and generating forecasts that can help society.
Wesley Early covers Anchorage life and city politics for Alaska Public Media. Reach him at wearly@alaskapublic.org and follow him on X at @wesley_early. Read more about Wesley here.
