Republished from “Can IoT Predict Earthquakes?” by Patrick Burns (Haystack Technologies), IoT For All. All rights belong to the author.
Lower cost, lower power, and longer range wireless technologies give us an opportunity to take another look at earthquake prediction hypotheses.
My seventh-grade son, Brendan, is basically familiar with what Dad does for a living — helping companies bring DASH7 products to market. He gets the basics of how the technology works and what it can do. He knows, for example, that companies are deploying our technology to track the precise location and condition of animals like cows.
The Hypothesis
One morning, he surprised me with a question: could Dad’s technology be used to predict earthquakes?
He had read a book about a guy named Jim Berkland who predicted the 1989 Loma Prieta earthquake here in California plus two later earthquakes, based on an increase in classified advertisements for lost dogs in the preceding weeks along with other weather and related data. You can get the book and decide for yourself whether you think his methodology is valid.
But beyond his “model” there are centuries of folklore about animals behaving in strange ways prior to an earthquake. My mother-in-law, to provide just one anecdotal example, insists that her two cocker spaniels living in Mountain View, California “went crazy” in the days leading up to Loma Prieta.
So my son decided for his eight-week science project that he would test whether it is possible to predict an earthquake by testing the question: do animals move more or less prior to a seismic event? There are lots of “little” or medium-sized earthquakes around the Bay Area every day and perhaps he’d discover something.
Regardless, it was a pretty ambitious goal given the timeline and his resources, but his science teacher OK’d the project and to my wife and me, the line of inquiry was novel enough for an 11-year old that it didn’t seem right to discourage him.
After a bit of emailing around to nearby cattle ranches to find a rancher willing to participate, we were introduced to the folks at the TomKat Ranch in Pescadero, California who specialize in educational/scientific research in agro/livestock topics. (NB: these folks are doing some groundbreaking research in agro and sold me some amazing grass fed bone-in-ribeye steak — highly recommend).
We decided not to use Dad’s technology in the interests of minimizing the perception of too much “help” from the parents — something that plagues science fairs everywhere, I believe. Instead, after some brainstorming, he decided to attach (cheap) FitBit Zips to the ears of a few of goats and cows for 30 days in order to measure changes in movement.
The results: according to USGS, there was no significant earthquake activity in the Santa Cruz/Half Moon Bay/San Mateo area during his month of testing so … there was no correlation of a change of movement with a seismic event. Only a decrease in movement on some days due to the heavy rains (El Nino) we received last winter in California. Also: goats move around quite a bit, and cows … not so much.
However, the reactions from the ranchers, science fair judges, and others who learned about the project can be summed up with one word: fascinating.
In California, those who have lived through a major earthquake (I thankfully have not, since 1993, been part of one) seem to react more viscerally to the topic of earthquake prediction. Most of the adults Brendan spoke to were not aware of Berkland’s research and I’ll predict that if someone were to sponsor a ballot initiative to fund large scale testing of this hypothesis, it would pass easily.
In fairness, this spring the U.S. Congress approved $10 million in funding to build an earthquake “early warning” system (not a prediction system) for California that would provide … drum roll please … a one-minute warning of an earthquake. For most of us, such a warning is close to meaningless.
An Opportunity For A Low Cost Earthquake Prediction System?
Previous research efforts to correlate animal behavior with earthquakes appear to be limited and/or very outdated. One study was based exclusively on interviews with farmers after the fact. Until recently, high data acquisition and device costs would have prevented the kind of long term, large scale longitudinal research that we might expect against something as valuable as predicting earthquakes.
Even in my son’s research, using FitBit Zips to measure movement requires manually acquiring the data via Bluetooth every, say, 30 days, which is time-consuming (you have to get within a few feet of each animal, which is tricky and messy) and expensive.
Lower cost, lower power, and longer range wireless technologies give researchers an opportunity to take another look at these earthquake prediction hypotheses. LPWAN technologies being deployed for the (business) purposes of animal tracking, using time stamped location and environmental sensor data, can serve a dual purpose of supplying researchers with valuable data on animal movement, temperature, and location.
A diverse range of domestic animals — dogs, cats, sheep, goats — are now more easily tracked, but non-domestic animals like deer or squirrels (my company, Haystack, has received a request to build a squirrel tracker — not joking) are also potential research subjects.
Getting farmers to give up their data might be easy, or not so easy. For researchers interested in studying earthquakes that want to cut to the chase and tag their own animals or animals they can easily access, the cost of executing a LPWAN-based sensing and data acquisition effort is dropping: tags in low volumes can be had for < $50 and gateways are only nominally more expensive.
Networking stacks offer opportunities to query an endpoint’s GPS location in real-time, which is of great interest to ranchers, but also offer the ability to structure queries in real-time to endpoints, which optimizes endpoint battery life and creates new opportunities for researchers to analyze the behavior of individual animals in real-time.
Whether the centuries of earthquake prediction folklore are just hindsight bias (I’m not accusing my mother-in-law of this, honestly) or a legit indication that animals can sense a coming earthquake, as IoT permeates the agriculture sector, it seems likely that someone will be able to re-test these hypotheses in more meaningful ways and perhaps without much incremental investment.
Can animals feel an earthquake before it happens? People have believed this for centuries. Low-cost, low-power wireless sensors now make it possible to test the idea with real data instead of stories.
An experiment from California
In his article “Can IoT Predict Earthquakes?” (IoT For All), Patrick Burns of Haystack Technologies describes a school science project by his son. The boy attached simple fitness trackers to goats and cows on a ranch near San Francisco for 30 days and looked for changes in movement before seismic events. There were no notable earthquakes in the area that month, so the test showed no link. It did show that animals move less on rainy days, and that goats move much more than cows.
The main lesson is about cost. With ordinary trackers the data had to be collected by hand, close to each animal. LPWAN tags and gateways remove this step: they send location and sensor data over kilometres, run for years on a battery and cost little.
Why this matters for Georgia
Georgia is a seismically active country. The 1991 Racha earthquake and the 2002 Tbilisi earthquake are part of living memory. Many villages in the mountains have no fast internet, but LoRaWAN and other LPWAN networks can cover them with one gateway on a hill.
IoT Lab already pilots early warning systems for natural hazards in Georgia, including seismic monitoring in Sameba village (Samtskhe-Javakheti) and landslide sensors in Adjara. The same networks can carry data from livestock tags that farmers use anyway. This gives researchers long, cheap and continuous records of animal behaviour, which is exactly what was missing in older studies.
Prediction or early warning?
Science cannot predict earthquakes today, and animal behaviour is still a hypothesis. Early warning is different: sensors detect the first waves and give people seconds to act. IoT helps with both: it makes warning networks affordable, and it makes it possible to test the old hypotheses honestly, with data.
