The traffic that is collapsing the city, the works in the street, the garbage truck, the disputes between neighbors, the terraces of bars: cities are, whether we like it or not, synonymous with noise. We are used to living with noise pollution as if it were background noise, but the truth is that it can cause serious health problems. And they are not anecdotal: these already affect 60 million adults, according to the World Health Organization (WHO). Moreover, in addition to being able to cause cardiovascular disease or mental health impairments, city noise can also interfere with the detection of earthquakes.
Try to Listen earthquakes among urban noise, Stanford University – in collaboration with the Chinese Academy of Sciences – has developed an artificial intelligence (AI)-based filtering system that remove noise, pushing it out of data from seismic sensors in metropolitan areas. It is, at least, explained in an article published in the magazine Science: In order to warn the population of the risk of earthquakes, experts have for years placed seismometers in areas prone to tremors, especially in urban areas where they cause more damage. For earthquake specialists, it has always been a big problem to separate seismic data caused by natural ground movements from those related to city life, that is, human activity in cities – like vehicle and train traffic – produce a lot of similar noise.
Faced with this situation, researchers from Stanford and China have developed this application which they claim is able to learn on its own – the general process called machine learning– how to filter which seismic data are natural and which are artificial. Something particularly useful according to Paula Koelemeijer, seismologist at Royal Holloway University in London: “Showing that the algorithm works in a noisy urban environment is very useful. City noise can be a nightmare, as well as very challenging.”
Four times more efficient
The new system is called UrbanDenoiser and it is an application that works with deep learning (in Spanish, “deep learning”), a set of deep machine learning algorithms included in the machine learning. Specifically, this system qualified itself with 80,000 samples of urban seismic noise and 33,751 of natural seismic activity.
Initially, the research group tested it with seismic data recorded in Long Beach, a city in the US state of California, to see how it worked: they found that it managed to improve the level of signals desired – relative to background noise – at around 15 decibels.
It has always been a big problem to separate the seismic data caused by natural ground movements from those related to city life.
Satisfied with the results, the researchers used UrbanDenoiser to analyze data from an earthquake that struck a nearby area in 2014, finding that the app could detect four times more data than unfiltered sensors. For now, according to believe, your tool can be used for surface noise, localized stress concentration, and intermediate blockage seismic monitoring. However, they also clarified that the system requires retraining with region-specific datasets before it can be implemented as a ground motion monitoring system.
This is not the only study that has followed this line of research in earthquake research: an article from Penn State Universityin the United States, also trained deep learning algorithms to accurately predict how changes in measurements might indicate upcoming earthquakes. With this model, it is possible to predict the size and timing of earthquakes from ultrasound data from active sources and from models of deep learning, which allow for more accurate predictions. It is a promising model: as they point out, “the predictions are accurate despite irregular seismic cycles”.