A volcanic eruption is one of the most challenging tasks to predict in nature. Of over 1,500 active volcanoes around the world, 50 erupt each year. This upsurge leads to numerous deaths, afflicting tourists and residents. That is why researchers are now using various machine learning techniques to predict eruptions and save lives.
Absence of perfect methodologies has proven deadly for human lives near the active volcanoes. This month, six tourists were killed and around 30 got injured due to sudden blaze in New Zealand. Unlike other active volcano sites where there is an absence of on-site observatories, New Zealand’s White Island volcano was actively monitored. Still, it could not precisely figure it out with only the traditional monitoring technologies.
New Zealand’s noted research centre GNS Science continuously monitors the activities of the volcano through tremors, seismic, and more signals. Although they raised moderate unrest (Volcanic Alert Level to 2), they were not sure of any immediate eruption. Consequently, tourists were allowed to visit the location, but to their dismay, they witnessed an explosion.
Why It Is Difficult To Predict
Many underlying factors lead to failures in predicting the volcano eruption with the desired accuracy. For one, the dearth of data due to the absence of observatory equipment makes it difficult for analysis. To be more specific, let us understand the activity of the White Island volcano from New Zealand.
The crater floor is very near to sea level, so the seawater infiltrates cracks and fissures of the volcano. This heated water then gets clogged and later it is released through the explosion. However, the lack of transparency within the active volcanos makes it strenuous to determine the time of the eruption.
How Researchers At University Of Leeds Solved The Problem
Today, adequate data is crucial for analysing it through machine learning models and predicting precisely.
However, due to a lack of ground-level data, scientists devised and executed a methodology that uses satellites data used satellites to keep track and created a method based on information received from satellites. This helped them in detecting warning signs, which was used for further processing using machine learning techniques.
Scientist utilised images from interferometric synthetic aperture radar (InSAR), which can determine centimetre-scale deformations of Earth’s surface. Images from these satellites were used to pinpoint the movement in the surface of the ground, thereby, assisting scientists in understanding the expansion of magma within volcanoes’ plumbing.
However, some volcanoes exhibit long-lasting deformation that poses no immediate threat, resulting in comparing new images with the older ones to identify the severity of eruptions. To carry out this, they deployed machine learning techniques to analyse the deformation over different periods.
The researchers from the University of Leeds published their research paper on ‘Journal of Geophysical Research: Solid Earth’ describing the methodologies of ‘Using Machine Learning to Automatically Detect Volcanic Unrest in a Time Series of Interferograms’.
This helped them in creating an alert system to notify if ML model identifies any unrest or unusual behaviour. Matthew Gaddes, Andy Hooper, and Marco Bagnardi trained their model using Europian satellites’ imagery data collected over a year before the 2018 eruption of Galapagos Island.
The machine learning model was tested with the genuine data from Sierra Negra, Galapagos Islands, and the model was able to predict ground inflation that started a year ago before the eruption, thereby, differentiating routine and catastrophic deformation.
How Big Is Its Impact
Approximately 800 million people stay near the active blast zone, therefore, makes it a crucial development for saving lives. There have been numerous efforts to predict with seismographs which provide earthquake data to understand the nature of eruptions, but they were ineffectual. However, the new technique will assist in effectively forecasting the hazard while eliminating the need for on-site equipment.
Outlook
This is a great example of how machine learning technique proved helpful in mitigating one of the longest-standing predicaments in nature. However, the use cases have to be scaled up by extracting data from different satellites for covering every active volcano and enable life-saving notifications before the hazard takes place though machine learning techniques.
