NASA and Amazon are collaborating on developing an application that can predict solar superstorms.

2020-01-20

The sun gets angry every now and then, and if the Earth gets in its way, bad things can happen. That's why NASA and Amazon are collaborating to use new machine learning applications to better predict the arrival of "superstorms" from the sun.We have seen these superstorms cause destruction in the past. In 1859, the Carrington Event, as it is known, disrupted telegraph communications. In the winter of 1989, thousands of residents in Quebec, Canada were plunged into darkness due to power outages. These worst-case events are rare, but they occur on average once every 50 years.


These events are caused by solar eruptions. The sun periodically emits flares (which can be seen through large telescopes), which are often accompanied by bursts of invisible radiation particles called coronal mass ejections. Normally, the Earth's magnetic field can shield against the radiation. However, occasionally there are so many particles that they generate massive solar storms, which can destroy satellites, power lines, and other critical infrastructure for humans.


Given how rare superstorms are, there are few historical examples available to train algorithms. This makes conventional machine learning methods, such as supervised learning, severely inadequate for predicting superstorms. Additionally, dozens of past and present satellites collect space weather information from different crucial vantage points around the Earth, resulting in a vast amount of data that is difficult to search for relevance using conventional means.


Amazon's AWS Professional Services and Machine Learning Solutions Lab have a different approach. They use both unsupervised learning and anomaly prediction to better forecast conditions related to superstorms. AWS can examine up to 1,000 datasets simultaneously, based on anomaly grading generated by NASA, to identify unique patterns associated with superstorms.


In the near future, NASA and Amazon plan to provide a data "lake" (or repository containing a large amount of raw data) for researchers to have hands-on access to the numbers. The plan is to make forecasts more robust by looking at anomalies and simulating current superstorms as well as processes related to extreme historical storms like the Carrington Event.


"There is a lot of data, and factors like time delays add complexity. With Amazon, we can get every piece of data from a superstorm and use the anomalies we detect to improve models that can effectively predict and classify superstorms," said Janet Kozyra, Chief Scientist for Solar Physics at NASA.

RECENT AI NEWS

RECENT AI TOOLS