Similarly to weather on Earth, there is also weather in space. Every now and then the Sun has intense behaviours that can take the shape of coronal mass ejections, solar wind, or geomagnetic storms and these events are not something to take lightly. For instance, when we talk about coronal mass ejections we mean great explosions of plasma and magnetic fields which can go as far as to impact the Earth. Nonetheless, although the impacts cover a range of sectors, today we will be focusing on the impact on spacecraft.
For the past years, there have been space weather related accidents that affected satellites. Why? Because events such as geomagnetic storms increase the air resistance (drag) of the objects orbiting the Earth. If unanticipated, this can change the expected location of the spacecraft and, in extreme cases, lead to the re-entering of the object and to the failure of the entire mission.
Neuraspace developed a Machine Learning (ML) model that is able to predict the thermospheric density. By having a better estimate of this value, accidents such as the ones above, could possibly be avoided. Moreover, considering that we are at the peak of the current solar cycle, with more frequent and more intense space weather events, being able to know in advance what to expect from the Sun is of even greater importance.
You may be wondering why ML is a good fit for this problem. Indeed, there are empirical models that estimate variables such as the thermospheric density. However, these models tend to oversimplify the problem, and the predictions’ accuracy is far from what would be desired. Diversely, by using ML, we are able to capture the nonlinear patterns of this phenomenon. Indeed, and as can be seen from the figure below, by leveraging ML we are able to surpass the current state-of-the-art empirical models and achieve outstanding performance, even for periods when the sun is more active.
Figure 1: Results in storm conditions from the Neuraspace beta version model.
Another differentiating factor from the empirical models is the blue bounds surrounding the Neuraspace model’s predictions. These are confidence intervals and offer a range of values within which we can reasonably expect the true outcome to fall. They reflect the uncertainty inherent in any prediction or forecast.
By understanding the width and values of these intervals, we gain a clearer understanding of the potential variability in future values for the atmospheric density, something the empirical models do not provide.
Figure 2: Neuraspace beta version model predictions for a 24 hour period at 400 km altitude, covering various latitude and longitude values.
Even with a very promising prototype, Neuraspace continues to work on this problem. Stay tuned!
---
Image credits: © Neuraspace. All rights reserved.
