Project GAINS: using HPC to improve stratospheric wind estimation
Stratospheric balloon operations take place in a dynamic and constantly evolving environment, where atmospheric conditions continuously shape mission planning and execution. At around 20 kilometres above the Earth, wind patterns and atmospheric conditions influence the floating and positioning of stratospheric platforms, affecting their operational capabilities and mission effectiveness.
Accurately understanding those winds is therefore a fundamental part of stratospheric operations. Yet it is also one of the most complex.
This is the challenge at the centre of Project GAINS (Guidance & Atmospheric Inference for Station-Keeping). Developed within FFplus (Fortissimo Plus), a project funded by the European High Performance Computing Joint Undertaking (EuroHPC), with support from the Digital Europe Programme, GAINS has recently been granted access to Leonardo, the Italian supercomputer hosted by CINECA and one of the most powerful high-performance computing (HPC) systems in Europe. An achievement born of the collaboration between BEES and Involve Space, respectively the project’s independent software vendor and end user.
From forecasting a trajectory to understanding the atmosphere
The goal of Project GAINS is to build a statistical model able to use real-time onboard observations of a stratospheric balloon to continuously improve our estimate of the wind field in the stratosphere. Simple, right? Let me explain it better.
Before launching a stratospheric balloon, meteorological forecast data are typically used to estimate the atmospheric wind field and predict the balloon trajectory. This is crucial for reaching specific target areas or performing station-keeping.
However, the stratosphere lies far above the altitudes where most atmospheric observations are collected. As a result, wind conditions and air currents there are less directly measured than at lower altitudes, and forecasts can carry a higher degree of uncertainty. And this is where the balloon itself becomes extremely valuable.
During flight, the balloon becomes a privileged observer of the stratosphere: through its motion and onboard measurements, it provides crucial information about the local wind field around its position. This is exactly where GAINS comes in.
By combining these onboard observations with atmospheric forecasts and background error statistics, we are developing a Hybrid Local Ensemble Transform Kalman Filter (HLETKF). This is a forecasting model capable of updating not only the wind estimate at the balloon’s location, but also across the surrounding spatial and temporal region.
So, why do we need HPC?
For a balloon measurement to provide useful information about the wind at a nearby location, conditions at the two locations must show similar statistical patterns. To estimate these correlations robustly, we analyse around 16 years of atmospheric data, building a climatological sample that captures as much variability as possible.
From a statistical perspective, the historical dataset can be divided into eight distinct atmospheric scenarios, defined by the combination of four seasons and two phases of the Quasi-Biennial Oscillation (QBO), a large-scale atmospheric phenomenon in the equatorial stratosphere in which high-altitude winds periodically reverse between easterly and westerly directions. Since each scenario can be analysed independently, computations can be distributed across multiple HPC nodes and processed in parallel.
Using this strategy, we have already built a statistical model of climate forecast uncertainty, leveraging the eight atmospheric scenarios to efficiently distribute the workload across the computing cluster. The next step is to analyse how forecast uncertainties are organised, identifying the dominant patterns that influence wind predictions across different regions of the atmosphere.
This information is crucial for the overall HLETKF forecasting model, because it determines how a local observation collected by the balloon can influence and improve the wind estimate beyond the exact point where the measurement was taken.
And what comes next?
The final goal is to build a forecasting system that blends long-term information from past atmospheric conditions with real-time observations available before launch, providing more accurate wind predictions.
During flight, the balloon will measure the local wind and transmit the observation to the ground, where the filter will update the surrounding wind field in real time.
By combining information from both the past and the current atmospheric state, the balloon can “surf” the stratospheric flow more effectively and improve the probability of successful station-keeping.
My Personal Experience
Personally, I’m really enjoying working on this project because it has pushed me into areas I have never explored before. Before GAINS, I had never worked with HPC systems or atmospheric wind models, and there was definitely a steep learning curve. But that’s also what makes it exciting. Seeing these ideas come together and run on a computing cluster still feels pretty amazing.
I’m looking forward to the next stages of the research and to discovering where this work can take us. There is still much to explore, so stay tuned for what comes next!
