It has been about two years since machine learning models got good enough at weather prediction to outperform traditional numerical weather models. Unfortunately, deep learning has an upper bound for accuracy over long timescales: real world data points. We are solving this problem.
Spyro Labs has used proprietary stratospheric balloon data to add a correcting layer on top of leading ML wind models to produce the world's most accurate model for predicting stratospheric winds.
This is a big deal; Integration of machine learning and physics based modeling now produces better results than either technique can achieve independently.
Applied within the stratosphere alone, this breakthrough has major implications for two industry categories:
1. Stratospheric Operators, who rely directly on high-fidelity wind models:
- Balloons and High-Altitude Platforms (HAPs)
- Supersonic and stratospheric flight systems
2. Forecast-Dependent Industries, which benefit from improved long-term weather prediction:
- Shipping
- Construction
- Insurance
- Commodities trading
- And more!
So what's next?
First, we are launching swarms of balloons carrying custom hardware to the stratosphere in both hemispheres, providing the data for our correcting layer to not merely outperform all other modeling techniques, but blow them away.
Once we've done that, we will move on to other categories of data and build a complete correction layer for the world's best holistic weather model.