The superfog in New Orleans, the debilitating atmospheric rivers in California, crippling dust storms in Illinois, flooding, lightning fatalities, rapidly intensifying hurricanes. What if we could prepare ahead of time for all of them? What if we knew that each was coming and could evacuate and prepare days in advance?
Artificial intelligence is all around us – social media, banking, facial recognition, digital assistants – but that’s not all. Scientists right here in North Alabama are using artificial intelligence to make destructive weather a little less destructive.
I sat down with some of the leading research and data scientists in the country to learn about their contributions to the machine learning field and how advances being made there could change weather forecasting and emergency preparedness as we know it.
Dr. Manil Maskey and Dr. Christopher Schultz work at NASA’s Marshall Space Flight Center in the Earth Sciences division. They are leading the charge, not only on research and development of new artificial intelligence tools, but also on the execution of them in real-world events – on research, development, and execution of AI tools.
Maskey explained to me that many still think of artificial intelligence as a stranger in the sky, so to speak. We know it’s there, we know it’s working, but we don’t know how or why. Maskey says that isn’t the case in the meteorological field at all. He and his team are using centuries of climatology data to change the scientific method as we know it.
Scientists usually create a hypothesis and use large scale and complex equations and, using a series of tests, each with different controlled variables, they determine whether or not their theory was correct.
To forecast, meteorologists analyze the jet stream, upper level wind movements, atmospheric moisture and energy, high and low pressure systems, and more. To do so, a chain of lengthy equations must be used and solved over and over again. They’re called non-linear partial differential equations. They are essentially unsolvable, and they take into account assumptions such as friction, oceanic heat, and more.
Maskey and his team are creating something called advanced neural networks and foundation models that intake data, and instead of programming it into an algorithm, it stores it and learns from itself based on impact and timing accuracy.
Schultz and his team are putting the neural networks to work.
Everything we know about lightning safety is reactive. Thunder roars, go indoors. Moving indoors takes time. Schultz’ advances and his team are currently working on software that adds physics to their predictive models and would allow a warning time of 30 minutes prior to lightning striking in an area. This would allow event organizers and fans in the stands to be alerted long before the threat became imminent, giving them time to take proactive safety actions.
Maskey’s neural networks allow disaster relief teams to see real-time data showing where the effects of floods and fires are most potent.
This allows for disaster relief agencies to have a plan of action long before boots are ever on the ground.
But artificial intelligence isn’t only creating an impact on a global scale. It’s processing information to make a difference right here in Alabama.
And this is just the beginning.
Schultz and Maskey tell me one day all forecasts will be made with help from machine learning. Whether it’s long range temperature forecasting, predicting tornado wind strength, or the likelihood of severe weather in any given system, Schultz tells me new neural networks will be able to intake infinitely more data, compute it faster, and spit out results ready for analysis from a local meteorologist, in a fraction of the time.
It’s not designed to take away human forecasters and meteorologists, but rather leaving those scientific minds free, with more time devoted to making the critical decisions, rather than crunching the numbers.
Now just imagine a world like that.
