As the recent catastrophe in Houston, TX has illuminated all too emphatically, natural disasters can touch any of us, anywhere, at any time. Houston is a city close to our hearts here at ENO8, and we send our warmest thoughts, prayers and condolences to everyone impacted by this terrible tragedy.
In reflecting on Hurricane Harvey, it occurred to me that one of the most important defenses against renegade weather is forecasting — the less time you have to prepare for something, the more likely it is to severely damage property and result in loss of life. If you know that a hurricane may hit your city’s shore but the winds and rain only warrant boarding up your house vs. requiring a full-scale evacuation, you can avoid mass hysteria, unconscionable congestion and basic shutdown of swathes of the country. In Houston’s run-up to Hurricane Rita a decade ago, the evacuation killed almost as many people as the storm’s landfall did. In cases like that, it’s incredibly valuable to predict the severity of the storm (or lack thereof) accurately to avoid unnecessary disruption of or loss of life.
Consequently, residents are more likely to stay home and “ride it out” the more often their local weather services or national weather services overestimate the severity of landed storms. When forecasters are wrong, it sows distrust in residents. If the experts wrong enough times, citizens stop trusting them altogether, and you end up with more people in harm’s path if a storm ends up being as big (or bigger) than anticipated.
Having an accurate weather prediction is super convenient and useful for planning purposes in our day-to-day lives — no doubt. I love being able to select what to wear knowing exactly what the weather conditions are likely to be that day. But, in times of life and death? Having an accurate forecast goes from a nice-to-have to a must-have. The more accurate the forecast, the more likely cities and states can accurately call for an evacuation vs. riding it out. The more accurate the forecast, the more time communities will have to brace for impact, especially given hurricanes’ nature of shifting course at the last minute. Obviously, natural disaster forecasting is hugely important in times of emergency.
The life-and-death essence of storm-based natural disasters make forecasting a natural candidate for some of the world’s biggest and baddest supercomputers. For years, some of the top performing number crunchers in the world were aimed at weather forecasting. The reason storms require so much computing power to come close to accurate forecasts? The gargantuan number of variables in any given weather equation.
In the case of North American hurricanes, that can be everything from ocean water temperatures at variable depths, shifting currents, high/low pressure system movement, the jet stream, atmospheric moisture, barometric pressure, air temperatures at varying, relevant altitudes… the list goes on and on. And, not only do we have billions upon billions of historical data points for storms, we also have an army of sensors out in the world at any given moment feeding incomprehensible amounts of real-time data back to our supercomputers for analysis.
This situation was and is ripe for an influx of deep learning and artificial intelligence to take modeling and forecasting to the next level. IBM, one of the leaders in machine learning, neural networks and A.I., saw that exact opportunity when it acquired the Weather Company in 2015. By bridging IBM’s Watson with the Weather Channel’s digital and technological assets, IBM wanted to help make weather forecasting that much faster, smarter and more accurate.
Used to be, humans had to build predictive models for weather forecasting that supercomputers then applied to real-time data to make a prediction. The model was only as good as the human designing it, and the machine could only analyze real-time data through the lens of the model it was tasked with applying. With machine learning in the mix, though, computer scientists can simply feed the neural network all the historical data points to both refine the model and apply it in real time. The more historical and sensor data the neural network gets its proverbial hands on, the more accurate the model and the forecast is likely to be. It’s a huge test case for the usefulness of artificial intelligence, and one in which I’m sure we’re all rooting for A.I.’s success.
It goes to show that some technology companies really are trying to solve humanity’s problems through the tactical application of targeted machine learning paired with raw computing power. While we can’t stop storm-based natural disasters, more accurate forecasting can certainly save hundreds if not thousands of lives over the coming years and decades.
That’s an R&D budget well spent.