Image
Is Winter Coming? Meteorologist Explains Why We Don’t Trust Weather Data
University of Virginia M.S. in Data Science alumnus Rohan Shroff is a meteorology software developer at World Climate Service whose primary focus lies in developing new tools to help make the lives of other meteorologists easier through improved forecast efficiency and new insights in a variety of markets. We spoke with Shroff about the incoming inclement weather, what makes weather data so unpredictable, and how AI is changing the field of meteorology.
Q: People joke whether we will get one inch or 22 inches of snow. What makes weather data so unpredictable?
Shroff: The biggest limitation is the accuracy and resolution of the input data that goes into a weather model. To create a forecast, a weather model first needs to ingest hundreds of millions of observations from around the globe — things like temperature, humidity, pressure, etc. Because the atmosphere is chaotic, even tiny errors at the start, often called the butterfly effect, can grow exponentially, leading to large forecast differences after just five to seven days!
Q: What kinds of indicators are meteorologists looking for in weather data that determines the severity of a snowpocalypse?
Shroff: To get a really good snowstorm, you need two key ingredients: a strong and resilient cold air supply and a steady source of moisture. When these two parameters line up, that’s when we start to get pretty excited. Unfortunately, with this storm, while the cold air will be very impressive at the surface (< 25 degrees Fahrenheit), warm air rushing in a few thousand feet above our heads will change the snow to a mess of sleet and freezing rain by Sunday afternoon.
Q: Compared to technology available during the 2010 Snowmageddon, how does artificial intelligence help make more accurate weather predictions?
Shroff: Artificial intelligence has already become a powerful new tool in weather forecasting. AI-based weather models, which take a fraction of the time to run compared to physics-based weather models, have proven to be just as good if not better in many complex weather situations. For example, with this upcoming winter storm, the American and European AI models were among the first to identify a northward shift in the storm track and associated lower snowfall totals for Virginia, compared to our traditional weather models.
Q: How do you see meteorology evolving as fields like data science and AI also evolve?
Shroff: The biggest thing that I see is the continued rise and adoption of AI tools to help us create more accurate weather forecasts. From predicting rapid intensification of hurricanes to major cold air outbreaks, AI-based forecasting systems have already proven very valuable and will continue to do so.
