In college, Paul Roebber reveled in the interdisciplinary aspects of meteorology. This was a sign to come, as Roebber, now a professor at the University of Wisconsin, Milwaukee, would go on to apply biological aspects in his research as he became one of the foremost experts in meteorology forecasting.
Ten years ago, Roebber designed weather forecast simulations that were organized like networks of neurons in the brain. The computer programs formed a system of interconnected processing units that could be activated or deactivated. This “artificial neural network” tool proved especially proficient at predicting scenarios with large data gaps and reams of variables. It significantly advanced snowfall prediction efforts—so much so that the artificial neural network is now used by the National Weather Service.
“For me, creativity comes from being open to broad interests,” said Roebber in a release from the University of Wisconsin, Milwaukee.
Recently, that broad interest extended to Charles Darwin’s evolution theory based on the finches of the Galapagos Islands—spurring Roebber’s next big weather innovation.
Metrology meets biology
Currently, weather forecasters use “ensemble” modeling, which predicts the weather based on the average of many weather models combined. But, ensemble modeling isn’t always accurate as each model is so similar, they end up agreeing with each other, rather than the actual weather. Essentially, more data diversity is needed to distinguish relevant variables from irrelevant ones. However, it’s expensive to obtain and add new data.
The importance of a weather forecast goes beyond you bringing an umbrella to work, or planning to host a party outdoors. In fact, an estimated 40 percent of the U.S. economy is somehow dependent on weather prediction. Even a small improvement in the accuracy of forecasts could save millions of dollars annually for the industries that are affected most—notably agribusiness and construction.
So, if the key to improving ensemble modeling is data diversity—how do you do it without first collecting new data?
Roebber found the answer in nature.
In 1835, Darwin observed what came to be known as natural selection in a population of finches inhabiting the Galapagos Islands. The birds divided into smaller groups, each residing in different locations around the islands. Over time, they adapted to their specific habitat, making each group distinct from the others—and all different from the original finches.
Applying this to weather prediction models, Roebber devised a mathematical method in which one computer program sorts 10,000 other ones, improving itself over time using strategies such as heredity, mutation and—of course—natural selection. The professor began by subdividing existing variables into conditional scenarios: the value of a variable would be set one way under one condition, but be set differently under another condition.
Then, his computer program picks…