Weather forecasters may not have the best reputation for accuracy, but with today’s computational modeling, it’s possible to make pretty reliable weather predictions up to 48 hours in advance. Researchers at MIT, however, believe that autonomous aircraft running smart storm-chasing algorithms could get that figure up to four days. Better weather forecasting could help farmers and transportation authorities with planning and even save lives by providing earlier warnings about storms and severe weather, says Jonathan How, principal investigator at MIT’s Department of Aeronautics and Astronautics.
Long-term predictions don’t necessarily go wrong because of forecasting models, but rather because initial conditions were inaccurately measured, says Martin Ralph, a research meteorologist at the National Oceanic and Atmospheric Administration’s earth systems laboratory, in Boulder, CO. Such inaccuracies come from gaps in the data, he says.
Ground-based sensors are already used to record temperature, wind speed, humidity, air density, and rainfall, but they gauge conditions only at ground level, says How. At sea, where many severe weather fronts originate, the coverage is much sparser. Satellite observations help build up a picture, but satellites are blind to a number of useful types of data, such as low-altitude wind speed and atmospheric boundary conditions, says Ralph.
To get the most accurate readings, you really want to get your sensors into the weather itself, says How. In theory, weather balloons can do this, but only if they happen to be in the right place at the right time. So weather services currently attempt to track down weather systems using piloted planes that fly prescribed routes, taking measurements along the way. The logistics of deploying such planes is so complicated, however, that it’s difficult to change their routes in response to changing weather conditions.
Consequently, says How, there has been a lot of interest in using unmanned aerial vehicles, or UAVs, instead. The idea is that there would be a constant number of UAVs in the air, continuously working together to position themselves in what would collectively be the most useful locations.
The problem, says How, is that calculating the most useful locations is an enormously complex task. It involves analyzing more than a million data states from hundreds of thousands of sensor locations, and using this data to predict the weather conditions six to eight hours from now. But that’s exactly the challenge that the MIT researchers tackled.
When designing an embedded system choosing which tools to use often comes down to building a custom solution or buying off-the-shelf tools.