When COVID-19 Emptied the Skies This Spring, It Likely Worsened Weather Model Predictions

The finding illustrates the value of regular weather observations made by commercial airplanes.
plane flying through a storm
Media credits

FTiare via Shutterstock

Catherine Meyers, Editor

(Inside Science) -- In all the disruption unleashed by COVID-19 pandemic lockdowns this spring, you probably didn’t notice a change in the accuracy of your daily weather forecast. But new research shows plummeting volumes of air traffic this March and April -- and the associated drop in weather observations -- likely worsened short-range predictions from key weather models that contribute to those forecasts. 

Many commercial airlines currently outfit their planes with sensors to measure air temperature, wind speed and direction, and sometimes water vapor. The airlines share the data with weather modelers who use it to update the initial conditions of their models and rerun them, regularly boosting the accuracy for the next few hours of forecasts. The data from commercial aircraft are one of the most important sources of weather observations because planes travel up and down through the atmosphere so frequently, said Eric James, a meteorologist at the National Oceanic and Atmospheric Administration’s Global Systems Lab in Boulder, Colorado. 

Before the pandemic upended life as we know it, James and his colleagues had already begun "data denial" experiments to help quantify this importance. When COVID-19 struck and the number of flights nose-dived, it made data denial not just an experimental "what-if," but an unexpected reality. The team decided to run a few more experiments to approximate what was happening in real life.

They found that removing all aircraft observations degraded the accuracy of air temperature, wind speed and direction, and relative humidity predictions made by the latest version of an advanced North American weather model in both summer and winter. Removing 80% of the observations also degraded the accuracy, though by less than 80% of the effect of total data denial. That’s probably because there is a core backbone of aircraft observations that is responsible for most of the impact, James said.  Removing, say, the first 10% of the observations might have almost no effect, while removing the last 10% could have a much more drastic effect.

An 80% reduction in aircraft observations is approximately what happened this spring -- at the end of April, planes were reporting 75% fewer observations than normal. It’s impossible to know for sure that short-range weather model forecasts worsened during COVID-19 lockdowns because in the real world, there is no alternate reality with a full set of observations for comparison. But it’s fairly safe to assume the predictions got worse, James said. Air traffic is currently reduced by closer to 50%, so weather model predictions could still be experiencing a reduction in accuracy, although probably not as extreme as this spring, he added.  

Weather models like the one the team examined contribute to the weather forecasts that the public often sees. The models are also used directly by airlines and airports to make decisions about flight plans. 

"We've clearly shown that partial loss of aircraft observations does affect the model predictions," James said. "I think that is the most important point."

The team reported their findings this month in the Journal of Applied Meteorology and Climatology

Author Bio & Story Archive

Catherine Meyers is a deputy editor for Inside Science.