In the lead-up to last weekend, computer models projected that southern Florida would see a deluge of several inches of rain. Such abundant rainfall would have been a sharp departure from the unusual heat and dry conditions that have gripped that part of the state so far this year.
In the end, though, Miami ended up seeing just 0.26 inches of rain, and the medium as well as shorter-term model projections turned out to be a complete bust. A forecast miss of such a magnitude is relatively unusual, and it prompted some meteorologists to wonder whether missing data because of the coronavirus pandemic might be to blame. Normally gathered by commercial aircraft, much of the data has stopped flowing since many flights have been canceled.
The answer, experts say, is that it’s unlikely that certain missing data was the cause, but it’s also unlikely that the data shortfall is not having some negative effect on forecasts.
Commercial aircraft are a forecaster’s secret weapon
For years, thousands of airliners and cargo planes have been involved in a side gig that few passengers or package shippers have been privy to: gathering and transmitting weather data that’s then used for improving weather forecasts issued worldwide.
To gather crucial information for operating the plane itself, such as determining its airspeed and motion relative to the ground, instruments on an airplane’s nose — they resemble pointy, metal needles sticking out of the fuselage, pointed forward, into the airstream — measure the characteristics of the air mass around them.
This includes the wind direction and speed, air pressure, temperature and, in some cases, humidity as well.
Government programs run in the United States, such as the efforts of the National Oceanic and Atmospheric Administration, along with international agencies and private companies have tapped into some of this data to pipe it into computer models that help meteorologists forecast the weather. This includes the National Weather Service’s Global Forecast System, or GFS model, and the world-leading European model, run by the European Center for Medium Range Weather Forecasts (ECMWF) in Reading, England.
At the global level, networks that vacuum up data from planes flying in Europe, Asia and South America have helped to ensure an ever-growing flow of data that’s benefited modeling centers worldwide, boosting model skill and forecast accuracy.
However, with the paucity of flights in the air now, the pandemic has disrupted that data flow, turning it from a fire hose to more of a trickle. Meteorologists don’t know exactly how much of an impact that’s having on forecast skill.
In the United States alone, airlines are operating skeletal schedules and have opted to store much of their fleets rather than continuing to lose money by flying near-empty planes. According to the trade group Airlines for America, U.S. airlines idled 3,162 planes, which accounts or 51 percent of their fleet, as of May 10.
Airports such as Tulsa International, Kansas City International and Pinal Airpark in Arizona have been transformed into parking lots for billions of dollars worth of jets.
U.N. agency sounds alarm as NOAA tallies missing observations
Meteorologists know that the aircraft data makes a measurable impact on forecast skill, but they’re not sure how big of a hit forecasts are taking right now. This uncertainty is potentially problematic, considering the start of what is expected to be a busier-than-average Atlantic hurricane season looms on June 1.
The World Meteorological Organization (WMO), a United Nations agency, issued a statement May 7 expressing concern about coronavirus-related deteriorations in aircraft-based and ground-based weather observations. The WMO found that, globally, measurements from aircraft have fallen by an average of 75 to 80 percent compared to average, “but with very large regional variations.”
For example, in the Southern Hemisphere, the data loss is “closer to 90 percent,” the WMO stated.
According to the WMO, when operating at full capacity, the global aircraft-based observing system produces over 800,000 weather observations per day using planes from 43 airlines and several thousand aircraft.
In the United States, as of April 20, “the daily output of meteorological data from U.S. commercial aircraft has decreased to approximately one quarter of normal levels,” according to NOAA spokesman Chris Vaccaro. When at full capacity, aircraft soundings, as such data is known, dwarf the volume of the twice-a-day weather balloons launched from 900 weather stations around the world.
In a typical year in the United States, more than 3,500 commercial aircraft provide more than 250 million observations per year, according to a NOAA statement. (Not every aircraft in the sky is part of the observing network).
Curtis Marshall, who manages the program, known as aircraft meteorological data relay, or AMDAR, for the Weather Service, said right now, only about 1,000 aircraft are sending the agency weather data each day, given coronavirus-related flight reductions.
“Our number of daily aircraft profiles is fluctuating between 20-30 percent of normal, depending on the day of the week,” he said in a statement.
Marshall says European nations have seen a greater reduction in observations, in part because the U.S. network involves cargo carriers that have not cut back on their operations. Also, the network’s largest domestic contributor, Southwest Airlines, is scheduling more flights relative to some of the other major U.S. airlines.
In addition, about 130 of the more than 3,500 commercial planes in the domestic network have a special moisture sensor as well as wind and temperature measuring equipment, Marshall said. Luckily for the NWS, those sensors are on Southwest and UPS aircraft, most of which are still flying.
“That moisture data is very valuable for the models, particularly as we approach severe weather episodes,” Marshall said.
Teasing the impacts of missing data is impossible
Aircraft provide forecasters with the advantage of traveling at all times and gathering unique profiles of the atmosphere as they change altitude over areas that may not have been closely studied by other observing platforms. Such areas include the jet stream level above the Pacific Ocean and the data-sparse region between Africa and Australia, along with the area above the North Atlantic.
According to the ECMWF, aircraft reports come in second, behind satellite data, in terms of their influence on forecast accuracy. In early March, as European countries began enforcing their shutdowns, aircraft-generated weather reports across Europe declined by 65 percent. There was also a 42 percent global decline in such reports during the same period.
The ECMWF has investigated the impacts that removing aircraft data has on its forecast model simulations and found that there is a particularly large effect at the jet stream level, between about 30,000 and 40,000 feet, where planes typically cruise during flight.
Bruce Ingleby, a senior scientist at ECMWF, said the loss of all aircraft data from the European model would decrease the accuracy of a 12-hour forecast of winds and temperatures at the jet stream level by about 10 percent, with a statistically significant degradation of 3 percent when it comes to surface pressure forecasts going out to three to four days later.
Ingleby said that, ironically, those who might suffer the most from the missing data may be airlines themselves, since forecasts of upper level winds are used to calculate fuel needed, flight routes and times.
But upper level data is also crucial for forecasting the evolution of weather systems, he said.
“I think it’s fair to say that covid-related reductions might cause a noticeable reduction in the forecast quality, but it gets really difficult to really pin it down,” Ingleby said.
“We don’t want to say there is no impact, but we don’t want to say there’s a great big impact and any bad forecast is due to it,” Ingleby said.
NOAA says it’s not seeing a clear trend in the data that jumps out and corresponds to the loss of AMDAR data.
“At this time, there is no clear signal/trend in the skill of the numerical guidance the models are producing, and determining any such impact of the lost aircraft data on model skill requires a rigorous and substantial scientific analysis,” said Brian Gross, who leads NOAA’s Environmental Modeling Center, which runs and tracks the skill of the agency’s computer models.
In addition, NOAA’s Vaccaro cautioned that degradations in model skill don’t necessarily translate directly into less accurate weather forecasts. “Even if a decrease in aircraft observation data impacts forecast model skill, it does not necessarily translate into a reduction in forecast accuracy since National Weather Service meteorologists use an entire suite of observations and guidance to produce an actual forecast,” he said.
Ingleby and Marshall said it’s impossible to pinpoint how significant the impact of missing data actually is, since one would have to run an experiment that involves models deprived of aviation data alongside a control group, which would have a complete set of data for the same time period.
Right now, such an experiment is impossible, since data is in fact missing.
To help combat the data loss, agencies are enlisting help from an airborne observing network that a private company, FLYHT, operates. They’re also ordering more frequent weather balloon launches in particular areas.
As for forecast situations like Miami’s this past weekend, the standard weather forecast disclaimer applies more than usual: All forecasts contain an element of uncertainty to them, perhaps more so now than at other times.