It’s got a Harvard pedigree, a presidential Twitter endorsement, and a surprising suggestion about the origins of the coronavirus pandemic.
The authors behind a new Harvard study say web searches and satellite pictures of increasingly full parking lots in Wuhan, China, show an unusual increase in traffic to hospitals at the epicenter of the outbreak months before the Chinese government first warned the world about the spread of the virus.
But some satellite imagery experts aren’t so sure. They say the imagery by the Harvard team obscures uncounted vehicles and doesn’t include enough evidence to support the idea that Wuhan’s hospitals had an unusual spike in traffic in October 2019.
“There are other high resolution images showing ‘surge’ days earlier in the year that are not included in their dataset. They just missed the contrary data,” said Jeffrey Lewis, director of the East Asia Nonproliferation Program at the Center for Nonproliferation Studies.
In particular, Lewis pointed to a satellite image captured by Maxar and published in Google Earth which showed a nearly full parking lot at Hubei Women and Children’s hospital on May 31, 2019, well before anyone believes the virus began circulating.
The study, still a preprint awaiting peer review, was authored by a team of researchers led by Dr. John Brownstein of Harvard Medical School. The team worked with the satellite imagery analysis firm RS Metrics to count cars in five major hospital parking lots in Wuhan and combine it with search query data about influenza-like symptoms from the Chinese search engine Baidu. The results appear to show a spike in traffic to hospitals in Wuhan in October and November, months before China officially notified the World Health Organization of an outbreak in the city.
“While we cannot confirm if the increased volume was directly related to the new virus, our evidence supports other recent work showing that emergence happened before identification at the Huanan Seafood market,” the authors wrote.
That conclusion has taken off on social media—President Trump himself tweeted out a Fox News segment about the researchers’ work—and led to questions about when the virus first started circulating.
The question of when the outbreak began goes not just to scientific curiosity but international culpability. China initially suppressed information about the outbreak before it officially notified the World Health Organization of a possible outbreak on Dec. 31. The Harvard study’s conclusions suggest the possibility that the virus may have been circulating as early as August 2019, suggesting Chinese officials could have known more about a potential problem and earlier than is commonly understood.
But some imagery experts aren’t so sure.
When the Harvard team shared imagery of its parking lot counts with ABC News, a number of imagery experts saw problems with the methods used to count cars.
The car counts visible in the published images were taken at off-nadir angles—not looking directly downward—and appear to obscure parts of the hospital complexes where uncounted cars could be parked.
Chris Biggers, a former satellite imagery analyst for the National Geospatial Intelligence Agency, reviewed the imagery released to ABC and found that the slant angle of the imagery obscured potentially parked cars areas at Hubei Women and Children’s Hospital, Zhongan Hospital of Wuhan University, Wuhan Tongji Medical University, and Wuhan Tianyou Hospital.
“An initial review of the imagery to determine if the areas were captured correctly should have been undertaken to understand ‘what’ and ‘when’ comparisons can be made. To make that determination an analyst would simply orient the image toward the satellite sensor to review. It’s by far one of the most rudimentary steps to doing imagery analysis,” Biggers wrote.
Rotating the images in the same direction as the satellite took them should have led the authors to understand whether the parts of the hospital parking lots targeted for analysis were actually captured, according to Biggers.
In an email to The Daily Beast, RS Metrics’ Tom Diamond wrote that he believes any undercounts due to the angle of imagery wouldn’t be enough to compensate for what he says was a broader and consistent trend that the research team observed across multiple hospitals. “Yes, there may be some error here in a few of the images of Hubei Women and Children’s because it has such a tall building, but no more than maybe 20 or 30 cars. If we take those images out, we still see the same trend with the remaining images. If we were to entirely remove Hubei Women from the study, the results are still the same for the other hospitals.”
The study is based on 111 satellite images taken over the past two years and yielded “140 successful daily extractions of parking lot volume from hospitals,” according to the authors. But imagery analysts have also questioned whether the research team’s selection of images may have missed potentially higher volume days at hospitals earlier in the summer of 2019, which could interfere with the appearance of a trend.
Biggers also found Maxar imagery of the Wuhan Tianyou Hospital from June 13, 2019, on Google Earth. The imagery shows a parking lot with about 231 vehicles visible, only 54 vehicles off one 285-vehicle tally at the hospital observed during the supposed peak period of traffic in October.
The study builds off a similar methodology used by the same research team for a 2015 study on Argentina, Chile, and Mexico. In that work, the authors used historical satellite imagery of hospital parking lots in the three countries to correctly predict influenza and respiratory virus outbreaks.
Biggers, however, still has questions about the researchers’ work on the Wuhan study. “If the comparisons published are examples of the methodology employed to get at the counts, I would have serious reservations about the count accuracy.”