One of the core competencies of meteorology, mathematical modeling, has been in the limelight in recent months, with epidemiological modelers producing a wildly varying array of model-based predictions on the potential spread of Covid-19. Across the board, governments have been heavily reliant on these to inform their responses to the disease. Now, modeling methods normally used to forecast weather have been borrowed to predict how rapidly Covid-19 could spread in different countries as lockdown is eased, as well as assess the effectiveness of measures put in place.
Meteorologists from the University of Reading and National Centre for Earth Observations in the UK were part of an international team that applied data assimilation – a technique that combines multiple sources of information to estimate how a situation will develop over time – to the pandemic. It suggested it is possible to make reasonably accurate predictions of how easing measures might affect the spread of the virus up to two weeks in advance.
This technique is usually used to pair computer simulations with real weather observations to forecast future weather. Previous computer model forecasts can be tested against the subsequent weather data to help make future short-term predictions more accurate.
When applied to the coronavirus, observations including hospital admissions, the number of patients in intensive care, and the number of daily deaths can be combined with models calculating risk of vulnerability, exposure, infection and death.
Dr Javier Amezcua, one of three Reading scientists that worked on the study, explained, “Most data is uncertain to some degree, but combining as much of it as possible from different sources can iron out some of this uncertainty when predicting future events. We decided to apply this technique we routinely use to the most uncertain situation facing the world right now – Covid-19. Although we are not experts on viruses, our findings demonstrate the potential of measures that reduce the amount of contact people have with one another to save many lives.”
Professor Alberto Carrassi, coordinator of the team at Reading, said, “This has been a genuine interdisciplinary exercise that demonstrates how methods and approaches rooted in one discipline can be applied beyond their original terrain of applications, creating opportunities for further cross-fertilization.”
The research, submitted to the journal Foundations of Data Science, allows estimations to be made of how situations will develop in different scenarios, and can be used to create longer-term forecasts. This means it could be useful to predict the impact changes to lockdown policies, such as reopening schools and shops or increasing permitted socializing, might have on the spread of infections.
The team applied the technique to estimate coronavirus spread in eight different countries around the world – England, France, the Netherlands, Norway, the USA, Canada, Brazil and Argentina – which have all seen the virus spread in different ways.
In England, the now-familiar R number – the number of people a person with Covid-19 is likely to infect, was altered by the team to create three possible scenarios: where it reduced to 0.5, increased to 1 or increased further to 1.2.
Although uncertain, the approximate number of deaths projected by September 1 in each scenario were 57,000 (R=0.5), 63,600 (R=1) and 76,400 (R=1.2). The number of deaths in England as of 1 June was approximately 45,000.
Dr Alison Fowler, the third Reading researcher who worked on the study, said, “Understanding the uncertainty of the model and measurements is crucial to the success of data assimilation. This was particularly challenging to quantify when dealing with reported deaths due to Covid-19, hospitalizations and number of positive cases, in which the collection of data is complicated by so many political and social issues.”