Last year Los Alamos statisticians beat out more
than 20 teams in a CDC competition aimed at
improving flu forecasting using supercomputing
power. The lab’s “Dante” model was most
successful in predicting the peak and short-term
intensity of the unfolding flu season — and
became the basis for the new COVID-19 model.
That model shows the likelihood that a state
has hit its daily infections peak and may be on
a downward slope. It also offers hurricane-style
probability forecasts for infections and deaths
in each state in the coming week, as well as a
longer six-week period.
For the state of New York, the most recent
forecast late last week showed a 60% chance
that infections already have peaked and may
be waning.
Behind the results is a model that runs on a
supercomputer at the national laboratory but
avoids complex, “what if ” assumptions, said
Sara Del Valle, a deputy group leader of the
Information Systems and Modeling Group that
creates the forecast.
Instead, it builds primarily upon infection and
fatalities data from John Hopkins University.
W ithout delving into state-by-state social
distancing measures, the model assumes
in general that the effects of implemented
interventions will continue into the future.
As leaders worldwide try to get a handle on
the coronavirus outbreak, they are turning to
numerous mathematical models to help them
figure out what might happen next. Some
models extrapolate the effects of existing social
distancing for a certain time period, or play out
a variety of assumptions about the effectiveness
of interventions against the virus.