Weather Forecasting is Fifty Years Ahead of Epidemiological Prediction: That Must Change

In October 2012, Hurricane Sandy approached New York from off the Atlantic.  Seven days before, a near perfect track forecast was made by the European Center and two days later the U.S. model (the GFS), run by the National Weather Service's Environmental Modeling Center, has locked on as well.

The stunningly accurate prediction of an extremely unusual event, saved countless lives, with only about 100 deaths in a region of tens of millions of people (and most who died made a bad decision to ignore the forecast).

Although a hurricane hitting New York is exceedingly rare, the excellent forecast was based on an extraordinarily weather prediction infrastructure that had been perfected over the past half century, with sustained investment and development:

•  A comprehensive observing system, based on weather satellites, surface observations, and more.
• A complex quality control and data assimilation system the ensured a good idea of what happening in real time.
• Highly complex numerical models for simulating the evolution of the atmosphere, models that had been tested and perfected over decades.  And a dedicated U.S. forecasting center responsibly for state-of-science prediction.
• A mature statistical postprocessing system capable of improving the model forecasts based on past performance.
• A comprehensive verification system to provide detailed evaluations of the skill of the forecast.
• A highly developed communication system, that provided the public with clear interpretation of the forecast.

Excellence in prediction took time and investment over decades, and paid off in warning the public and guiding public officials in protecting the population.

Stunningly, U.S. epidemiological modeling has almost NONE of the above components or systems, and the performance of the U.S. Centers for Disease Control and Prevention (CDC) as poor.    Consider:

• There is no large, permanent epidemiological prediction center in the U.S. government analogous to the NOAA/NWS.   U.S. response to COVID had to depend on a hodgepodge of university forecasting efforts, some spun up for this event.  Untested, unverified, and often very wrong.
• Unlike weather prediction, U.S. epidemiologists do not have an accurate description of what is happening NOW.  Testing was slow to begin, in fact, the CDC made serious errors in test development.  Even today, six months into the pandemic there are not enough tests.  There is no attempt being made to randomly sample the population to understand how many folks have or were infected.  Unbelievable.
• Unlike the comprehensive and smoothly running weather observational system, there is poor organization to the data collection by CDC. There are not even standard reporting approaches.
• The epidemiological models are generally primitive affairs, many of which do not consider the complex, variable transmission properties of a heterogeneous population, and lack clear information of what is occurring right now (called the initialization in weather prediction).
• Communication by CDC of both the threat and how to deal with the disease has been inconsistent and often wrong.  For example, they initially discouraged the use of masks, before reversing their guidance 180 degrees.  Similarly, CDC downplayed the threat in January and February, before reserving in March.  The NWS works very hard to start with reliable forecasts, to communicate the uncertainties, and not to go back and forth in their warnings.  They are masters at this.

So let us imagine what the Hurricane Sandy forecast would have been like if the NWS followed the CDC approach.

• The storm would have been out there, but without comprehensive observational assets, they would not have known where it was.
• Without a government prediction organization, the NWS would have asked for volunteer forecasts of University modeling research groups (like the University of Washington).  They would have had 5-10 forecast from various universities that would have diverged by position and intensity.  Without good initialization data, none of the forecasts were skillful.  And furthermore, without long-term verification, no one knew how good the forecasts were.
• Based on this guidance, the NWS could not provide specific, accurate forecasts, suggesting that there was a storm out there, but it could hit anywhere from Georgia to Maine, or might go out to sea.
• With such uncertain forecasts, political leaders pressured to evacuate the entire coast from Georgia to Maine at a huge cost.  Many would not evacuate under such vague warnings. With landfall on NY and many remaining in their homes, nearly 4,500 people died

The Political Opportunists

        There are some folks and many media pundits who are claiming this COVID disaster is all the fault of President Trump and that things would have been much better under a Democratic President.

This is either very naive or very cynical.

There is no doubt that the President and his administration has been startlingly misinformed and ineffective. His abysmal leadership has made things much worse.

But the problems noted above are not recent developments and have been allowed to fester in recent administrations, including the 8-years of President Obama.  I suspect we would not have been in much better place if Hilary Clinton would have won, because the basic institutional infrastructure was not put in place.

That is what we must do together as a nation, following the example of the weather prediction community.  And speaking as one of them, we would be glad to help.

from Cliff Mass Weather Blog https://cliffmass.blogspot.com/2020/07/weather-forecasting-is-fifty-years.html