lllustration: USA - Election Day 2020 by Phil Roeder from Des Moines, IA, USA [CC BY 2.0] via Wikimedia
How can our weather forecasts be so wrong? At least in some way, it's related to our awareness that our forecasts cannot predict very well beyond a week out. Why? Technically, the variability in the forecasts, as represented in the global ensemble forecasts, becomes so large that there is no signal within that forecast ensemble that we can rely on. Or, there is no signal within the noise of uncertainty.
Long range forecasts do rely on longer scale changes that are predictable, allowing one to look at deviations in the mean temperatures and/or precipitation from weekly or monthly averages. These longer scale changes are connected to "forcings" related to ocean currents and temperatures (like La Nina/El Nina) or changes in snow cover over the normally colder areas of the arctic and Siberia. Yet, there are so many teleconnections among the oceans and large scale atmospheric circulations that errors in the prediction of even one of them can lead to large errors in the seasonal forecasts or even predictions for regional climate change. Perhaps, that’s why so many seasonal forecasts are not useful.
Forecasts that can be inconveniently wrong, are not just limited to weather. Hillary Clinton was forecast to have a 91% chance to win the 2016 election. Even though Clinton was ahead by roughly 3 percentage points, the pollsters made their prediction, ironically ignoring their own statement that "As a rule of thumb, we should expect Election Day polling averages to miss by as much as 3 to 4 points." Yet, as noted in the aftermath of that 2016 election by Dr. Pradeep Mutalik, a research scientist at the Yale Center for Medical Informatics: “It’s the overselling of precision.” How pollsters calculate the margin of error affects their models to predict the improbable, deeming it impossible.
One might also take umbrage with predictions concerning the continuing saga of the coronavirus. A top Israel professor predicted in April that the coronavirus would disappear after 70 days. On the other hand, at the beginning of the pandemic, a top British scientist erroneously predicted: "Our estimates suggest the impact of the unfolding epidemic may be comparable to the major influenza pandemics of the twentieth century” in which as many as 50 million people may have died. Earlier projections from the Imperial College in London suggested that 2.2 million would die in the U.S. by September, while the Centers for Disease Control predicted 200,000 to 1.7 million deaths. So far, the actual numbers in the U.S. are about 230,000 and about 1.2 million have died worldwide.
So, you can see that forecasts or predictions are sometimes just wrong. All predictions suffer from a combination of difficulties. Both tomorrow's and next month's weather forecasts require an accurate assessment of the oceans (currents, temperatures), land (vegetation and ice) and atmosphere (temperature, humidity, etc), as well as programs to simulate the interactions between each of them.
Pollsters need to better estimate people's preferences, their desire to hide them, and their likelihood to act upon them when voting. Epidemiologists must estimate infection rate, and how the infection rate may change due to mutations and people's behavior. Mutations can also alter bacteria and viruses to be more or less infectious or deadly. And degrees of expertise and experience can affect the accuracy of forecasts in very different disciplines.
It's at times like this that we could use a "Harry Seldon," a fictional ‘psychohistorian’, who predicted the future in probabilistic terms to suggest appropriate actions to prepare for events in history. Were he to appear, I would ask him, “Will we have a winter of plentiful rains, who will finally win the U.S. election, and when will the coronavirus pandemic be behind us?” I would want to check my answers: yes, rain will be at least closer to normal than predicted; Trump will win; and next year will see the end of the pandemic.
Last Sunday, I thought that you could ignore my prediction about Trump and the coronavirus because I was just guessing based on intuition (and intuition based on faulty knowledge can also be wrong). Yet, even now in the immediate aftermath of the election, the results showed pollsters did indeed underestimate Trump’s votes, although in the end not enough to bring him a second term. It's also unlikely we will have such low seasonal rain amounts, and maybe even less likely that the seasonal models can be right in any year.
Dr. Lynn is a lecturer at The Hebrew University of Jerusalem, Earth Sciences Department. He is also CEO of Weather It Is, LTD, a company that specializes in reducing weather risk. Click here to read more of this writer’s work in The Jerusalem Herald.