Whether the Weather Is Fair

forecasts, meteorology, science, capitalism

“It’s raining,” I told my mom.

“No, it doesn’t start raining until later,” she replied. She was looking at the forecast on her laptop. I was looking out the window.

Photo by Atahan Demir.

I’ve had a near identical interaction with other people. And on a daily basis I hear people say things like, “it’s going to get up to 80 tomorrow,” or “the snow will stop around 9” or “it’s going to be cloudy all week next week.”

The salient feature here is matter-of-fact statements about the future without knocking on wood, opening the back door to uncertainty, or paying the least bit of homage to the spiraling, prediction-defying chaos that constitutes our universe. As though the future were the latest thing people could own. Or rent access to as a subscriber to this or that app.

As I began noticing this tendency more and more, I started to see how people would revise their confident descriptions of the future as their favored weather app changed its predictions, stating the new future with the same certainty, and insisting that the new forecast was the same as the old forecast, that they hadn’t changed their tune by a single note. I’ve practically gotten into duels insisting that people are editing their memory of the future. (Truth: I wish I got in more duels.)

It’s like Stalinist revisionism projected forwards. Rain? There was never any rain in the forecast. Zinoviev? Who’s Zinoviev?

Faced with this distressing trend, I did what any normal person would do: I started a file where I would record forecasts, and then when the actual, in-the-flesh weather arrived, I’d record if the forecast was accurate or inaccurate. I periodically repeated the process over the following 10 months.

And boy do I have things to share!

Weather forecasts were only accurate about 50% of the time! Now, I don’t have any degrees in statistical analysis or modelling, and only 107 data points. However, the sample was random and impartial: I randomly chose days to record forecasts and I included all my prediction/result data points. I took predictions from 3 different apps that people swore by (including the standard Apple app and Weather.com) as well as the National Weather Service.

Total nerds can get more info on the numbers at the very end of this newsletter.

So… does this mean that I’m a flat-earther who doesn’t believe in science? Let’s find out!

What the hell does any of this have to do with capitalism?

Hold your horses, we’re getting there!

The basis of weather forecasts in this day and age is meteorology, the science of studying weather systems on the basis of numerous measurements (ground temperature and temp at different altitudes, humidity, wind speed and direction, dew point, air pressure, wet bulb, dry bulb, etc.); observations of direct relationships (e.g. what happens to humidity when a hot high pressure center moves over a body of cooler water); and increasingly complex computer modelling that attempts to plug in billions of data points past and present, extrapolations of the laws of physics, and numerous long-observed patterns regarding how different elements in this complex system interact. The goal is to predict how the present conditions will change in twelve hours, twenty-four hours, two days, five days, and so on.

Before anyone carts me off to the tinfoil hat room, please let me place my hand on the good book and say “I believe in meteorology!” Contending with an immense quantity of factors in one of the most chaotic systems we can observe (the planet’s weather), meteorology and its close cousin climatology can give us insightful and accurate analysis of what the heck is going on up there. Moreover, they provide a strong basis for fairly accurate predictions of how existing conditions will change moving into the future.

Weather forecasts based in meteorology are clearly not 100% accurate. Similarly, climatologists have pretty consistently underestimated how quickly climate change and related effects will unfold. In other words, they bring useful tools to the table, but there’s still a bug or two in the machinery.

For me, these accuracies and inaccuracies bring up a number of questions.

Do these mistakes in climate forecasting have anything to do with a tendency towards false optimism shared by both scientific institutions like the IPCC and financial institutions like the WEF, which reflect the interests of those who profit off ecocide, and who have an incredibly hard time imagining that capitalism might bring about its own collapse?

What about the use of meteorology in a consumer market? Decades ago, scientific weather forecasts became, in part, a form of TV entertainment to capture people’s attention. TV stations wanted to project confidence and certainty: here are the experts, telling us what the future will bring. Producers of the capitalist forecast didn’t want to teach their viewers about probabilities and multiple possible futures.

Nowadays, the weather app is just like any other app: a tool to take advantage of and reproduce a person’s alienation from the world around them. Producers of apps are more interested in extracting active data from their users, and delivering data and users over to advertisers, then in delivering accurate data to the users. (Similar to how, say, DuoLingo is almost completely ineffective at teaching someone a language, and yet highly effective at getting users to tune in every day.)

What about weather forecasters that aren’t businesses? In my mini-study, I got more data points from the National Weather Service than from any app. And granted, the NWS presentation puts more focus on probabilities than a story of certainties, and they generally offer less information for the forecasts that are multiple days out (whereas plenty of apps will mock up an hour-by-hour forecast days into the future). Still, the NWS exists side by side with the Weather Channel and the apps that condition the expectations of the scientists who work there, and they probably also have bureaucrats whipping them on with performance indicators including website hits and other dross.

Given we live in a society of amnesia, the producers of weather forecasts are less likely to be punished if they provide an inaccurate forecast than if they provide a boring or confusing forecast.

If we can permit ourselves a sophism, what kind of weather forecasts would a counterfactual pure science produce? Of course, there’s no such thing as pure science, since information can never exist without context, and science is inseparable from those who produce it, who are themselves inseparable from their funders, their sponsors, their institutions, their professors and ancestors…

I’ll come back to this relationship between capital and science in future essays. For now, I want to encourage people to get off their phones, go outside, and just breathe. Observe how all the creatures around you respond to changes in the weather, and wonder about what keeps humans from doing the same. What humans seem the most insulated from the real world weather? What are the causes of it?

Do any of you know anyone who is so intimately connected with a territory—maybe they’ve lived there their whole life, maybe they’re just good at paying attention—that they can tell you with high accuracy when it’s going to rain or if the clouds will break? Can you imagine what it feels like to be in that kind of relationship with the world around you?

Until next time, take care,

p

Okay you numbers nerds!

With a sample size of 107 data points (accurate/inaccurate), I found an accuracy of 53.2%. According to https://www.qualtrics.com/experience-management/research/margin-of-error/ the characteristics of my survey give it a 12.4% margin of error. Looking only at results from the National Weather Service, which potentially has access to a larger data set and better technology, there was 55.9% accuracy in a sample of 59.

Corroborating my self-evaluation that my study is low precision and decent accuracy, there is a wide margin but also a general similarity between the accuracy in the winter predictions (55.3%) and the summer predictions (51.7%).

Basically, what I’m saying is my results won’t be very precise, but in broad terms I think they should be accurate. Think of it like this: if a machine can give you the right answer 90% of the time, there’s a decent chance you could pick two random answers and get a right one and a wrong one. However, if you pick 107 random answers there would be an extremely low chance that half of them were wrong.

Does anyone out there know how to calculate how improbable it is? Leave the equation in the comments!

I’ve been looking into a number of studies that claim much higher accuracy in weather forecasts, but I’m still trying to find their original sources and see if I can understand their methodology, because from what I’ve seen I just don’t believe that 24 hour forecasts are 90% accurate and 5 day forecasts 80% accurate unless we’re using a reeeeeally gentle measure of what constitutes accurate.

Here’s the metric I used, and I think it’s pretty generous. I considered a forecast inaccurate:

- if weather predicted as sunny, partially sunny, mostly cloudy, or cloudy fits fully within a different characterization.

- if a same day temperature prediction is 3º off

- if a future day temperature prediction is 5º off

- if the probability of precipitation is given as below 20% or above 80% and it does/ does not occur

- if predicted snow turns out to be entirely rain or predicted rain turns out to be mostly snow

An easy way to notice that weather forecasts have low accuracy unless we grant them a huge margin of error: compare the predictions from a few different apps and weather stations, and throw in the NWS or some other government or scientific body as well. Most of the time, their temperature predictions will vary by a couple degrees or more, their precipitation probabilities by several percentage points, and their predictions for cloud coverage will often also vary. There’s no way that they’re all accurate unless we consider ballpark figures to be accurate.