Tuesday, May 28, 2013

Why Can't We Handle Probability in Weather Forecasts?

Our minds don't easily handle probability. Complex systems like the global economy, the financial sector, ecosystems or quantum mechanics are riddled with counter-intuitive randomness. We can visualize the movement of an electron as a point particle flying around a nucleus of an atom. Yet the real picture is filled with uncertainty. The electron's existence is a hazy cloud of probability. Its nature is not just a particle but BOTH a particle AND a wave! Say what? 

Much of our human experience is filled with such truths. We especially hate probabilities in our weather forecasts. Why? For our minds to grasp probabilities, we need to be able to handle multiple possible outcomes at once. Just our luck, weather has many, many outcomes over a large area over a significant period of time. Change the initial weather conditions (humidity, wind flow, frontal position, upper level energy, etc) and you create more uncertainty. Factor in time and the probability becomes significantly higher.

Typically, our brains work much better with a theme that is linear: A story that has a beginning, middle and an end. We want to visualize a line of showers that moves in at a specific time, stays for a select amount of time and then moves out without fanfare. Unfortunately, rain events rarely behave in this manner.

Here is a quick radar loop from May 28, 2013. Notice the disjointed nature of the rain/storm clusters and how they evolve. Some smaller cell develop independently of the main cluster. I guarantee that by the time they made it to Ohio, they looked nothing like what you are seeing here.

The radar loop above is an excellent example of why—much to the chagrin of the general public—probabilities are the only way to tell the weather story. We use 90% chance of rain, 40% chance of rain, etc. Yet if it doesn't rain over their house when the probability is 90% chance of rain, the forecaster is wrong even if the rest of the area was hit with a good downpour. We want to know if it will rain or not; a black and white scenario without caveats. Yet the behavior of some small scale weather events like warm frontal rain/storms can behave semi-independently of the overall large scale pattern. I’ve tried multiple times to convey this idea on the air. The explanation of small scale rain clusters as behaving somewhat “on their own” falls on deaf ears.

It all goes back to basic human nature. A good weather narrative (a feel-good forecast with some folklore) is desired versus something data/science driven. Nebulous weather data and science makes most of us feel uncomfortable even if the on-air meteorologist has the best of intentions. We have created some sophisticated models of the weather that can make some very good “probabilistic” outcomes for weather events and situations. Yet a level of uncertainty still remains and we humans don’t like it!  We try to rationalize the irrational. Our biases quickly dismiss the probabilistic science as irrelevant or at the very worst, an excuse.

Instead, we favor more simplified stories even though that story might gloss over important details. Our minds involuntarily cherry-pick elements of the story so that it fits our biases. Think of a time when someone told you a weather fact or forecast which you didn’t believe. You felt uneasy. Your mind shrugged it aside only to be replaced by a story, forecast or explanation that made you feel better…accuracy be damned.

A great financial blog called The Big Picture written by Barry Ritholtz explains the narrative vs data idea succinctly: (I inserted the weather components)

* Narratives (straight forward simple weather forecasts) are about hitting emotional buttons making the reader feel good by focusing on less qualitative aspects (weather science and probability) of an issue.

* Narratives (weather forecast) are/is about the outcome not the process (explanation of the science and probability)

* The process (weather science) is important in developing solid results

 So remember the psychology. How you react when you hear a weather forecast?  Do you dismiss the science? How do you handle probability?  Do you like hearing an explanation to why the weather does what it does? Do you overly simplify the weather? Are you aware of your biases?

The science of the atmosphere is never as straight forward as we'd like it to be…and never will.


Eric Barnett said...

This was a great explanation. Thank you for sharing.

Eric Barnett said...

This was a great post. Thank you for sharing!

Dennis Boylan said...

So, would it not be better to provide an hour by hour forecast on a daily basis vs an 8 day forecast?

I rather be able to wake up in the morning and see an accurate detail forecast for the day.

Would also rather like to see the detail forecast by cities like "Here is todays forecast for Westlake, Painesville, Port Clinton" since as you have said many times, the weather can change significantly from one location to another.

This was I have a better and safer understanding of what the day holds for me.

would also like to see you offer some training classes on what a person can do to figure out what the day has in store. If I look at various cloud formations what does that mean? What does red sky at night mean?

Scott Sabol, Meteorologist said...

Dennis, very good points. Details for specific cities for each day are very difficult. We would be grossly generalizing large scale data for locations. Believe it or not, the forecasts would be harder than you think. We don't have computer model guidance for resolutions down to say 5 km that would be needed to create a computer model projection for say Westlake or any city. We are decades and decades away from being able to determining when a pop up thunderstorm will occur over a specific city at a specific time. So highly accurate hourly forecasts for specific cities are still loaded with uncertainty and randomness. Those type of details are still beyond our reach even if some apps say they can do it. Believe me, they can't.

As fro the 8day forecast, polls continue to show that people want to see an extended forecast more than anything. I agree that an 8day is too long...a 5 day is more appropriate.

Dennis Boylan said...


Thanks for the update and the comments. I can see your point about the accuracy of any one given location.

To my other question; what about some training/seminars that would be helpful in understanding what is going on. If I were to get one of those home weather stations, would they be helpful in understanding what MIGHT be happening for the day? How can I look at cloud formations and understand what might be coming.

I have attended a couple of classes taught by Mark Thorton at www.LakeErieWX.com. His focus is on understanding the weather on the lake for boaters which is one of the drivers for my questions. As a boater understanding the weather is a major safety issue. Any education is helpful

I plan a lot of my trips based on your forecasts and they don't always quite work out they way we would like them to!!:)

So any classes, seminars, etc. would be appreciated!

Dennis Boylan said...

The article today by Dr. Ricky Rood,that you referenced further validates the challenges with understanding a forecast. In the paragraph you referenced ( point 1 paragraph 3), he states" Probability and likelihood are notoriously difficult ways to communicate in quiet consultation, and even more difficult in newspapers, on the radio, television and online. Probability and risk are just made for conflicting headlines.

So when we see a forecast with an 80% probability of rain/snow, there is a lot of room in that forecast to be either 100% right or 100% wrong; ending up some where in the middle.

Thank you though for your comments and insight as I continue to try to understand if its going to be a good day or not to take the boat out!