Weather Forecasting: What It’s All About

As a meteorologist, it is difficult for me to turn that part of myself off wherever I go.  When people try to make small talk about the weather with me, they usually regret it, because to me weather isn’t just to pass the time, but it’s interesting.  However, for most people it is small talk.  Invariably whenever I go out in public I overhear conversations about the weather, whether in restaurants, museums, the locker room at the gym, etc.  There is a common theme I hear which is that forecasters don’t know what’s going on.  I hear and read things like prediction is not even possible.  I have had people come up to me and jokingly say, “Meteorology must be a great profession because you can be wrong 50% of the time and still have a job.”  Though they clearly are telling one of those jokes that they actually believe to be true, the chuckle I give in return is not nearly as sincere.  To explain would require more time than I often have, so I thought maybe I should write a post explaining some of the basics, and explain some of the most common things people misunderstand about the weather.

Why Forecast?

There is a lot to the history of forecasting, but I think it’s fairly clear why people want to have some knowledge about what sort of weather was coming their way.  Whether you wanted to know when to plant crops, went to harvest, or where and when to sail your ship, having knowledge of the atmosphere, knowing what weather is coming your way has huge advantages.  The beginnings of forecasting as a science were driven by WWI when aviation was added to warfare and they quickly realized that having weather observations and an ability to know what weather was coming was a huge tactical advantage.  The weather forecast is something almost everyone utilizes today, whether you think it means something or not, people will still look to try to make the best guess about what to wear, whether to bring an umbrella, or whether travel or not will be hazardous.

How Accurate are Forecasts and Why Do People Think They Are Not Accurate?

First things first.  Currently the National Whether Service is accurate for 1 and 2 day  temperature forecasts to within 2.5 – 4 F, and has an 82% accuracy rate when it comes to precipitation.  I am not going to spend a lot of time proving accuracy here, you can check out these links as they have already done the work.

What I will say is that it is important to understand how our cognitive biases shape our perception.  The one at work here is that what sticks in memory are the misses, and not the hits.  When the forecast is right, you don’t think about the forecast.  When it’s wrong you do.  This creates a data point in your brain only when there is a missed forecast, but it’s poor way to draw meaningful statistical conclusions.  I think it’s also important to note that I see a lot of click bait type headlines for upcoming weather and this may be what’s drawing our attention.  Extreme gets clicks, but may not be what’s being endorsed by the National Weather Service.  It’s also not clear whether people are staying current with the latest forecast.

Finally I think it’s important to remember that in extreme weather situations forecasters will err on the side of caution.  It is a difficult line to walk.  When extreme predictions don’t happen, the public loses trust in your forecast, and this can cost you lives in the future.  If on the other hand you don’t communicate the possibility of an extreme situation, that can also lose you lives.  So in erring on the side of caution, more often than not people will find that it might not have been quite as bad as predicted.  Erring on the side of caution is the right thing to do, because there is an inherent error to the forecast.  Sometimes in the margin of error, the extreme end of that error can be the difference between life and death.

Hard Rain

Precipitation is the hardest variable to forecast for and some of the reasons for that are given in the next section, but a few points are worth talking about here.  First, many people don’t understand the precipitation forecast.  This has been a criticism of the National Weather Service to change their way of forecasting precipitation, but for now, there seems to be no better way of doing it.  When the probability of precipitation (PoP) is reported, it is reported as a percent.  But what does that percent mean?  This probability is actually the product of two other numbers.  One is the actually chance of precipitation but the other is the percentage of the forecast area that will be impacted by precipitation.  Each National Weather Service office has a specific region in which they are suppose to forecast for and they usually break those down into smaller regions for the purpose of precipitation forecast, but the fact remains that incorporated into that PoP is areal coverage.  So 50% chance of precipitation is not the coin toss that some make it out to be, but it could mean that there is a 100% chance of rain over half the forecast area.  Of course it could also mean that there is a 50% chance over all the forecast area.  But it’s also important to remember that even in the latter case it’s not a coin toss, but rather based on evidence that pegs precipitation as more likely. The difference between rain and no rain can often be very small and requires knowledge of atmospheric properties at high resolution.  A far higher resolution than we have.

Snow forecasts are often worse, and this is largely due to two factors.  One, is that it depends on temperature whether you get rain or snow.  It takes a very slight error in the forecasted temperature for rain to suddenly become snow or vice-versa.  So being 2 F off in our forecasted temperature may make no difference in what you wear for the day, but it can have huge impacts on what driving conditions are like.  The second important factor here is that water expands when it freezes such that the ratio of snow to liquid precipitation is 10:1.  Forecast models only determine the precipitable water for a particular area.  If that prediction is off by 0.2 inches this could be the difference between 1 and 3 inches of snow, which is a rather big deal when it comes to driving.  But it’s not always a matter of the forecast model being wrong in terms of precipitable water.  Across any storm system there is going to be variation in the amount of precipitable water and thus getting the storm track exactly right also matters.  Mix this in with a possible slight error in forecast temperature can lead to a vast difference the amount of snow accumulation for a particular location.  On top of that the 10:1 ratio is more like 7:1 if the snow is really wet, so this adds error into the forecast as well.

Weather is a matter of Scale

Image result for synoptic scale mesoscaleA lot get’s said about the difference between weather and climate, but very little is said about differences among various types of weather systems.  Typically, the average meteorologist separates scale into 3 categories.  Turbulent eddies near the surface to convection currents in clouds make up the microscale (< a few km) A thunderstorm or a system of thunderstorms or series of cloud bands for lake effect snow would be part of the mesoscale (about 10-100 km, several hours), and then things like low pressure systems would be on the synoptic scale (about 1000 km, several days).  Our ability to forecast events along these scales depends largely on our ability to make observations smaller than the scale we are trying to predict in both space and time.  For instance if I am at a station 100 km away from the nearest station, even if I make continuous observation a thunderstorm that happens somewhere in between will not be observed by me.  When you look at the number of tornadoes in the U.S. over a 100 year period, you will see a dramatic rise in the number from a few hundred to over a 1000.  This is no climate change phenomena, but a matter of our ability to observe tornadoes, and the advent of a national radar network that dramatically increased our ability to determine where tornado producing storms were.  Similarly if I make observations only two times a day, I’m unlikely to be able to resolve well the changes that occur between those observation times.

Computer models that forecast weather have similar problems.  Computer models operate by breaking up the atmosphere into a 3-D  grid that then processes the physical equations that describe the atmosphere at equal time intervals.  The size of these grids and the spacing of the measurement network that gives the initial data for these models to work lends itself best to the forecasting on the synoptic scale.  What this means is that we are likely to best forecast the development and movement of low pressure systems and high pressure systems, and forecast widespread rain.  The timing and movement of individual thunderstorms represent processes that occur at the sub-grid level.  In essence, noise.  Obviously a potential hail or tornado producing thunderstorm is not really noise, but this is why your forecaster is pretty good at telling you when that cold front is coming through the next day, but not so good at pinpointing where thunderstorms will be the strongest.  That type of accuracy is usually only made several hours in advance.  Although we’re pretty good at assessing a day or two in advance which day will have a high potential for thunderstorms.

Practical vs. Theoretical

When it comes to the theory about how weather works we are, in general, ahead of the game, but practical considerations take precedent.  For instance we could do an excellent job of forecasting if we had weather data every 10 km over the surface of the earth and sent up weather balloons once every hours.  The cost however of such an enterprise would be enormous.  Especially considering it’s very difficult to get this information over the ocean.  Remote sensing devices like satellite and radar are making strides in provide better spatial coverage, but even those have limitations.  We are never going to have perfect data over as wide a range and as often as we need it, and this is always going to lead to some error.  Computer power is also a practical limitation although it has accelerated greatly since the first model.  Previously, with all the theory we knew, trying to create a model that matched our data network would have taken the computer so long to produce an output that the time we were trying to forecast for would have been past.  This is no longer a terribly relevant problem, but it is if we really want to be able to break into models that compute both synoptic and mesoscale features.  It’s a bit hard to explain but you can think of a computer model as potentially like a nesting doll.  We could run a model at the smaller scale within in each grid of the synoptic scale model.  So a model within a model.  That becomes computationally laborious and can take intense amount of computer processing power.

Image result for sounding stations north america weather
It might look like a lot, but these are all the weather balloon stations in North America. Typically one or two per state and at minimum more than a 100 miles apart, and in some cases more than 500. Canada is worse. Not near enough for getting good data for thunderstorm formation.

Then we have the reality of cost-benefit analysis.  Decisions about weather research and preparedness have a lot to do with what the costs are.  This is hardly surprising.  If snow is rare in your city you might find that it’s easier for the city to just close down for a day than spend a lot of money on snow plows.  As mentioned above, to take ideal amount of measurements would be of great cost and despite the scientist’s love of data, the question must be asked do the gains in forecast accuracy outweigh the costs.  Improved technology can help reduce cost and make instruments more maintenance free, but instruments still need to be recalibrated, replaced and maintained.  These instruments are outdoors and can get pelted by hail, get dirty, or get spiderwebs or hornets nests, etc.  You will find the densest network of measurements in areas where lots of people live.  Sparsely populated areas, areas with complex terrain, will have less measurements and this means they will experience greater errors in forecasting.  In addition to the complex wind flow that occurs in mountainous areas leading to a large variability in conditions, there are far few weather observing stations.  If you live in such a region you are likely less than impressed by your local forecaster.

The Answer is Blowing in the Wind

The prevailing wind direction in mid-latitudes (where most of the U.S. resides) is from west to east.  Thus even being downstream of areas that have sparse observing stations also are more poorly forecasted.  The best way to know what weather you are about to get is to have good measurements upstream of your location.

Communication Breakdown

Finally, there are also communication issues.  The National Weather Service has put a lot of effort into this area, to think how to better communicate and disseminate weather information.  For instance if we have a particular graphic showing probabilities for where the eyewall for a hurricane is going to hit, is that graphic communicating what it needs to the person who needs to see it, whether that’s an emergency response worker or the average person?

Image result for hurricane eyewall probability map
                          How well do you understand what this graphic is showing you?

In this day and age of instant media and social media, it should in some ways make communication easier, but what I’ve noticed that it’s not always clear whether people are paying attention to the most current information, if they’re getting their information from a good source, and even if it seems whether or not they are aware what location a particular forecast is for and may think a forecast was bad even if it wasn’t for where they live.  As I mentioned at the beginning there is also a lot of clickbait and alarmist language being used.  Things like “bomb cyclone” and other colorful adjectives.  At the same time there has been criticism that the normal scientific tendency to temper their language in communicating important information may make people pay less attention to situations they should pay attention to.  Undoubtedly there are going to be consequences of both extremes.  Overuse of strong language, especially when conditions end up not being that extreme can numb the public to more dire warnings. Trying to find the best way to get people to understand, and pay attention is difficult, but this is a challenge the weather community takes seriously.  In the end, there probably is no perfect way to communicate, and it is up to the consumer of the information to educate themselves as well as to what this weather stuff is all about.

Hopefully this little piece helped explain a few things.  If you have any other questions, let me know.  I’ll add to this so this remains a fixed guide to helping people understanding the challenges in forecasting and why we might have misconceptions about forecasting accuracy.

 

 

It’s the Thought That Counts

It has been discussed by many that our brains are wired on an evolutionary scale, and that the rapid change of society through technological advances has outpaced us, leaving us with many disconnects between what we see every day and what we can actually handle.  In many ways, we might be happier if we lived in small tribes and were closely surrounded by wilderness, instead of surrounded by brick and cement, drive vehicles and get visual stimuli from computer or television screens.   One aspect of this disconnect, that I find quite intriguing, and I think is central to our ability to understand the world we find ourselves in, is what I call and order of magnitude problem.

Think about early man in those hunter gatherer days.  Counting is a base cognitive skill, important for our survival.  But what is that we might count?  You might count the amount of fruit gathered on any particular day, the number of children, or people.  Such numbers might get you into the 100s.  You might count seasonal cycles.  If you were lucky maybe you had 80 of those to count.  You might count lunar cycles.  Getting you to about 1040.  Even this would require some note making, because this is counting over time, and surely you would not sit there and count something that high.  Such cycles of time were the only things worth keeping track of.  We had no need to measure time beyond that.  No need for small units of time such as a second.  It might make sense to come up with some unit of measurement for distance. Something comparable to arm lengths or hand widths…something we might use to size an animal, measure height of people or spears.  When it came to traveling, you might then simply use something like phases of the moon, or number of diurnal cycles.  Once again such counting would leave numbers small.  Occasionally you might find yourself thinking about numbers in terms of fractions.  Maybe something like half a day, or a quarter of an armlength.  For things very small, you probably would no longer use armlength as your standard, but perhaps finger width.  Such techniques are ones that we still use today.

The reality is that if you think about numbers, you probably won’t get very far.  Now do a little exercise for me.  If you think of the number 1000.  How do you think about it, to picture a 1000 of something?  You might think what a $1000 can buy, but money is a fiction that represents a quantity of stuff you can buy which varies depending on what stuff your buying.  If you wanted to actually count, what would you think about.  Maybe 1000 people in a room.  You might have a sense for how big a group that is.  Chances are you won’t get it exactly.  Go down to a 100 and your chances of picturing 100 things gets better.  Now do 10 of something.  Pretty easy.  Now do 1.  Even easier.  Let’s go down another order of magnitude.  Try to think of something that is 0.1.  Here as we move down an order of magnitude we can no longer count whole things.  So think of 1/10th of a person probably gets a bit graphic, so what are you thinking of to imagine 0.1?  For that you now have to think of some standard.  Maybe a mile, an inch, a meter?  Depending on what you choose, you can do okay.  Now try 1/100th.  Again with the right starting point you might do okay, but even dividing by 100 can be hard for someone without a formal education and once we get to 1/1000th our ability to guess at the meaning of that fraction is severely reduced regardless of our starting point.  So if you are keeping track this puts the human mind, on a good day our brains are capable of somewhat accurately sorting out 5 orders of magnitude (10-2 – 103).  However, if we look at the scale of the universe in size we span 52 orders of magnitude from the plank length to the size of the observable universe (please see this very cool interactive graphic that allows you to explore the different spatial scales of the universe).  In terms of time, our quantum clocks can measure up to 1  ten billionth of a second (10-10) .  Meanwhile we know the universe has been around for about 14 billion years (1015 seconds).  If you don’t have trouble digesting such numbers you are a super genius, because everybody should.  Those are just the extremes, but unless you are within that 5 orders of magnitude range I discussed earlier, it makes little difference.  And this is also important because it means that a million miles, might as well be a billion miles in our head.  However, the difference between those two numbers is meaningful.  In science, to consider two numbers like that the same would be to make a grievous error on the order of 100,000%.

Scientists, through years of working with the numbers that shape our world are often better at dealing with these things, but even scientists tend to use conventions to make numbers easier to manage.  There is a reason why you don’t measure the distance from New York City to Boston in inches.  We have developed different units of measurement for distance.  In the old English system we have inches, feet, yards and miles.  In metric, we have prefixes that span numerous orders of magnitude so that we don’t have to always report distance in meters.  For objects in space in our solar system we might use astronomical units to keep those distances in more manageable numbers.  For things outside our solar system, light years.

                  Image of radar reflectivity.

Whatever we measure in science can change over large ranges and change at massively different rates.  Change is rarely linear, but very often exponential.  As a result, we might find ourselves dealing with quantities which very over several orders of magnitude.  In my field a good example for this is radar reflectivity.  You may not be familiar with it, but you’ve certainly seen radar images if you’ve paid attention to the weather.  Higher reflectivities indicate bigger drops and faster rain rates.  Lower reflectivities represent light rain or drizzle.  The difference in size between a drizzle drop and a basic rain drops is no more than a factor of 10, but the reflectivities span over 10- 1,000,000.  Thus, meteorologists convert those reflectivity values using decibels.  The decibel system was initially used for sound given the large range of frequency for sound waves, but now is a common tool for expressing values that vary over several orders of magnitude by taking the logarithm (base 10) of the value. This reduces the number to its order of magnitude.  For example, instead of 106 if I take the logarithm with base 10 of that number I get 6.  And 6 is much easier to wrap our heads around than 1,000,000.  I know I’ve gotten kind of technical here with this example, but the point is that nature, as we’re discovering, does not conform to the numbers our brains had to deal with when we evolved.  And most scientists, while they might have some understanding of the microscopic or macroscopic numbers and the wide ranges of values science employs, to objectively analyze and come to some meaningful conclusions we very often have to be able to visually see those numbers between about 0.01 and 1000.

You might say that such numbers make little difference to most of us unless we are in science, but let’s talk about where our everyday lives might be impacted.  First let’s start with the population of the world.  There are 7 billion people.  Try to wrap your head around that number.   Is your soul mate really just one in a billion?  Could such a large group of people create an environmental disaster? How many bodies could certain countries throw at you in a war? About 700 million, globally, live in abject poverty.  Do the numbers seem so voluminous that it’s easier to ignore human suffering, or make you feel defeated before you try?

What about some of the more important educational and scientific controversies that still exist today? Evolution has been happening for several billion years, but many would like to believe that we’ve been around for only 6000 years.  Religious dogma aside, isn’t it possible that part of the reason that some people resist what science clearly demonstrates is because we are talking about a length of time that few can relate to?  The vastness of time threatens to humble us all as blips in a universe far older than we can fathom. And its size and origin similarly attacks our human conceit at being the grandest and cleverest design in a creator’s eye.

                                         The Geologic Time Scale

Vast amounts of people also create vast sums of money.  Billionaires have almost unimaginable wealth that people still commonly believe that can obtain too.  Politicians and media constantly throw large dollar values in our faces to intimidate us.  When one wants to high light wasteful spending we can put point to something costing 100’s of millions of dollars and we shudder at such an amount being wasted.  Forgetting that with 100 million taxpayers, something in the 100’s of millions is costing us a handful of dollars a year.  I have seen the tactic used frequently.  Once again we might on some level realizes that a 100 million, 10 billion, and a trillion dollars are different, but they are all unimaginably large sums of money that in the battle for what’s important and what’s not, they can all be seen as being on equal footing. The idea that public television and radio need to be cut for austerity is quite simply a joke when compared to a 10% increase in defense spending if anybody thinks that’s going to balance the budget.

One might argue that the microscopic matters very little (no pun intended), but I do think an appreciation for that scale is valuable, if for no other reason helping us appreciation the vast variation of scales that make up our known universe.  Scientists often take very small numbers that might exist for pollutants or toxicity in foods or water, and change the unites of those numbers so that they are bigger.  I understand why, because of course we don’t want to underwhelm in those situations, but maybe it’s also a problem that we continue to cater to this limited range of numbers that our minds most easily manage.   It’s probably best to start incrementally, and perhaps a good example of how we can begin is with time.  John Zande over at his blog, The Superstitious Naked Ape, offers up a good first step towards our lack of comfortability with numbers outside of our “sweet spot”.  The start of our counting of years begins with the birth of Christ, but this is a religious and faith based reason to start the counting of the years.  Why not use Thai’s bone which is our earliest evidence of careful astronomical observations of the sun and moon over a 3 ½ year period.   Instead of the year 2017, it would instead be 15,017.

                                                                          Thai’s bone

It might seem like an arbitrary difference, but I think it would give us a better feel for the vastness of time, and a better appreciation for the numbers that shape the universe we’ve come to know.  Since there seems to be little stopping the advance of science in technology, perhaps we better find more ways to help these brains, made for a different time, catch up.