How Weather Models Work

we used to try and predict the weather. UH. In the first episode in this series, we concentrated mainly on the science of weather itself, and that's important to understand because you begin to pick up on how complicated a system whether actually is, especially when you start expanding out from a small region to a larger region to a global region and you see how interdependent all of these different regions are on each other, the brain starts to

swim a bit. In Part two, we looked at the various sensors and tools used to capture information about what is going on with the weather. These are the tools that meteorologists use in order to feed that information into their various weather models. So UH, Today's episode is going to focus on those weather models. These are based on our understanding of the behavior of weather under different conditions, and it's what gives us the confidence to make predictions

of what will happen next. Now that being said, predictions, as we all know, are not guarantees. I'm sure there are many of you who have walked out of your homes with confidence, dressed in your best clothing, only to be plagued by an unexpected downpour at an inopportune time, turning into Charlie Brown with that one rain cloud just

directly overhead. Or you might be one of those people that have convinced him or herself that if you have an umbrella in your hand, it virtually guarantees that not a single drop of rain will fall, that you, by virtue of holding the umbrella, have prevented rain from happening.

As it turns out, predicting the weather is really hard for a lot of reasons, though the biggest one is that weather is just an incredibly complicated system affected by hundreds of variables, and those variables may have a lesser or greater effect in different situations. That means there's a lot of potential outcomes for any given scenario, and until we have a really comprehensive understanding of what's going on at all times in our atmosphere. Weather predictions will be

based primarily on statistical probabilities, not certainties. Before we really had a handle on all of those variables, and to be honest, we don't completely have a full handle on them right now. We based weather predictions off of empirical

rules that we formed observation. So, in other words, generally speaking, if you woke up and looked outside the window and you thought, hey, it looks like it might rain today, because a couple of weeks ago I looked out the window and it looked just like this and it rained that day, well that's about as complex as it got. I mean, you might have a weather map, like a literal map, and you have some things you've written down

on it. You know that to the west of you there's a low pressure system, so you might start using that as a guide for what could end up happening. But it wasn't a very precise science. It wasn't what people were calling a rational approach. It was an empirical approach, and over time we began to understand that weather is

dictated by a host of very complex variables. So meteorologists are gathering all of this data about air pressure, temperature, windspeed, weather patterns that are nearby, and tons of other factors. These things are changing quickly and constantly, meaning it's important to look over those observations regularly and adjust projections. And if you're not lucky, you may only have a few observation stations placed in strategic locations within your particular region,

which gives you a limited on resolution. So weather forecasting is a lot like your displays or your televisions resolutions. Very important resolution is all about how many points of data do you have within that given area, how representative

are those observation stations. If you have a single observation station for several square miles, well that's not going to give you very good resolution, right You're going to have a very specific idea of what's going on at one point within that area, but everything further out from there it's going to be some variation of the information you've

pulled down. So if you want high resolution, you have to have lots of observation stations throughout that same area, and you collectively are able to determine what's happening by looking at all of them, but obviously that adds a lot more information to your calculations. In an ideal world, you have all areas densely packed with observation stations, giving you amazing consistent resolution and the processing power necessary to take all that raw data and crunch it to produce

reliable weather forecasts at any given moment. But we just aren't there yet. So weather models, what's the story on those? Well, it helps to look back on the birth of meteorology and weather forecasting in the form of numerical weather prediction. That's really what we get down to when we start talking about weather models. And to do this we actually have to backtrack a little bit. I know, we talked a lot about the various tools in the last episode and we got pretty up to date, but we're gonna

have to go back to talk about weather models specifically. Now. In the nineteenth century, so this is the eighteen hundreds, physicists were beginning to suss out the law of thermodynamics. These are the basic laws of energy that we UH that everything is is UH has to obey, at least

everything on the macro scale has to obey. And it was relatively easy to put hypotheses to test with modest experiments, right, Like, you could do tests about fluid dynamics with small contained systems and you could limit the variables and make lots of observations. That was pretty easy, relatively speaking, But it was much more challenging to step back, and I mean way way back and see how those same laws applied

to something as massive as our atmosphere. If you want to think about another way, it's one thing to look at a maze that has rats running in it. It's another thing to try and figure out what the maze is like. When you are inside the maze and you can only see a small part of it, How do you know what the entire layout of the mazes from that perspective? So, in other words, the perspective of the rats we are inside the actual environment that we want

to describe. That makes it way more difficult for us to uh isolate and weigh every single variable in the system. Well, in nineteen o one, there was a professor named Cleveland Abbey. He published a piece in a journal called Monthly Weather Review. Now I was really sad to discover that this wasn't actually a list of reviews for actual weather like Weather

Today was pretty good. I give it three stars. No it It actually was a scholarly journal on the subject of meteorology, and Abbey's piece had the title The Physical Basis of Long Range Weather Forecasts. And in that piece, Abbey pointed out that forecasts of the time were based on experience rather than any real knowledge of how weather works, and that the physical theories explained the development of whether we're either superficial or non existent. So it goes back

to that example I gave earlier. Weather forecasting was based on people's experience with weather, but not knowing how the weather was actually working. So you might say, well, I think that it may snow tomorrow based upon what the conditions are right now, and the fact that I remember a day that was like this where it snowed the

next day. But that's not a very scientific approach ultimately speaking, and it doesn't have It is not based upon understanding the factors that lead to things like a snowstorm, and so Abby was arguing that in order to have real weather forecasting prowess, we would have to gain that understanding of the underlying factors of weather. Abby asserted that we just had to understand the laws of mechanics and heat of the atmosphere. Only then, he posited, could we use

the information to start making more accurate forecasts. And his peace would go on to outline what he saw as the necessary steps to get there, including a thorough investigation of the behaviors of the atmosphere, and he said that the science of meteorology is quote essentially the application of hydrodynamics and thermodynamics in the atmosphere end quote, So he was calling for the establishment of a new area of science,

specifically within meteorology, something that that would require people to dedicate a lot of time to try and understand this complex system that is our atmosphere. Then you have a Norwegian scientist, Vilhelm Biekness. He was born in Christiania, Norway, in eighteen sixty two, and as a young man he worked with a notable physicist, Heinrich Hurts Hurts I mentioned in our episodes on the history of electricity, you know

the Hurts. He went on to he being Bakness went on to teach applied mechanic and mathematical physics at the University of Stockholm, where he sussed out some theorems that helped him create a synthesis of hydrodynamics and thermodynamics for large scale atmospheric motions, the very thing that Abby was

calling for. Bierk Nous was the one to to develop a very comprehensive model, the first really comprehensive model for that, and this led to the development of air mass theory, one of the principal ideas upon which we base weather forecasting. Now in n Berk Now published a work titled on the Dynamics of the Circular Vortex with Applications to the atmosphere and to atmospheric vortex and wave motion. And it's a real page turner, guys. This is a pretty dense

piece of of scholarly work. It's considered one of the most important scholarly works in the field of meteorology, and it was the basis of our understanding of general weather pattern behaviors and why they take on the forms the way they do. In other words, it was pretty much what Professor Abbey was saying was necessary before we took

a rationally scientific approach to forecasting the weather. Now, what Wilhelm illustrated was that the atmosphere and thus weather does can be described in math through fluid dynamics factors like temperature impact those behaviors, so you have to take that into account. And as things change, there's a sort of ripple effect. If you change one part of the the system, that ripples out and affects the rest of the system

in different ways depending upon other factors. And nothing in the atmosphere is remaining completely unchanged, and as each element shifts or cools down, or heats up, or the density changes or whatever it may be, it affects other parts. So the math gets pretty challenging pretty fast. He further went on to create a two step process for rational forecast nesting of whether now. The first step was diagnostic, which is, in other words, using observations to determine what

is the present state of the atmosphere. This is where you take all those readings and you say what is going on right now? That's the diagnostic step, and it's absolutely necessary before you can do anything else. You can't say what's going to happen next until you have an

understanding of what is happening right now. The second step was prognostic, which meant that you would use that information from the diagnostic step and project outwards and say, all right, well, based upon what we know is happening right now, what is going to happen twelve hours from now, or a day from now or two days from now. And you would have to use the information you had gathered in the diagnostic step, combined with our knowledge of the laws

of motion for atmospheric masses, to predict what would happen next. Now, if you could strip everything away and just look at the math, you'd be looking at a collection of what are called partial differential equations. The mathematicians out there know exactly what I'm talking about. These are equations that deal with rates of change with respect to continuous variables. So you're not just talking about variables like temperature, pressure, and velocity.

You are talking about those, but you're also talking about the rate of change of those variables. How quickly is the temperature changing, how quickly is the pressure changing, etcetera. Now, the way you frame and solve these equations defines your weather model. Different weather models place different emphasis on these variables and the rates of change. All of this boils down to a computer program ultimately that solves these equations

as you have directed. So one model might approximate different equations one way and another model does so in a different way, and thus you're going to get two different forecasts. Consulting these two different models, they might resemble one another, but they're taking different pathways to get to their destination, so sometimes they might be very different from one another. And it's all because of the way you have told the program to prioritize the various processes, which ones you know,

which factors have the most weight and under what circumstances. Now, this doesn't necessarily mean one model is by its nature superior to the other. Some models are for specific regions in the world, and those regions, due to geography and general atmospheric motions, may require more importance to be placed on certain sets of variables rather than others. Now Berkness identified seven variables he saw as critical for accurate weather forecasting.

That includes pressure, temperature, density, humidity, and then three different components of velocity. He also identified seven equations, three therma hydro dynamic equations I should say three hydro dynamic equation, emotion, the continuity equation, the equation of state, and equations expressing the first two laws of thermo dynamics. Now, keep in mind that the atmosphere is three dimensional. You have to essentially consider any region within that area a three dimensional grid.

So your grid has an x, y, and z axis, and the events within one part of that grid can affect other parts of it, particularly atmospheric motion. And you need to figure out how to take partial derivatives which computers can't really handle, and then turn them into approximate partial derivatives that computers can handle instead. This approximation adds in a bit of imprecision by its very nature. But then's the brakes. And speaking of brakes, let's take a

quick one right now to thank our sponsor. So getting back into forming weather models. There's a meteorologist named Lewis Fry Richardson who was influenced by Berkness, and he did his best to tackle the problem of numerical weather forecasting, but he stated that the sheer amount of computation was impractical for the time. This would be in the early twentieth century, first couple of decades of the nineteen hundreds.

He did say, quote, perhaps someday in the dim future, it will be possible to advance the computations faster than the weather advances. But that is a dream end quote. So he's saying, by the time I'm able to work out the math, whatever the weather was gonna be has already happened. I'm predicting what happened hours ago. Uh. And

that was a real problem. Was just that again, you had these very complex equations with lots of points of data and lots of variables that you had to solve for, and by the time you would be done with all the calculations, the time had passed. So he was saying, there is a need for some sort of engine that can do computations faster than what humans can do. So the math was just too complex to complete without the

use of that computational engine. One person determined to help design such an engine was John von Neumann, who was a mathematician who made numerous contributions to the sciences, and he realized that some of the more advanced problems in hydrodynamics and weather forecasting would benefit from a powerful automatic computational machine. He worked on a project at Princeton at the Institute for Advanced Studies and it would become known

as the Electronic Computer Project. Meanwhile, over at the University of Pennsylvania's More School of Electrical Engineering, you had J. G. Brainerd who was heading up a project, and J. Presspur Eckert and John W. Malchley who were working on what was called the Electronic Numerical Integrator and Computer or NIAC for short, and computer nerds out there, I consider myself one of them will kind of bristle at the sound of ENNIAC. They might prick up their ears and say, oh,

I know, I've heard of ENNIAC. ENIAC being one of those early early computers in the dawn of the computing age. Well brain Nerd, the man who was heading this project, invited von Neumann over to the University of Pennsylvania to check out ENIAC, and von Neuman would end up having these very deep discussions with the team, and those discussions would help inform the design of the successor to ENIAC, and these early computers would become some of the first

capable of tackling those difficult computational problems. I was mentioning a second ago, and that brings us to the concept of the weather model. Now, your weather model is an advanced computer program that runs all of these sorts of equations and then calculates outcomes. So, in other words, it generates your weather forecast based upon those points of data. Many of these weather models have been written in four chan for Tran, I should say, largely because that's how

it's been done for decades. So if you've ever chatted with somebody and you're saying, why are we doing it this way, and they say, it's because it's how we've always done it, that's sort of the case with Fortran in weather models. Uh, that's one of the reasons. But it's also a very useful language that has evolved over time. It's not like it was developed and then forgotten about. It has received a lot of I hesitate to use the word love, but development over the years. Now, this

four trend program is compiled into machine language. That's the language that computers understand, and we'll talk more about that in the Programming Languages episode that will be coming up soon. So keeping the year out for those episodes, they should

be following this one shortly. The program, the weather model takes all this information, the data fed to it from multiple sources, all those observation stations, and all those equations based off of fluid dynamics and thermodynamics, and steps through in time to simulate what will happen next. So the computer is actively trying to simulate the behavior of weather patterns based upon our understanding of hydrodynamics, thermodynamics, and all

of these variables. So it's it's actively simulating the outcomes. Now you're getting these simulations in the form of numeric answers. It's not like you're looking at a graphic representation of weather. You're not, you know, looking at your computer and you see a massive storm is roiling across the screen. I'm

pretty sure that's how Hollywood would do it. But we're talking more about lots of numbers, so not as sexy as say, watching Twister on Netflix and you're saying, that's what the way, what's the what the weather is going to be? Uh, that's not exactly the case. Sadly, Maybe one day we'll get there, but not not right now now. Some simulations can project out for multiple days, and some

are more immediate. Some look at short range forecast, some do mid range and long range as well, and the highest amount of accuracy typically is within the next several hours.

And then, of course the further out you go from the moment you gathered all that data and did your diagnostic stuff, the more you are likely to diverge from reality for your forecast, more uncertainty enters into the picture because it's hard to predict how all of those different variables are going to uh what what their state will be at any given point in the future. And the

further out you go, the more uncertain you're going to be. Typically, So let's make up a hypothetical situation to kind of explain what I'm talking about here. Let's say I'm in the north of wester Ross and I know winter is coming.

My weather forecast model is very much focused on atmospheric movements from beyond the wall and less concerned with other variables like maybe humidity or I don't know, dragons, because humidity and dragons don't play such a large role in the weather patterns of my region, right, I mean, I'm one of the Starks in this hypothesis. The model I have created is as accurate a representation of how patterns emerge and behave in my region as I can get

my hands on. So that's what I rely upon. But let's say you in your fancy pants King's landing house are concerned with trade winds coming in from out over the sea, because that's a large influencer of the weather in your area. So your weather model takes that into account and gives it greater weight than some of the variables that I'm concerned with. And this is so that you can create accurate forecasts for your area. Your model and my model aren't equivalent. Your model would not work

as well in my region, and vice versa. And neither model is comp inhensive, which means neither model covers all of wester Roast. That's very much regional. Now in the real world spoiler alert, Game of Thrones isn't real. We sometimes find that our computer models are mostly good, but not perfect. Some may, under certain conditions under or over

estimate the temperature for example. Now, this could happen for lots of different reasons, such as, you might have a region that's close to the ocean, and the ocean could affect the temperature in ways that the model is not quite capable of accounting for. So you might experience more windy conditions than other areas, and the wind may affect

temperatures in ways that the model can anticipate. The wave dynamics could affect whether in ways that the model can't anticipate, so you still have meteorologists who are dealing with this, and they're fudging the numbers a bit once they've been processed, because you learn over time how well your model does versus reality. So you can look at the results of your model, what does the predictions say, and then you can compare that against the actual results that you get

just by waiting around. Right, you wait around and you see what actually happens. You compare that to the forecast that your model gave, and you start to look and see if there's any any adjustment that needs to be made, or in some cases you may just say, well, this model is frequently about two degrees warmer than what really happens,

so we're going to build in an adjustment. We will automatically no to decrease the temperature forecast by two degrees from this model, and that we are more likely to

hit on what the actual temperature will be. This happens all the time with lots of computer models, not just for temperature, but for other variables as well, and really we should expect this to continue to happen because we cannot have a perfect understanding of how everything is going to be and builded into a computer, not yet, possibly not ever. It is so complex and so dependent upon so many different variables. But what we can do is

we can correct for those known problems. If we know that there is an issue and it's not likely to cause a ripple effect, which I'll talk about a little bit later in this episode, then we can just correct for it at the end and say, all right, let's bump up or bump down the temperature by a couple of degrees based upon our knowledge of how this model performs against what really happens. It's kind of interesting because it really nails home how computers are all about precision

and replication. They don't tend to give you vague guesses. They can create different answers that have different probably probabilities of being correct, and then choose whichever one is the most likely to be correct. But they are about making these precise uh outputs, and we as humans are the ones who have to add extra levels of interpretation on top of that, which means that there's still a human being associated with this process. It's kind of what you

have to do with an old scale. If you had an old like weight scale and it was off of its calibration, you couldn't quite get it to reset at zero. So let's say you notice that your scales giving a reading that's always two pounds less than what it should be. You know that when you weigh something, you need to add two pounds to your scales reading whenever you weigh something. Meteorologists will often do the same thing, and sometimes they'll do it to just a very specific region within a

computer models area of coverage. They know that the model has a history of under or overestimating things, and so they just correct for it. Now, for that reason, we still have human beings involved in meteorology and all these different phases. Meteorologists use their training and expertise to interpret the output from weather models, and they learned the quirks of the models, even as new versions of those models come out to correct for inaccuracies or to increase resolution

or frequency. So remember whether forecasting depends upon the quality of the weather model, the accuracy of the information being fed into the model, and the frequency with which that information comes in, and the density of the observations within that region. All of these things will affect the accuracy of the ultimate weather forecast to come out of that computer model. If you have the best model in the world, it's still not going to give you a very accurate prediction.

If either you don't have enough observation stations so your resolution is low, you aren't consulting your observation stations frequently enough, so you are relying on older information, or the information you're feeding into your model is somehow inaccurate. Let's say that you have some sensors that aren't working properly and are giving you, uh the wrong reading for some element here,

whether it's air press or temperature, whatever it might be. Well, if that information gets fed into your computer model, then you would expect that the outcome is not going to be accurate because it wasn't accurate information going in, or, as some people say, garbage in garbage out. Your outcomes are only going to be as good as your data, as well as the fact that you have to worry about the the quality of your model itself. Now, if this sounds like it's a ton of processing, it is.

Traditionally one of the big applications we have for supercomputers is for weather models. So whenever you hear about supercomputers and the massive amounts of processing power they have. Often these computers are being put towards the task of simulating weather, taking these weather models and trying to get more and more accurate simulations of what is going to happen. ANIAC

itself was used to generate weather forecasts. But even though NIAC was a big jump forward on just working out the equations by hand, it was still limited and the weather model wasn't much more than a barotropic equation. A barotropic equation is a fluid dynamics problem in which density is a function of pressure only. Even this limited interpretation of the factors that affect weather was still a big

leap forward. However, any act success led to the development of new models, including multi level models, and one such model was the product of several scientists work in the in the in the wake of a massive storm system that took place on Thanksgiving Day in nineteen fifty. So this big storm ended up being a great opportunity for the scientists who were trying to make a weather model, and the model they developed seemed to simulate actual events accurately.

They were very excited. This multi level model appeared to be much more accurate than the barotropic model that had been used as it turned out their model was only really accurate for that one set of circumstances. They found that as they ran more simulations, it was not giving accurate forecasts, at least not in every situation. So it turned out that that weather model was really great for one set of circumstances, but it didn't handle other ones

nearly as well. It didn't get nearly as accurate a result, and the barotropic model actually was superior. The older model that had been running on any act was superior to the multi level model, at least in some situations. So throughout the nineteen fifties, meteorologists were mostly relying on this older barotropic model because it was more accurate more often

than multi level models that had been proposed. Starting in ninety eight, multi level models began to gain more acceptance as they tuned into the right waitings for the various variables and weather forecasting, and from that point forward we saw more varieties of weather models arise, each with its own pros and cons, And what followed were numerous symposes about computational models and the machines that would be needed to crunch the numbers in a reasonable amount of time.

And it gets super duper technical. Now today we have many models, most of them covering specific regions. Creating a global weather model is an enormous task, and not just to combine our understanding of weather behavior from around the globe, but also to find a computer capable of processing such an enormous amount of data regularly enough to give us an accurate weather forecast at any given time for any given location. But here's some of the models that we

use today. One of the big ones is the European Center for Medium Range Weather Forecasts or e c MWF. They provide one of the more important models in the world. The Journal of Computational Physics describes the model as quote a spectral primitive equation model with a semi Lagrangian, semi implicit time scheme and a comprehensive treatment of physical processes.

I'm pretty sure that means it can summon cthulhu. In addition, this model is coupled with an ocean wave model, and the basis is the Integrated Forecast System or i f S. The model runs on high performance supercomputers capable of performing several terra flops of calculations, and just a reminder, a flop is a floating point operations per second. So generally speaking, the number of flops the computer can perform gives you an idea of its processing power or speed, and terra

flops means a lot. All Right, we're in the home stretch for meteorology. But before we jump into that final section,

let's take another quick break to thank our sponsor. Alright, I just talked about the weather model over the big one over in Europe, and keep in mind there are dozens of weather models, but over here in the good old US of A, you've got a big one with the National Centers for Environmental Prediction or in c e P as it is known, and it has a globally gridded set of data about the state of the Earth's atmosphere.

And there are tons of other models too. As I was just saying, some of them are more localized than others. Some of them are capable of much higher resolution because they consult more observation systems with respect the area covered by the model, and those grids are important. You want um smaller grids, You want the the sides of each of the grids. To keep in mind, this is three dimensional. It's not just um land area, but elevation as well.

You want smaller grids because that increases that resolution, right, because each grid represents an area where you understand what is going on inside of that area. The smaller you make the grids, the higher the resolution is. This is again a lot like your television or computer display. If you make a picture out of just a few pixels, it will be blocky. It has very low resolution. Like think about eight bit graphics back in the day. Every all the characters on video games were made up of

these blocks. They all had very jagged edges. It was not a high resolution. If you use more pixels to make your photo, that improves the resolution to a point. Anyway, there gets to a point where we can't really perceive it anymore, but you certainly can perceive it at those early stages. So the smaller, the smaller, and more numerous the pixels, the higher the resolution is and the higher quality you get of an image up to a certain point.

The same is true for weather models. So if you have a grid with small squares, such as on the order of a few kilometers per side, you would have a high resolution, and those grid points can have a single value per atmospheric variable per observation. So another words, you get a value for temperature, a value for wind direction, a value for wind speed, a value for atmospheric density, et cetera. Uh, all of that tends to be consulted

about once an hour with most of these models. So once an hour you pull all that observational data for every square or cube if you prefer, within that grid. So you pull all of the information for all of the grids within that area or all the cubes within that grid, and you crunch the numbers from all of that to see how weather will progress from that moment forward. Now, if one variable from one grid is way off, it can cause bigger errors and forecasts further down the line.

That garbage in, garbage out thing I was talking about, and this is the infamous butterfly effect. The butterfly effect refers to a small effect that can have much larger consequences further on in time, and you've probably heard about

the effect before. The classic example is that you have a butterfly flapping its wings in South America and the force from the breeze generated from the flapping ends up contributing to a system that eventually grows in power and ultimately culminates in a massive typhoon in Asia, for example. Now that's just a thought experiment, obviously, but with weather

models something similar can happen. If you have a large grid and each square in the grid is representing a relatively small area, and one of those areas within that grid produces data that doesn't reflect real conditions, your forecasts will be affected by this. Now, depending upon the weight of the variable in question, it could make the entire forecast inaccurate after a certain amount of time. Generally speaking, the further out you go in time, the more you

have to depend upon numerical forecast models. A short term forecast might not require a full numerical analysis. It could depend and more on what's going on right now and the likelihood of how weather will change over the next few hours, so you can refer more on experience in those cases. Beyond that, however, you'll need some more numerical analysis to get a better than UH to do better

than just giving a wild guess. So even so, you might run the data through a couple of different models to look at potential forecasts, and from that point and experience, meteorologists might look over the data to see which predictions appeared to be the most realistic. Sometimes computer models get stuff wrong. They might predict an extremely unlikely outcome. Other

models might have a very different forecast. The meteorologist has to determine which of these outcomes best represents what is likely to actually happen, based upon how well they seem to handle the current weather situations. So you might look at a computer model and say, well, how is it handling what's going on right now. If it's doing that well, then we can at least lay some assumptions that the any predictions coming from this computer model are going to

at least be semi accurate. If it's not handling it well, then we may need to consult a different weather model for this particular forecast. Now, if observation data is affected, as in, if there are problems with sensors, then the information you will get out of the models will not be dependable. The high resolution can help smooth this out. So if you're getting one sensor with erroneous data, but you've got lots of other sensors in the area, you

could possibly smooth that out. You might say, well, this is clearly an anomaly. If most of your observation stations are reporting that the temperature is about seventy degrees fahrenheit, and one of them is saying it's degrees fahrenheit. You could say, well, this one clearly there's an anomaly. Maybe something is going on in that area. Maybe it's close to a fire or something not too close but fairly close.

Because pretty off the track for everything else, you can perhaps week your model to ignore that particular sensor so that way it doesn't affect the rest of your forecast

and throw things into disarray. But if you have just very few observation stations, then the loss of even one could be enough to throw your forecast off anyway, So you may end up having your forecast thrown off either because a sensor is giving you incorrect information or because without that sensor you don't have enough information to build

a reliable prediction. So it's a delicate thing. Meteorologists also have to keep up with what is actually happening at any given time, and this seems pretty evident, but it's important to either verify that a computer model is in fact forecasting, whether accurately or if it's off track. And we learn more from our mistakes than we do from successes.

Just like in that example of that first multi level model, the scientist thought that they had really hit on something because their models seemed to handle the conditions of that Thanksgiving Day storm and give very realistic outcomes, But as it turned out, it wasn't good at handling other situations. If we succeed, we might think that what we've done is perfect, that it's worked, but it may turn out that that's not the case. When we fail, we realize, oh,

something is not quite right here. We need to figure out what that is and how to correct for it. This is true, by the way, in all areas of science, we learn more from our failures than we do from our successes. As computers get more powerful, the simulations can take in more data and in theory, will generate more accurate forecasts. The addition of some other elements, such as deep learning algorithms that can assign probabilities to outcomes, might

also help. These probabilistic models assigned statistical probabilities to various outcomes, letting meteorologists or even an artificially intelligent program determine which one most likely represents what is really going to happen. But some challenges will remain. I'd like to end this episode by quoting from the second edition of Introduction to Tropical Meteorology to illustrate just how dann complex this is, and I quote from chapter nine of said textbook, Tropical

weather is difficult to forecast. Mid latitude weather is dominated by synoptic systems moving in the westerly's, which formed the basis for the weather analysis methods developed in the nineteenth and twentieth centuries. In the mid latitudes, baro clinic instability results from air masses with contrasting temperature and density. Their energy is concentrated in extra tropical cyclones that can be

tracked fairly easily. By comparison, the tropics have a relatively homogeneous air mass and fairly uniform distribution of surface temperature and pressure. Therefore, local and mesoscale effects are more dominant than synoptic influences, except for tropical cyclones. For example, service temperature and pressure can change quickly with convection and sea breezes. So you see, we cannot We cannot apply what works for one area across all areas of Earth because it

just doesn't work that way. There are some areas where things that are of a major impact on weather patterns are almost non factors, and so until we develop very specific particular ways of observing, measuring, and predicting weather across all of Earth, and then synthesizing that so that we can give global weather forecasts that can then narrow down to the hyperlocal area, we're going to continue to have.

This uncertainty is a phenomenal area of science. It is remarkable to see technology applied to that area of sience in such a way that is easy to illustrate the

importance of pushing back the barriers of computational power. It's why a lot of people look at stuff like Moore's lawns said say, it's really important that we keep that going, even though it's getting harder and harder to keep Moore's law relevant, because we do have these needs for heavy computational loads that have real effects and and and powerful

outcomes for billions of people on this planet. I mean, accurate weather forecasts have the the potential to affect or to save people from calamity, or to help businesses determine when and where they're going to transport goods to get to people more effectively, thus reducing lots of things like environmental impact and the economic impact. You start to see how intrinsic our weather is to everything else we do.

It's more than just small talk, and it's more than just whether or not you need to grab your umbrella before you leave the house. And I hope that these episodes were interesting to you. I find meteorology to be absolutely fascinating and I would love to learn more about it. And I love chatting with meteorologists because even though they can get super technical and they can really talk about some heavy duty math that I can sometimes kind of

sort of follow. There dedication to understanding something so complex I find inspiring. Now, our next episode will be about the history of programming languages. It's likely going to be a two part episode, and I'm going to study the origin and evolution of computer languages. But if you have suggestions for future episodes of tech Stuff, please share them with me. You can send me an email. The address is tech Stuff at how stuff works dot com, or you can drop me a line on Facebook or Twitter.

The handle of both of those is tech Stuff hs W. You can always drop in and watch me record these episodes live that you can find at Twitch dot tv, slash tech Stuff. I record on Wednesdays and Friday's today. If you had joined us in the studio, you would see that I'm in a brand new studio, or at least a different one from the one I'm usually in, and that Dylan has been separated from me by a pain of soundproof glass, as nature intended. And I hope

that you guys can join me for future episodes. Just go to twitch dot tv slash tech Stuff and you'll find a schedule right then and there, and I'll talk to you guys again really soon for more on this and of sense of other topics. Is it how stuff works dot com, wh