What is a Swell Model?

Ben Freeston

by on

Updated 1815d ago

We’ve come to depend completely on model-driven surf forecasts. Even the most skeptical of surfers keep an eye on the weather and swell charts generated from computer models - but where and how are they made and how can understanding this help us score more waves?

This is the first of a series of features breaking down exactly how modern surf forecasts work and how you can use them to score more waves. We’re going to try to put things together in as accessible a way as possible, but this is complex stuff and we’re not going to pull any punches. If you love surfing, if you travel and search for the best waves you owe it to yourself to figure this out and we’re here to help. You WILL score more waves.

Of all the forecasting issues and questions we answer 90% relate to misunderstandings about how swell models work.

You often hear surf forecasters talk about a ‘swell model’. This forecasting oracle is the centre point of all modern surf forecasting. It informs our discussions and powers the surf forecasting websites that help make our decisions. Yet it’s poorly understood, even by those whose living or leisure time increasingly depends on it or those who claim to be surf forecasters. Of all the forecasting issues and questions we answer 90% relate to misunderstandings about how swell models work and how to use the data we present here to predict the waves. Most surfers reading this won’t even know what a swell model is - and the blame for that is ours - no one has every attempted to actually explain it.

What is a swell model - the basics.

A ‘model’ is a simplification of a system into a set of rules that can be applied to explain and predict behaviour. For swell the following is a very simple model:

STRONGER WINDS = BIGGER SWELL

This model can be broken down into three parts, each of which will be important to understanding the more complex swell models we use to predict the surf:

Input

We can only predict the future with some idea of the present. In the case of this really simple example the input to our model is the wind strength. If we don’t know, or can’t measure this accurately we can’t predict the outcome.

Rules

While our basic example isn’t going to win any prizes for it’s physics it’s still telling us something about the fundamentals of the ocean: That wind is part of the process of creating swell, and that there’s a correlation between the wind strength and the swell size. These rules deliberately simplify reality - if the model was as complex as the thing it was trying to predict we’d never actually be able to use it. The art is enough simplification to make it usable with enough accuracy to make it useful.

Output

Our model is attempting (poorly) to predict swell size. It’s limitations will be numerous, it won’t be predicting breaking wave size, swell period, consistency, quality or any other factor important for surfing.

Our in-house cluster computer we use to crunch the complexities of an operational forecast model.

Our in-house cluster computer we use to crunch the complexities of an operational forecast model.

© 2019 - magicseaweed

How does a modern swell model work?

The first difference between our simple example and a working swell model is in scope and scale. Because swell can travel for thousands of miles unobstructed, forecasting for most locations means we need to consider the whole globe. As we mentioned above a model works by simplifying - we couldn’t possible calculate the movement of every drop of water in the ocean. Instead we divide the Earth into a grid of squares, commonly about 30 miles by 30 miles across and calculate the swell in each square, and how it travels to surrounding squares over time. For this we need a computer. While the inputs, rules and outputs could all be calculated by hand it’d take longer than the process itself. That’s to say we’d end up predicting tomorrow’s swell some time next month. It’s only with the advent of powerful computers that we’ve been able to make enough calculations fast enough for our model to be useable. In fact the physics outdates the model - we knew how to do this before we'd invented the computers to make it possible.

Inputs

We need to know exactly the state of the wind at the start of the forecast. How this happens is one of the hidden wonders of weather prediction. In a nutshell we can measure the wind strength accurately over the oceans' surface using satellites (which detect ripples in the surface created by the wind). But on land we rely on weather stations and hundreds of weather balloons released by meteorologists all over the globe. This is happening all day, every day. If individuals in the far flung reaches of the Earth weren’t heading out every single day and inflating and releasing these measuring balloons our forecast wouldn’t have the accurate start conditions it needs.

For our swell model we need not just the winds at the start, but for the whole forecast, for this a swell model talks to a wind and weather model. This takes the current state of the atmosphere and calculates how it’ll change over time. It’s a computer model of a similar type to a swell model, but more complex in operation.

We also need to know something about the geography of the oceans: Where they're shallow and deep. Where the coastline is and what is land and sea. We need to know the extent of current sea ice, which impedes or blocks swell depending on it’s thickness. We need to break all this information down into our grid representing the world.

We need to know the sea state at the start of our forecast, we normally get this from the last forecast rather than observation.

NOAA weather balloon launch. This happens every day and is crucial to your forecast accuracy.

NOAA weather balloon launch. This happens every day and is crucial to your forecast accuracy.

© 2019 - NOAA

Rules

A swell model attempts to understand the physics of the ocean. We’ve already represented the Earth's surface as a grid of squares. We now represent the exact state of the sea at every location with another grid representing each period and direction and how much energy is present at each. This is called the model spectra.

The rules are implemented with a large amount of complex mathematics that take the data about every period and direction in our grid square and neighbouring grid squares and calculate how they’ll evolve over the space of a few hours. This mathematics models the physics of ocean processes including: waves being created or increased by the wind, waves interacting with other waves, waves refracting in shallower water, waves being obstructed by coastline or offshore islands and more. We’ll write more about some of these processes in a future article.

Outputs

The swell model outputs the complete model spectra for every grid square on Earth for every few hours for up to about the 16 days we run it for here at MSW. From this spectra the model creates a range of simplified numbers that we’ll recognise from our forecast, for example the significant wave height and period. How this simplification occurs and what’s lost in the process will be an important part of understanding your forecast and something we’ll cover in detail in a future post.

A few key facts.

- Wave buoys aren’t involved. Sorry. This myth that the forecast comes from wave buoys is worth putting to bed here. We can use them to check that our model is working well and as a small part in gathering the wind data we need to initialise the model, but that’s the extent of it.

Wave buoys aren’t involved. Sorry. This myth that the forecast comes from wave buoys is worth putting to bed here.

- For those who still insist that relying on a pressure chart is how ‘real’ surfers call the swell: Our main model implementation operates on a grid of over 220,000 locations. For each of these locations it calculates the physics described above calculate the swell energy for 1080 combinations of direction and period. For every three hours we’re calculating for almost 250,000,000 data points. Over the course of the 16 day model that’s almost 31 BILLION data points and a great many more calculations to implement the complex physics. Even if you have the mathematical ability you’re going to long since miss the swell you’re forecasting if you try to do this yourself. That isn’t to say an understanding of how surface pressure relates to wind and how this relates to swell isn’t important, in fact those links are the ones we’ll be trying to make in these articles.

- A swell model is as accurate as the winds it uses as its input. For the most part operational ocean models are powered by the same very small handful of wind models. These wind models are so complex (both in terms of computer power needed and the inputs we described above) that there almost exclusively the preserve of large, well funded, national agencies. Historically getting a more accurate idea of current winds has been the battle ground in making more accurate models. Almost every operational surf forecast you’ll see uses the American ‘GFS’ model as it’s input, including MSW. This means that the results are broadly similar in deep water.

- A modern, well implemented swell model is very accurate in the short term. The issues arise when it’s outputs are misinterpreted or expectations nearer the coast are unrealistic. This is our battle as surf forecasters - bear in mind it’s a ‘swell model’ not a ‘surf model’. Getting from the offshore swell in deeper water to the surf on the beach is another whole model process and one for which there’s no short answer. It’s something we’re constantly evolving.

- Swell models themselves are, in principle, as accurate in the long term as they are in the short term. This is because the physics that forms their rules them means that small errors quickly shrink and disappear. However the swell model needs a wind model as it’s input and these are the exact opposite. Small errors in the starting data quickly amplify and this is why a swell model can become increasingly inaccurate over time depending on the location and nature of the swell.

- Because a model, by definition, simplifies things, the outputs are also a simplification. For example the model grid of 30 miles by 30 miles can miss small islands or subtle coastal features entirely. Likewise very small but powerful tropical storm centres can be under or over called. We’ll cover this in more detail in a future article because it’s critical to interpreting your forecast.

- Most ‘surf forecasting’ websites are simply presenting the results of one publically available American model (NOAA/NWS WaveWatch). That’s a simple way of saying they’re cut and paste clones of each other. There’s nothing wrong with this in principle (it’s very well implemented and we use the data ourselves in some parts of your forecast) but in practise the inability to get beneath the hood and adjust things when they’re not working so well for surfing or being able to extract detailed data for a specific location means their scope can limited. Further a lack of understanding of their operation can mean coastal forecasts created from them are poorly designed. It also means checking a few different websites is often just a check of the same data in a different colour scheme.

The challenges of doing things properly are cost and expertise. At MSW we’ve invested tens of thousands of dollars building our own cluster computer to handle the complexities of running an operational model. We also employ a full time model engineer to maintain and improve this system. As far as we know only Magicseaweed and Surfline run their own global wave models and only ours has the very latest physics operational.

- We run our swell model four times a day. The battle is to harness enough computer power to do this in less than the six hours between the next time the model runs. On a larger scale this means that thousands of detailed satellite and local measurements are fed into the US supercomputers that run their wind model, these results are downloaded by us and fed into our own cluster computer and the output created, processed and uploaded back to our servers in the US so that it can appear on the website. All this is happening constantly, from a lone weather observer in a far flung Antarctic outpost through to our office servers and back out onto the web data is being collated, interpreted and moved around. Living in the internet age this seems ‘normal’ - in fact that this is happening simply for the business of surfing waves is beyond staggering.

- Because our swell model relies on open source software code and those thousands of bits of input data it’s entirely reliant on the generosity of national agencies. While we’re based only miles from the UK Met Office (who produce their own model data) the reality is the world of public surf forecasting owes it’s debt of thanks entirely to the NOAA / NWS in the US - whose data sharing practises and tireless support of their products is instrumental.

- With the complete data from a well implemented modern swell model, coupled with a really good understanding of exactly what it’s telling you and a really good understanding of your beach and how it responds to different conditions you can reliably predict the surf anywhere on the planet. This is really important to appreciate. These models are complex. Our attempts to simplify them for the ‘average surfer’ lead to a host of problems that can damage their reputation for accuracy with those who aren’t prepared to invest in understanding them. Don’t be the ‘average surfer’. Put in the time to really understand what’s going on and with this data you’ll be able to reliably score surf at home and abroad. We do.

This graph shows our various swell models compared to the actual conditions observed by a wave buoy (blue crosses). Your standard forecast is the red dots (and this is what you'll find on every other forecast site. MSW also run a series of high resolution models (black and blue dots) which offer significantly improved forecasts in some tricky coastal areas.

This graph shows our various swell models compared to the actual conditions observed by a wave buoy (blue crosses). Your standard forecast is the red dots (and this is what you'll find on every other forecast site. MSW also run a series of high resolution models (black and blue dots) which offer significantly improved forecasts in some tricky coastal areas.

© 2019 - magicseaweed

Please ask any questions you have and I'll do my best to answer them below.