Under the Weather Hood: an Explainer on Traditional Weather Forecasting versus Long-Range Weather Forecasting Approaches

A typical 10-day forecast from a weather app.

Folks often ask me how the weather forecasts on their phone’s weather app are different from the long-range weather forecasting that Planette does.  As in, when we provide long-range weather forecasts on Eddy™ or the Sura™ platform, are we just using a regular weather model and using it to predict into the future for a lot longer?  And wouldn’t this mean that the forecasts that we produce are going to be worse than your typical 10-day or 14-day weather forecast, since you’re using a similar model but are using it to predict further into the future when things are even more uncertain?

It turns out that there is a huge difference between the weather forecasts that you usually see on websites and apps, which predict 10 to 14 days out, and long-range forecasts, which can stretch from weeks to months to years into the future.  This difference lies in the complexity of the models that are used for these different types of forecasts.  Let’s unpack these.

Traditional Weather Models are (Mostly) Atmosphere Models

In a typical weather model, the most important and interesting part is the atmosphere.  The atmosphere is complex, and to model it properly, you need many different parts.  You have a bunch of mathematical equations to model how the atmosphere moves around its bulk – the dynamical core.  Then you have other major sets of equations to model how the sun’s light and radiating heat move through the atmosphere.  You have sets of equations to model when the atmosphere is unstable and there will be a storm – a convection model.  And more sets of equations to model clouds and tiny particles (microphysics and aerosols, respectively), chemical constituents, and the special interactions that occur between the atmosphere and the Earth’s surface (the boundary layer).  All of these equations are put into forms that can be solved on a computer, and – voila! – you have your weather model.

The takeaway is that a weather model is (mostly) an atmosphere model.  But while the atmosphere is an important part of the Earth system, it’s not the entire Earth system.  The Earth system consists of the atmosphere, ocean, terrestrial surface (i.e. land, including soil and vegetation), and ice (including the sea ice, which floats on the ocean, and land ice, which forms the Greenland and Antarctic Ice Sheets).  Weather models deal with all this complexity by treating the surface – either the ocean surface or the land surface – as a boundary condition.  This means that information about the temperature of the ocean, or the vegetation cover or soil moisture of the terrestrial surface, are given to the model as static quantities.  This means that there is not a physical representation of the ocean or land surface or ice in the weather model.  

While weather models are very good at doing what they do, which is to predict the weather about 10 days into the future, they are not good at looking anywhere beyond a couple of weeks.  This is because the atmosphere, more or less, has no memory of where it started from beyond a couple of weeks.  There are arguments about whether we can stretch this limit and squeeze out a little bit more predictability from the atmosphere, possibly by creating models with even better physical representations of atmospheric processes, or better constraining where the atmosphere starts from at the beginning of the forecast.  But there are real physical limits to how predictable the atmosphere is by itself, which manifest in the atmosphere’s two-week forgetfulness window.  

Earth System Models: Modelling Everything, Everywhere, All at Once

So to make predictions beyond two weeks, and to understand anything about the Earth system beyond hours and days (climate, anyone?), we need to have more complex models.  These models are known as Earth System Models – referred to as ESMs.  Unlike traditional weather models, ESMs contain representations of all the important parts of the Earth system.  This includes the atmosphere, ocean, sea ice, land ice, and land surface.  These models even include representations of the terrestrial and oceanic carbon cycles, which are extremely complex.

But it’s not enough for us to model each of these parts of the Earth system.  We also need to model how these different pieces of the Earth system interact with each other.  The coupler is the part of the model that contains a set of physics equations that model these interactions.  The ocean, atmosphere, land, and ice are constantly talking to each other, and it is through the coupler that these messages pass from one model component to another.  These interactions add another layer of complexity to Earth system modelling – it is not about the individual parts, but how the parts interact to make up the whole.  

It’s also important to point out that each of the components of an Earth system model are extremely complex.  Remember all of those sets of equations that we need to model the atmosphere?  Our Earth system model contains all of those equations, and also contains similar equations that represent the physics of these other model components, including the ocean, land, and ice.  Because we need to make sure that each of these parts of the Earth system are represented by our latest scientific understanding of the underpinning physics, Earth system models are enormous (1+ million lines of code), unwieldy, and computationally expensive.  

An atmosphere model is just one part of an Earth System Model, which includes physical representations of the atmosphere, ocean, land ice, sea ice, and land surface.

A Complex Modelling Approach for a Complex System

But if you want to understand anything beyond the weather over the next 10 days, you need to represent the parts of the Earth system essential for that long-range outlook.  As we’ve described in earlier blog posts, the land surface, ice, and ocean are all essential for being able to look deep into the future.  While the atmosphere forgets after a couple of weeks, the land surface, ocean, and ice have much longer memories. In fact, the deep ocean can take centuries to forget where it started from!  It is these parts of the Earth system that allow us to make accurate predictions a few weeks, months, or years into the future.  And if you’re looking to mid-century or to the year 3000, Earth system models are essential for understanding how our climate might evolve going forward.