Anyone can run a model. Building a reliable framework is the challenge

When someone says to me, “I’ve learned how to run WaveWatch III” or “I can launch cases in SWASH now”, I’m often tempted to reply: “Great… and then what?”. Because running the model is just the tip of the iceberg.

Those of us working in coastal numerical modelling know that opening the software, setting up a basic input, and running a simulation can be learned in a matter of days or weeks. Even generating a nice animation of the free surface can look impressive at first glance. But that alone doesn’t make you an expert.

The real challenge begins when you have to build a modelling framework that’s robust, efficient, and reusable. That’s where the amateurs and the professionals part ways.

It’s Not Just the Physics – It’s What You Do With It

Understanding the physics behind the models — energy transfers, dispersion, friction, nonlinearity, slope effects, infragravity generation, etc. — is fundamental. But it’s not enough. Knowing the theory without knowing how to implement it is like memorising chess rules without ever playing a proper game.

The hard part is translating that physics into a consistent set of decisions: which boundary conditions to use, what grid resolution you need, how to handle margins and buffers, which cases to compare, which parameters to vary, and when to trust a result or raise a red flag.

That process alone can take months of trial and error, and sometimes a full year of continuous work to arrive at a solid scheme. And by “solid scheme” I don’t just mean something that works once — I mean a setup where, if your client or supervisor asks for a modification, you can tweak a single input or rerun a script and have the whole system regenerate seamlessly, with full control over what’s going to happen, how long it’ll take, and what the outputs will look like. That level of stability and traceability doesn’t happen by accident. You build it.

Welcome to the Land of Raw (and Heavy) Data

Another underrated skill is managing data efficiently. Sure, the model runs — but it can leave you with output files weighing several gigabytes per case. And you can’t afford to read everything every time.

Writing custom extraction scripts that pull only what matters is a skill in itself. Which section of the domain matters? Which variables? What frequency? What format? If you don’t ask these questions upfront, you’ll end up drowning in irrelevant data or crashing your machine with unnecessary processing.

That said, today’s “useless” data may be tomorrow’s gold when new research questions emerge. That’s why robust data organisation and backup strategies are also part of the job. Clean directories, clear naming conventions, metadata tracking — they’re not just housekeeping; they’re survival skills.

Post-Processing Is Where You Earn Your Stripes

Once you’ve got your raw output, the real work starts: extracting relevant, actionable results. And this is rarely straightforward. Sometimes you need to convert velocity into discharge, integrate quantities over time or space, isolate infragravity contributions, or apply extreme event criteria — and do it for dozens or hundreds of cases.

That’s where your personal toolbox of scripts and routines becomes invaluable. Automating these tasks isn’t just about saving time — it allows you to validate, compare, detect systematic errors, and rerun only what’s necessary, rather than starting from scratch every time.

Experience Is Something You Build, Not Download

After years of working with models like SWASH, ADCIRC, SWAN, STWAVE, WaveWatch III and proprietary codes, I’ve come to a simple conclusion: real expertise isn’t about launching simulations, it’s about using models as effective tools under pressure, with clear objectives, while managing dozens of moving parts and countless iterations.

When you finally move from a fragile, manual setup to a fully optimised workflow — with every step streamlined, from input to plots — that’s when you know you’ve levelled up.

In Summary

• Running the model is easy. Building the model is what matters.

• Knowing the physics helps — but knowing how to implement it well is what sets you apart.

• Having the raw data is pointless if you don’t know how to extract what matters without crashing your system.

• Expertise isn’t about knowing which button to press. It’s about developing your own tools to make the work sustainable, scalable, and adaptable.

• And when last-minute changes come in, you should be able to rebuild the entire workflow with one input file, without panic or chaos. That’s real modelling.

Everything else is just pressing “run”.