Any scientific discipline that involves numerical modeling has three kinds of modelers:
- People who develop models
- People who run models for their or other people’s research
- People who analyze model output for their research
Neither is better or worse than the other. Model developers (1) often also run models (2) to test and validate them. People who run models (2) typically do it for a specific research goal (3). Occasionally, they’ll also tweak the code to fit their need, which gives a glimpse into the model developer’s world. And then there are people who don’t fiddle with models at all but simply mine the data for patterns (3). As model output data become increasingly more abundant and easier to access, the barrier to entry for group 3 slips lower.
Each group thinks differently. Model developers think like engineers. After all, numerical models are nothing more than machines made of simple building blocks. People who analyze model output think like scientists — they look for emerging patterns and meaningful connection. They usually don’t concern themselves with the internal workings of models, such as whether energy is conserved, whether mass is positive-definite, or what is the truncation error in the Taylor series expansion of the finite difference operators. Instead, they ask questions like how much does the ocean warm up over time if it’s consistently more cloudy. People who run models must maintain both mindsets — setting up a model simulation is an inherently engineering task, and you need to keep the scientific question in mind while doing it. The simulation must be appropriately configured for the problem at hand.
Each group is smaller than the next. For every model developer, there may be ten people who run models. For every person that runs models, there may be ten that use their output. Being a model developer is lonely. You often work solo or with a few people at most. You can work for three years without producing any publishable results. When you’re building a car, you can’t half drive it when you’re half way done building it. Same with numerical models — it takes a long time before you can show that it works.
Model developers feel pressure to perform. If the model doesn’t work as advertised or is difficult to run, people who run models blame model developers. If the model prediction is off, who do you blame then? Not the person who ran the model, but the person who wrote it.
But it’s rewarding in the long run. Because they’re scarce, model developers are sought after and have job security. If and when you publish, chances are your papers will have an impact and your idea will spread. If you don’t publish, people will still use the tool that you made. At least those people that paid for your work.
So, you want to be a modeler? The first question to ask is “Do I want to build tools for others to use or do I want to use other people’s tools to do my research?” Whatever your preference, the good news is there’s plenty of work to be done on all fronts and that’s not going away any time soon. Models are only getting better, computers are getting bigger and faster, data becoming less scarce. Most of all, humans will never stop reaching for better prediction and understanding of the world.