Groundwater Modeling Pitfalls

This is the fifth article in our six part series on groundwater modelling.

In previous posts, we explored the journey from the messy, unpredictable Real World, through the simplified Model World, and into the quantitative Math World (Decoding: Bringing Groundwater Models Back to the Real World). Along the way, we saw how models sharpen understanding, support sound decisions, and help us act with purpose when they’re built thoughtfully and with consequences in mind.

But even good models can go wrong. Often, the problem isn’t catastrophic failure but a collection of small missteps, value judgements, or misunderstandings about what modeling is meant to achieve. Modeling is rarely a straight path from question to answer. It’s a looping, iterative process of forward, back, adjust, repeat.

Knowing where things commonly go off track is one of the best ways to keep your models useful. Here are three pitfalls that regularly turn promising models into misleading ones and how to avoid them.

Fuzzy Purpose: When the “Why” and the “What If” Aren’t Clear

Many authors, including Franklin, Dym, and Starfield have repeatedly stated that every useful mathematical model starts with a clear purpose. Without that anchor, even the most elegant math won’t help you. But purpose alone isn’t enough. You also need to understand the consequences of being wrong.

The purpose tells you what decision the model will support. The consequences tell you how much rigor, detail, and effort that decision deserves. Together, they guide what to include, what to simplify, and how much time and data to invest.

If the consequences are low, shortcuts are often fine. Take a simple example: before a pumping test, you need to size a pump. That still requires a model, but at this stage, a basic Theis solution might be all you need to make a reasonable first estimate. You’re not forecasting regional drawdown for decades; you’re just trying to get the right pump on site.

When the consequences are high, shortcuts become dangerous. If the model will guide infrastructure design or inform environmental risk assessments, the stakes of being wrong are far greater. The model’s purpose must then be tightly defined, and the effort scaled to match those consequences.

Failing to align purpose and consequences leads to “model drift.” A groundwater model designed for a narrow compliance check might be perfectly valid for that use but unreliable if reused for long-term water planning. Purpose and consequences are inseparable, and losing sight of either one undermines the value of the model.

How to avoid a Fuzzy Purpose

Define both the purpose and the consequences at the start. Ask not only “What decision will this model support?” but also “What happens if we’re wrong?” That level of clarity will anchor every decision you make as you build and apply the model.

The Complexity Trap: When More Detail Hurts, Not Helps

It’s easy to believe that more detail makes a better model. In reality, it often does the opposite.

Anthony Starfield reminds us that our ability to process complex interactions is limited. The more clutter we add, the harder it is for us to understand what the model is doing. David Nordstrom calls this the complexity paradox: as models become more sophisticated in an attempt to mimic reality, they become harder to test, harder to explain, and ultimately less useful.

Economists John Kay and Mervyn King go even further warning that excessive mathematical sophistication (especially in areas of deep uncertainty) often leads to “calculable ignorance.” It creates the illusion of precision without improving understanding.

How to avoid the complexity trap

Follow Barbour and Krahn’s advice and “start simple.” Use Occam’s Razor as a guide: include only the complexity needed to explore the behaviour you care about, and add detail gradually as new data or insights justify it. A model that’s simple, transparent, and closely tied to the question at hand is far more powerful than one weighed down by unnecessary intricacies.

The Prediction Illusion

Barbour and Krahn note that engineers often expect models to predict the future when their real value lies in understanding it. Natural variability and data limits make precise prediction nearly impossible. A model judged only by how closely it matches future measurements will always disappoint.

Erica Thompson, in Escape from Model Land, echoes this point. Once inside “Model Land,” it’s easy to mistake a neat set of equations for reality itself. The illusion of predictive power can replace the real purpose of modeling: to explore possibilities, test ideas, and reveal sensitivities.

How to avoid the prediction illusion

Treat modeling as an ongoing conversation with reality, not a one-time prophecy. Compare results with field data whenever possible, learn from the gaps, and use models to guide better questions rather than to deliver final answers. The most valuable model is not the one that forecasts perfectly, but the one that helps you think clearly.

Final Thoughts: Useful Means Honest

Good models don’t need to be perfect; they need to be honest about their purpose, their structure, and their limits.
Clarity of intent, disciplined simplicity, and realistic expectations will keep any model grounded in the real world rather than lost in abstraction.

In our next post, we’ll tackle the question that follows naturally: if models are only ever approximations of the real world, how do we build trust in the decisions our models inform?

How to spoil a perfectly good model

Stay in Contact

Legal

Book to join us at our next informal 1 Hour drop in session