Environmental systems, traffic, drainage and economics contain more detail than we can control or experiment with directly. Modelling lets us select important relationships and simplify them to understand a process, test a scenario or discover a problem before implementing a real decision. Models can address the same variable while differing in purpose, time representation, formulation and solution method.
The question determines what belongs in the model
Suppose we model a ship to ask whether it will float or sink. We concentrate on the factors needed to answer that question. We do not necessarily represent everything required to calculate its speed or its response to currents and wind. Those matter to complete ship design, but may lie outside this particular model.
A useful model does not need to contain all of reality. Its simplification needs to fit the question. Similarly, a calculation of air concentration at a receptor does not automatically answer every question about pollutant fate, exposure routes or project decisions.
Scientific models: understanding mechanisms
Scientific models focus on understanding and explaining a system. How does it work? How do source, pathway and receptor relate? Which external influences control the output? The model can help develop or test an explanation against evidence and scientific knowledge.
For example, we might investigate the response to a change in source strength or surrounding conditions. The value lies in examining that relationship, even if the calculation does not independently determine the final management decision.
Management models: acting on the understanding
Management models are used more directly to support decisions. Should the design change? Is mitigation needed? Should another monitoring point be added, an operating condition modified or a buffer considered? The focus becomes how to use the assessment to choose an action.
This is not a rule preventing the same formulation from serving both purposes. A calculation may explain a relationship and then compare options. The important point is to identify the purpose against which its usefulness will be judged.
Conceptual models: before software and equations
A conceptual model can begin with boxes and arrows. Boxes represent components and arrows show pathways or relationships: source, transport route, receptor and possible impact. A simple diagram can reveal a missing route or an assumption that has not been stated.
Creating it does not require specialist modelling software, but it does require understanding. If the study team has not agreed what the source, receptor and connection represent, sophisticated numerical work may answer a different question from the one that matters.
Physical models: something that can be tested
Physical models represent a system materially, often at reduced scale. Examples include building models, channels and water-flow experiments in which behaviour is observed rather than calculated only on a screen.
Scaling has conditions of its own. A small representation does not automatically reproduce every full-scale physical process. Definition of what is represented and what is omitted therefore remains important whether the model is a physical object or software.
Static conditions and change through time
Static models describe a condition or simplified state; dynamic models follow system evolution. Environmental concentrations, flows and operating conditions can change within a day or between seasons. The relevant question is how the formulation treats that change, not simply whether the output has a date column.
For example, a steady-state assumption can be applied to a sequence of hourly meteorological conditions. Outputs are then indexed by time without every process being represented as fully evolving within each interval. Knowing that distinction helps explain suitability for a particular application.
Deterministic, stochastic and uncertain
A deterministic model produces the same result when inputs and settings remain unchanged. An AERMOD run can be reproduced by retaining emission rates, meteorology, terrain, receptors and configuration.
A stochastic model represents selected components probabilistically, which can help examine variability or distributions of possible outcomes. Deterministic models can also be explored using scenarios, sensitivity analysis or probabilistic methods around them. Determinism does not imply that reality or the input evidence is free from uncertainty.
Mathematical models express the relationship
A mathematical model writes relationships as equations. Distance = speed × time describes a vehicle moving at constant speed for the purpose of calculating distance. More complex transport and dispersion questions can involve differential equations with initial and boundary conditions.
The strength of the formulation is that relationships and assumptions become inspectable. Writing an equation does not, however, guarantee that it is straightforward to solve. A direct analytical solution may not exist for the desired case or may become impractical with additional detail.
Numerical solutions turn the formulation into calculations
Numerical methods address that difficulty. A computer executes digitally represented instructions and arithmetic rather than interpreting symbols as we do on paper. At the electronic level, logic gates such as AND, OR and NOT help implement that representation and execution.
A solution method can be expressed as smaller repeated operations, performed rapidly to obtain an approximation. Many applications divide time or space into steps or a grid and calculate values on that representation. Not every numerical method requires a spatial mesh; it depends on the problem and algorithm. Accuracy, convergence and computational assumptions still need attention.
Mathematical and numerical can describe the same model
A numerical model has an underlying mathematical or computational representation. The choice is therefore not always “mathematical or numerical” as mutually exclusive alternatives. A mathematical model may have its equations solved numerically on a computer.
This returns us to the learning question: an easy interface does not remove the theory. Understanding simplifications, fixed assumptions and solution methods is necessary to explain why two models differ or where a result can be applied.
Using the classifications in practice
I ask whether I need scientific explanation or comparison of decisions, how time matters, whether probabilistic treatment or scenarios are needed, which variable is being studied and what evidence is available. I then examine application limits and evaluation methods. Model evaluation guidance relates suitability to the intended use. [1]
A model does not replace reality or become preferable simply because it is more complex. Its value lies in simplifying enough to answer a defined question and support a clearer decision while making the limits of that understanding visible.
References & original writing
Adapted and expanded from the author’s LinkedIn writing. Technical references are listed below.
