When politicians debate measures to cut emissions, when Statnett plans grid investments, or when research scientists estimate how much wind power Norway will need in 2050, they often lean on the same kind of tools: energy system models. These are large computer models that simulate how the entire energy system might evolve over the coming decades. They have become some of the most important tools for planning the energy transition.
But a model is always a simplification of reality. The interesting question is not whether energy system models are wrong but where, and what we can do about it.
That question is the subject of the first report from FME InterPlay, a Norwegian research centre studying how we can make better decisions for integrated energy systems. Rather than only reviewing the scientific literature, we asked 21 Norwegian modelling teams directly: where are the blind spots in your models? The answers were given anonymously by researchers and experts from academia, research institutes, public administration and industry. It is worth noting that the survey was answered by researchers, not by the models themselves. The answers are therefore based on the researchers’ own experience and perceptions from using the models in practice. They show how the models are experienced by the people who use them, but are not necessarily an objective account of what the models can do or how well they perform.
What is an energy system model? An energy system model is a computer tool used to explore how the energy system might evolve in the future. The model describes, among other things, energy production, the power grid, industry, buildings and transport. Scientists can use it to investigate questions such as: if we are to meet the climate targets, and therefore use far more electricity than today, which combinations of wind, solar, hydropower, energy storage, and grid expansion could meet the need? And how can district heating and energy efficiency measures reduce the electricity demand? These models typically search for the solutions that meet the targets at the lowest possible cost.
An honest confession about the models’ weaknesses
The idea behind anonymisation was to make it easier for modellers to speak openly about the weaknesses.
Most of the modelling teams believe that their models can often overestimate the profitability of renewable electricity. Meanwhile, models can favour large, centralised solutions over local alternatives.
On the other side, many pointed out that models often underestimate two challenges that recur in the Norwegian energy debate: constraints in the power grid, and the energy system’s need for flexibility. By flexibility we mean the ability to use, store or produce electricity at different times so that supply and demand are always in balance. Put simply, flexibility is about using energy at the right time. For example, electricity consumption can be shifted to hours with plenty of renewable power, or heating can switch to other energy sources when electricity prices are high.
Why does this happen? The cause is not necessarily bad data, but the fact that all models must simplify reality in different ways. To analyse how an entire country’s energy system might develop up to 2050, the models have to simplify a highly complex reality. When details disappear, important challenges can become less visible. For example, a model might treat all of southern Norway as a single area, so that bottlenecks in the power grid are not captured. It might also represent a whole year using a handful of “typical days”, smoothing away the cold, calm periods that in practice determine how much power generation, storage and flexibility the system needs. This is what we call structural uncertainty: uncertainty that stems from how the model is built, not from the numbers fed into it.
Two kinds of uncertainty: Input uncertainty comes from the assumptions fed into a model (for example, projections about future technology costs, fuel prices and demand). Structural uncertainty comes from the model’s own architecture: how finely it divides time, space and technology detail, and which real-world mechanisms it simplifies away. The second kind is harder to spot, because it is baked into the tool itself.
If a model systematically underestimates grid bottlenecks and flexibility needs, its “optimal” energy system may look cheap and elegant on paper but prove difficult to implement in the real world. And with Norway facing major decisions about electrification, grid expansion, energy efficiency, and new industry, that is more than an academic concern.
Diagnosed, but not yet treated
Perhaps the most sobering finding is not the biases themselves, but how they are handled. Asked how they typically deal with their models’ known weaknesses, most teams gave the same answer: they discuss them. A caveat in the report, a paragraph on limitations, a qualitative judgement of which results to trust. Beyond that, uncertainty from model inputs is usually addressed via sensitivity analysis. This means to test how the model outcomes respond to changes in input data.
That is honest scientific practice, but it is not a fix. More systematic remedies, like adding detail where it matters, linking models together, or optimising for several objectives at once, are used far less often. The gap between knowing about a bias and correcting it – and adding additional modelling perspectives – is where we aim to work at FME InterPlay.
No single model has all the answers, so the models must talk to each other
The obvious response to a “too-simple” model is to make it more detailed. But detail has a price: computing time grows rapidly, and at some point, the model simply becomes unsolvable or too big to manage. You cannot build one model that simultaneously captures decade-long investment decisions, hour-by-hour power flows in every part of the grid, and consequences for biodiversity.
An alternative is model coupling: connecting specialised models so that each contributes with what it does best. A long-term investment model can propose an energy system for 2050; a detailed power market model can then stress-test that system hour by hour and report back where it breaks.
The survey shows that Norwegian teams frequently couple models. It also shows why the practice can disappoint: the models being connected often rest on different assumptions about the future. Only about a third of the teams reported that key input assumptions were fully harmonised across their coupled models. When two models disagree about basic premises (future energy prices, energy use, technology costs), combining them can produce results that are fragmented or inconsistent. Roughly as many teams were dissatisfied with past coupling exercises as were satisfied.
One of the most important proposals is not about new models, but about something more fundamental: better coordination of data and assumptions. In the report we propose a shared “model hub” where data and scenario assumptions are stored and managed independently of the individual models. Then every model in an analysis can build on the same picture of the future.
Cheapest is not the same as best
There is a more fundamental limitation, too. Most energy system models are designed to find the most cost-effective ways of meeting energy needs. But real decisions involve more than cost alone. A wind farm’s price tag, for example, says little about its consequences for biodiversity, landscape or reindeer herding. The cheapest solution is therefore not necessarily the best one, or the one with the broadest support in society.
Our survey found that considerations such as biodiversity, environmental impacts and how people actually make choices often fall outside the models. Biodiversity is a telling example. Many models take land use into account, but that does not necessarily mean they capture the natural values found in those areas. Nature is often represented through a few constraints or indicators, while the complex relationships between species, habitats and ecosystems are largely left out.
One promising way forward is what is known as modelling to generate alternatives. Instead of searching for a single cheapest solution, the model is used to find several solutions that cost slightly more. That makes it easier to choose alternatives that also take account of nature, land use, security of supply and other societal goals. For many, a slightly more expensive energy system is a much better one.
Why this matters beyond the modelling community
The energy transition is becoming ever more complex. We need to electrify transport and industry, build new power generation and, at the same time, take account of nature, land use and security of supply. That is why energy system models are becoming ever more important. The choices they shed light on involve billions of Norwegian kroner and can have lasting consequences for nature, local communities and the energy system for decades to come.
Our report does not show that energy system models are just “playing with numbers”. It gives us a better map of where the models have their blind spots, including electricity demand, the power grid, flexibility and considerations that don’t show up on a balance sheet. It also points out a direction for further research. We have identified the blind spots. The next step is to fill them in.
Pernille M. Sire Seljom leads InterPlay’s Research area 1: Integrated energy system models, and is lead author of the FME InterPlay report “Advancing Integrated Energy System Modelling: Research Gaps and Priorities” (Chang, Seljom, Sazon, Reigstad, Backe & Straus, 2026), the first deliverable of InterPlay’s Research Area 1. The full report is available here: Advancing Integrated Energy System Modelling: Research Gaps and Priorities.
Comments
No comments yet. Be the first to comment!