Urban design testing in Reykjavik Iceland

Solar optimization

The Sun, the City and the Algorithm

In the video I am sharing with this post you will see a 3D model of a mixed-use neighbourhood constantly shifting: heights step up and down or disappear. Nothing is “hand-drawn” in the usual sense. Instead, an algorithm is quietly running in the background, testing hundreds of massing variations on a curved grid to find a balance between two things cities usually fight over:

Maximum daylight / solar access, and
Maximum usable floor area for homes, workspaces and ground-floor activity.

This is the kind of systemic, optimisation-driven work we have been moving towards at Thor Architects for years – using parametric tools and systems thinking to let performance and context shape form.

From rules of thumb to algorithmic urbanism

Traditional urban design relies on a series of rules of thumb:

Keep certain distances between facades.
Don’t exceed particular height-to-street-width ratios.
Align streets roughly east–west or north–south, depending on climate.

These are useful, but they are also very blunt instruments for highly complex problems. A mixed-use development in Reykjavík, for example, needs generous solar access in winter but must avoid glare in summer, with less likelihood of overheating. At the same time, planning policy and viability models push for higher density and more floor area.

Parametric design and environmental simulation now allow us to treat these as concurrent, measurable objectives rather than vague competing instincts. Tools such as Ladybug and Honeybee plug into Rhino/Grasshopper to calculate sun paths, solar radiation, daylight autonomy and energy demand directly from a live parametric model.

On top of that, optimisation engines like Galapagos, Octopus or Opossum use evolutionary algorithms to search through thousands of possible configurations and highlight families of “good” solutions that balance multiple criteria.

What are we actually optimising?

For a typical mixed-use block or small district the algorithm can be asked to juggle, for example:

Solar access to public space
- Hours of direct sun in key plazas and courtyards in winter and shoulder seasons.
- Maximum depth of overshadowed zones at street level.
Daylight for apartments and workspaces
- Target daylight autonomy or simple daylight factors for a percentage of units.
- Limits on north-facing single-aspect units in cold climates.
Floor area and density
- Gross floor area or floor-area ratio (FAR) for different use classes.
- Mix of residential, commercial and 3rd-space functions, as I have written about previously. thorarchitects.com

Urban comfort and microclimate
- Optional metrics for wind comfort and outdoor thermal comfort using microclimate scripts at block scale. MDPI+1

Wind factor: In Iceland wind is a major factor in environmental and comfort considerations. This is done through simulating wind using computational fluid dynamics. However coupling this with other parameters is difficult due to the complexity of the computational calculations. In the video, the fluid dynamic adjustment has be done in a previous step.

Instead of testing one massing option at a time, we let the algorithm mutate the geometry – within constraints – and then rank each generation of solutions based on how well they satisfy these objectives.

How the optimisation loop works (without equations)

In practice, the workflow looks something like this:

Parametric model
We define the site as a set of adjustable parameters:
- grid spacing and curvature,
- building footprints and heights,
- setbacks, courtyard radii, street widths, etc.
Performance engine
The model is connected to environmental tools that can:
- read local weather data,
- compute sun paths, shadows and solar radiation,
- estimate daylight and, if needed, energy demand. ResearchGate
Objective functions
We translate qualitative aims into quantitative questions:
- “Maximise winter sun hours on the main square.”
- “Maximise net developable area.”
- “Keep at least X% of dwellings above a daylight threshold.”
The algorithmic “search”
A genetic algorithm (for example Galapagos or Octopus) generates many variants, keeps the ones that perform better, mutates them, and repeats. Over time, this converges on a Pareto front – a set of solutions where you cannot improve one objective (say daylight) without sacrificing another (say floor area). Ladybug Tools | Forum
Architectural judgement
Crucially, the computer does not decide. We, together with the client and the municipality, choose from this Pareto family which trade-off fits the project’s cultural, economic and political context.

In other words, we are not asking the machine for “the best masterplan”. We are asking it to map the landscape of possible cities on that site, so we can make informed, human decisions.

Why a curved grid?

The example in the video uses a curved grid rather than a pure orthogonal one. There are several reasons for this, especially in Nordic conditions:

Primarily to minimize wind effect, as wind cooling will minimize the comfort level of the external spaces, and reduce the time these are used into the colder seasons.
A curved street pattern allows us to open “sun pockets” – widened, south-facing spaces where winter sun can penetrate deep into the block.
The curvature can be tuned so that street canyons avoid permanent shade while still creating urban intimacy.
Building heights can be stepped along the curve so that taller elements sit where they do the least damage to solar access, and where they can potentially host solar panels with good exposure.

Because the grid itself is parametric, the algorithm can nudge radii, angles and block depths as it searches for combinations that push daylight and density up together, instead of treating them as enemies.

What this gives clients and cities

For clients, municipalities and communities this approach changes the conversation:

Instead of discussing a single “artist’s impression” of a masterplan, we can compare evidence-based scenarios generated from the same rules.
We can attach clear numbers to questions like “what do we lose in daylight if we add one more storey?” or “how much sun do we gain in courtyards if we rotate the grid by 10 degrees?”.
We can document the optimisation process and its assumptions, making the design more transparent and robust for planning negotiations and public engagement. ScienceDirect

For me as a designer, the most important point is that algorithmic optimisation is not a style. It is a method – another way of doing what architecture has always done: balancing many competing demands into a coherent, liveable place. The difference is that we can now see that balance more clearly, and use the sun, the weather file and the algorithm as active participants in the design process, instead of treating them as after-thoughts.

If you are working with a complex mixed-use site and want to explore how this kind of parametric, optimisation-driven urban design could apply to it, feel free to get in touch through Thor Architects.

By Gudjon Thor Erlendsson