GIS Site Selection: How Multi-Criteria Analysis Finds the Best Location

Choosing where to put a school, clinic, warehouse, solar farm or water treatment plant is a decision that lasts decades. Get it wrong and the cost shows up every day: long journeys for users, expensive foundations, flood damage or a facility that sits half-empty because demand was somewhere else. GIS site selection exists to compare every possible location against the same rules, so the shortlist comes from evidence rather than from the sites people happened to know about.

This guide walks through the multi-criteria decision analysis process we use, with a worked example, the common mistakes, and how to present the results so they hold up in a meeting with sceptical decision-makers.

What is GIS site selection?

Site selection in GIS evaluates locations against a set of criteria, each represented as a map layer. Some criteria are absolute: a site cannot be inside a protected area or on a flood plain. Others, however, are matters of degree: closer to a main road is better, gentler slopes are cheaper to build on, more people within 5 km means more demand. Therefore, combining these layers shows which areas meet the requirements best and by how much.

In fact, the formal name for this approach is multi-criteria decision analysis (MCDA). Teams use it in urban planning, public health, energy, logistics and environmental management because it makes the reasoning behind a recommendation visible and repeatable.

The multi-criteria decision analysis process

1. Define the goal and the criteria

Start with a precise goal, such as "find three candidate sites of at least 2 hectares for a district hospital". Then work with stakeholders to list what makes a site suitable. Typical criteria include travel time for the population served, distance to main roads, slope, land use, land cost, distance from hazards and access to water and power.

Keep the list short. Seven well-chosen criteria are usually better than fifteen overlapping ones, because overlapping criteria quietly count the same factor twice.

2. Apply constraints

Constraints are yes-or-no rules. For example, typical exclusions are water bodies, protected areas, existing buildings, slopes above a threshold, high flood risk zones and buffers around airports or industrial sites. As a result, you get a mask of land that you could, in principle, use.

3. Prepare and standardise the criteria

However, each criterion comes in its own units: metres, degrees, people per square kilometre, rupees per square metre. To combine them, the analyst rescales each layer to a common suitability scale, for example 1 (least suitable) to 10 (most suitable). In addition, the rescaling can be linear or follow a curve, and the direction matters: for slope, low is good; for population served, high is good.

4. Weight the criteria

Weights express relative importance and must add up to 1 (or 100 percent). Stakeholders can agree them directly in a workshop, but the Analytic Hierarchy Process (AHP) is more reliable when several stakeholders are involved. AHP asks people to compare criteria in pairs, which is easier than assigning numbers to everything at once, and it flags inconsistent judgements with a consistency ratio.

5. Combine with weighted overlay

Multiply each standardised layer by its weight and add them together, then apply the constraint mask. The output is a continuous suitability surface.

6. Identify candidate sites

However, a raster surface is not yet a site. Group adjacent high-scoring cells into contiguous areas, remove patches smaller than the minimum site size, and rank the remaining areas by their mean score. Consequently, this produces a shortlist of real, buildable parcels.

7. Test sensitivity

Rerun the analysis with different weights. If a site ranks highly only under one weighting, it is fragile. Therefore, sites that stay near the top across several scenarios are the ones to recommend.

8. Validate on the ground

Then check the shortlist against current imagery, land ownership records and site visits. GIS data is never perfectly current, and a vacant plot on a two-year-old dataset may already have a building on it.

A worked example: siting a public facility

A model for a public facility, similar in structure to our multi-criteria site suitability study, might look like this:

For instance, constraints would exclude water, protected land, existing buildings and slopes above about 15 percent. After weighted overlay, we rank contiguous areas of at least the required size, test three or more weighting scenarios, and the sites that perform well in all of them go forward for field checking. The weights shown are illustrative; on a real project we agree them with the client and stakeholders.

Using demand data in site selection

For any facility that serves people or businesses, demand matters as much as physical suitability. A perfect, flat, cheap site in the wrong place will still fail. Population grids, customer locations or employment data can become demand surfaces using kernel density or travel-time catchments. Our employment density mapping of Dübendorf is an example of the kind of layer that can feed directly into a site selection model.

Common applications

• Public facilities: schools, health centres, police and fire stations, where travel time and coverage gaps dominate.
• Retail and logistics: stores, warehouses and distribution hubs, driven by customer demand, competitors and road access.
• Renewable energy: solar farms (irradiance, slope, grid distance) and wind farms (wind speed, distance from homes, aviation constraints).
• Waste management: landfills and transfer stations, where groundwater, soils, distance from settlements and haul routes matter.
• Agriculture: crop suitability based on soils, climate, slope and irrigation access.
• Housing and urban expansion: identifying land that can be serviced without high hazard exposure.

Mistakes that weaken a site selection study

• Double counting. Including "distance to roads" and "distance to towns" when towns sit on the roads gives that factor twice the intended influence.
• Mixing constraints and criteria. If a slope above 15 percent is unbuildable, treat it as a constraint, not just a low score that other factors can outweigh.
• Distances in the wrong units. Always calculate distance surfaces in a projected coordinate system; otherwise a "2 km" buffer will not be 2 km.
• Ignoring data dates. Land use, roads and facilities change quickly. Check the age of each dataset.
• One scenario only. After all, a single weighting hides how sensitive the result is to opinion.
• Presenting a heat map as an answer. Above all, decision-makers need named, ranked candidate sites with the reasons behind each ranking.

Presenting the results

The final report should let a reader follow the logic without being a GIS specialist: the goal, each criterion and why we included it, the constraints, the weights and how they were agreed, a suitability map, a ranked table of candidate sites with their scores on each criterion, the sensitivity results, and the recommended next steps. In addition, a short interactive map alongside the report helps stakeholders explore the candidates themselves.

If you are planning a new facility or network of sites, our spatial analysis team can design a transparent site selection study and run the scenarios with your stakeholders.