Site suitability analysis in GIS: weighted overlay explained in plain English
"Where is the best place for this?" is one of the oldest questions in GIS, and weighted overlay is the standard way to answer it. You take several maps of things that matter, put them on a common scale, decide how much each one counts, and add them up. The result is a map where every place has a score.
The method is simple enough to explain on a whiteboard, which is exactly why it is so often done badly. This guide walks through it step by step for people who do not use GIS every day, covers the mistakes that make a suitability map indefensible, and shows how to run one in NeuraCities by describing your criteria in a sentence.
What site suitability analysis is
A site suitability analysis answers a ranking question about places: which parcels, cells or neighborhoods are the best fit for a given use, according to stated criteria. Planners use it to find land for housing or parks, developers for retail and industrial sites, utilities for substations and solar, and resilience teams to find where need is greatest.
Weighted overlay is the most common technique because it is transparent. Every part of the result traces back to a layer, a scale and a weight that someone chose and can explain. Esri's documentation, How Weighted Overlay works, is the standard reference for the method, and this guide follows its steps.
The whole analysis comes down to five decisions:
- 1Which places are ruled out entirely (constraints).
- 2Which characteristics make a place better or worse (factors).
- 3How each factor's values translate to a common suitability scale (reclassification).
- 4How much each factor counts relative to the others (weights).
- 5How to check the result is sensible (validation and sensitivity testing).
Constraints vs factors: filter first, then score
This is the distinction that matters most, and the one most often blurred.
| Constraint | Factor | |
|---|---|---|
| Question it answers | Is this place possible at all? | How good is this place? |
| Result | Yes or no | A score on a scale |
| Examples | Inside the FEMA floodway; parcel under 10 acres; protected land; zoning that prohibits the use | Distance to a highway; population within a drive; land value; slope |
| How it is applied | Removes places before scoring | Combined with other factors in the weighted sum |
Esri's tool handles constraints with a special Restricted value, which assigns a cell "the minimum value of the evaluation scale set, minus 1" so it can never score as suitable (Esri). On parcels, the equivalent is simply removing them before you score. Either way, the point is the same: a constraint is not a factor with a very low score. It is a hard no.
Sometimes the constraints alone are the answer. In seven dry parcels in Nashville, four tests (vacant, over 10 acres, within half a mile of an interstate, outside the 100-year floodplain) took Davidson County's 286,920 parcels down to seven. With a list that short, scoring is optional. Weighted overlay earns its keep when the constraints leave hundreds of candidates and you need to rank them.
Step 1: reclassify every factor to a common scale
Factors arrive in different units: metres to a road, dollars of income, percent impervious surface, degrees of slope. You cannot add metres to dollars. So each factor is translated into the same suitability scale, where a higher number always means better.
Esri describes reclassifying each input "into a common preference scale such as 1 to 10", and its Weighted Overlay tool's default evaluation scale runs from 1 to 9 in increments of 1 (Esri). The scale itself matters less than using the same one for every factor.
Two things to decide for each factor:
- Direction. Is more better or worse? Closer to a highway is better for a warehouse; closer to a highway may be worse for a school.
- Breaks. Where do the values change score? You might give 9 to anything within 1 km of an interchange, 6 to 1 to 3 km and 3 beyond that. Breaks can follow a standard (a ten-minute walk), an engineering limit or a simple even split of the range.
An alternative to class breaks is normalising: rescaling each factor linearly so its lowest value becomes 0 and its highest 1. It keeps more detail but lets a few extreme values compress everything else, so look at the distribution before choosing.
Step 2: choose the weights (and admit they are a judgement)
Weights say how much each factor counts. In Esri's method each input gets a percentage influence, and "the total influence for all rasters must equal 100 percent" (Esri).
There is no formula that produces the correct weights. They encode priorities, and priorities belong to people: a council, a client, an investment committee. The honest approach is to choose them deliberately and write them down.
- Equal weights are a reasonable default when there is no basis for preferring one factor, and they are easy to explain.
- Stakeholder ranking asks the decision-makers to rank or allocate 100 points across the factors.
- The Analytic Hierarchy Process (AHP), developed by Thomas Saaty in the 1970s (overview), breaks the choice into pairwise comparisons ("is flood risk more important than road access, and by how much?") and derives weights from the answers, with a check on whether the comparisons are consistent with each other. It makes the judgement more structured, but it is still a judgement.
Step 3: multiply and sum
This is the easy part. For each place, multiply each factor's score by its weight and add the results. Esri's description is exactly that: cell values of each input are multiplied by their weight, and the results are added to produce the output (Esri).
An illustration, using the urban forestry weights above and made-up scores for one place on a 1 to 9 scale:
| Factor | Score (1 to 9) | Weight | Score x weight |
|---|---|---|---|
| Heat | 8 | 35% | 2.80 |
| Social vulnerability | 6 | 35% | 2.10 |
| Stormwater need | 4 | 30% | 1.20 |
| Suitability score | 100% | 6.10 |
Do the same for every place and you have a ranking. Because the weights add to 100%, the result stays on the original 1 to 9 scale, which makes it easy to read.
Raster or vector: grid cells vs parcel scoring
Weighted overlay was born on rasters, and that is still how Esri's tool works: it takes integer rasters as inputs, so continuous rasters are reclassified first (Esri). But the arithmetic works just as well on a table of polygons.
| Raster version | Vector (parcel) version | |
|---|---|---|
| Unit of analysis | Grid cells (for example 30 m) | Parcels, census tracts or hexagons |
| How factors are measured | Each factor is a raster: distance surfaces, slope, land cover | Each factor is a column: distance to nearest road, zonal mean of slope, joined attributes |
| Best for | Continuous surfaces, regional screens, environmental factors | Decisions about real properties, ownership, anything with an address |
| Output | A suitability surface | A ranked table you can sort and export |
For most real estate and planning decisions, the parcel version is more useful. Nobody buys a 30 m cell. A common hybrid is to compute raster factors (slope, heat, land cover) and summarise them to each parcel with zonal statistics, then score the parcels.
Composite scores on tracts are the same idea. In where does Miami flood first, NeuraCities joined FEMA flood zones to Miami-Dade census tracts, measured distance to the county's 25 cooling centers and 48 evacuation routes, and combined them into one vulnerability score per area.
Validate the result and test its sensitivity
A suitability map always looks authoritative. That is the danger. Two checks separate a defensible result from a pretty one.
Validation: does it agree with what you know?
- Look at the top ten places one by one on a satellite basemap. Are any obviously wrong (a cemetery, a highway interchange, a lake)? That usually exposes a bad input layer or a missed constraint.
- Check places you already know are good or bad. Existing successful sites should score well; known failures should not.
- Look at the score distribution. If almost everything lands in a narrow band, the reclassification is not discriminating.
Sensitivity: does the answer depend on one arbitrary weight?
Change each weight up and down by a modest amount and rerun. If the same sites stay at the top, the result is robust and you can say so. If the top list reshuffles completely when one weight moves a little, the ranking is telling you more about the weights than about the places, and the decision-makers should know that before they rely on it.
Common weighted overlay mistakes
- Double counting correlated factors. Median income, poverty rate and share of renters often move together. Include all three at 20% each and you have quietly given one idea 60% of the score. Pick one, or combine them into a single index first.
- Mixing hard constraints into the score. Giving the floodway a score of 1 instead of excluding it means a site in the floodway can still win if it scores well on everything else. If it is a no, make it a constraint.
- Unstated weights. A map with no weights listed cannot be reviewed or reproduced. Put them in the legend, the report and the question that produced the map.
- Inconsistent direction. One factor scaled so high is good and another so high is bad will cancel each other out without anyone noticing.
- Scoring before checking coverage. A factor layer that only covers half the study area scores the other half as if it were the worst possible. Check every input covers the whole area first.
- Treating the winner as the answer. A suitability analysis is a screen. Site visits, title, environmental review and engineering still decide.
How to do suitability analysis without ArcGIS, in NeuraCities
Esri's Weighted Overlay tool requires the Spatial Analyst extension at every ArcGIS Pro license level (Esri, Weighted Overlay tool). And before any tool runs, you still need to find and prepare every input layer, which is usually the slow part.
In NeuraCities, you describe the constraints, factors and weights in one sentence:
In Austin, TX, find parcels of at least 5 acres outside the FEMA 100-year floodplain and not zoned residential. Score them on distance to the nearest highway (40%), distance to the nearest transit stop (30%) and land value, lower is better (30%). Show the top 25 on a map with a table of the scores.
- The orchestrator writes a checklist that separates the filters from the scored criteria and lists the weights, so you can correct its reading before anything runs.
- The Data Agent finds the parcels, zoning, FEMA flood layer, roads and transit stops from its registry of 600+ open data portals, federal sources and OpenStreetMap, and probes each one for coverage before loading it.
- The GIS agent applies the constraints with standard filters and overlays, measures each factor, normalises the factors and combines them with a weighted overlay using your weights (or equal weights if you give none).
- You get a map styled by score, a ranked table that exports to Excel, charts and a written summary with the criteria as applied and the sources used.
Before scoring, check whether the flood study covers every candidate. Use this flood-risk screening workflow to keep coverage gaps out of the confirmed shortlist.
Sensitivity testing becomes a follow-up question: "Rerun with highway at 30% and transit at 40%. Which of the top 25 drop out?" Because each step is deterministic GIS code, changing a weight changes only the ranking, not the inputs. If a criterion has no public data, the answer says so and you can upload your own layer.
If you would rather click than type, GIS Studio in NeuraCities includes weighted overlay, reclassification, zonal statistics and buffers as manual operations on the same data. When the method is settled, save the prompt as a workflow with {{city}} and weight variables, so the next run uses exactly the same criteria.
Frequently asked questions
What is a weighted overlay in GIS?
A weighted overlay combines several map layers into one suitability score. Each layer is reclassified to a common scale, multiplied by a weight that reflects its importance, and the results are summed for every cell or feature. Higher scores mean more suitable places.
What is the difference between a constraint and a factor in suitability analysis?
A constraint rules places out entirely, such as land inside a floodway or a parcel below a minimum size. A factor measures how good the remaining places are, such as distance to a road. Constraints should filter places out before scoring; factors are combined in the weighted score.
How do you choose weights for a weighted overlay?
Weights reflect priorities, so they are a judgement made by the people the analysis serves. Common approaches are equal weights, stakeholder point allocation and the Analytic Hierarchy Process (AHP), which derives weights from pairwise comparisons. Whatever you choose, state the weights and test how sensitive the result is to them.
Can I do suitability analysis without ArcGIS?
Yes. The arithmetic is the same anywhere. In NeuraCities you describe the constraints, factors and weights in plain English, and the agents find the data, apply the filters, run a weighted overlay and return a ranked, mapped result with sources. GIS Studio also offers weighted overlay as a manual operation.
Should I use raster or vector data for site suitability?
Raster suits continuous surfaces and regional screens; vector suits decisions about real parcels. Many analyses compute raster factors such as slope, then summarise them to each parcel with zonal statistics and score the parcels.
How do I know if my suitability analysis is right?
Check the top results on a basemap for obvious errors, confirm that known good and bad sites score as expected, and run a sensitivity test by changing the weights. If the top sites hold up across reasonable weightings, the result is robust.
More guides on site selection and land
Start with a study area, then specify the exclusions, factors and weights.
Run a suitability analysis