Multi-Criteria Site Suitability with AHP

In Design with Nature (1969), Ian McHarg drew each site factor - slope, drainage, soils, vegetation, scenic value - on its own transparent sheet, shading the least suitable areas darkest. Stack the sheets on a light table and the darkest composite was the worst place to build; the lightest, the best. It was ecological planning made visible, and it is the direct ancestor of everything in this lesson.

GIS does exactly this, but with numbers instead of ink, so it can weight the sheets unequally and compute the composite precisely across millions of cells. The logic is unchanged; the rigour and the reproducibility are new. Naming the lineage matters: suitability analysis is not a software trick, it is a fifty-year-old design philosophy that a raster engine finally made exact.

First, list the factors that genuinely bear on the decision - and only those. Padding a model with weakly relevant layers dilutes the ones that matter. For each factor, decide its direction (is steeper worse? is nearer a road better?) and reclassify it (Module 5.1) onto one shared suitability scale - commonly 1 (least suitable) to 5, or a simple 1-2-3.

This is where honesty lives. The class breaks encode real judgement and, where possible, real standards: a maximum buildable slope, a statutory flood level, a walkable distance to transit. Also separate constraints from factors: some things are absolute vetoes (a legal no-build zone, water body, protected area) and should be masked out to zero regardless of how good everything else is, rather than being allowed to average away.

Not all factors matter equally, so each gets a weight, and the weights must sum to 1. The naive approach is to assign weights by gut feel, which is neither defensible nor reproducible. The Analytic Hierarchy Process (AHP), developed by Thomas Saaty, fixes this.

Instead of rating all factors at once, AHP asks you to compare them two at a time on a 1-to-9 scale: is flood risk equally, moderately, strongly, or extremely more important than slope? Those pairwise judgements fill a matrix, and its principal eigenvector yields a set of weights - a priority vector - that is mathematically consistent with your comparisons. Comparing pairs is far easier and more honest than conjuring ten weights that sum to 1 out of thin air.

AHP's quiet genius is that it can catch you contradicting yourself. If you say slope is more important than flood, and flood more important than access, but then rate access above slope, your judgements are inconsistent. AHP measures this with a Consistency Ratio (CR), comparing your matrix to what a perfectly consistent one would look like. Saaty's rule of thumb: keep CR below 0.10. Above that, revisit the pairwise comparisons - somewhere your judgements disagree with each other.

This check is why AHP is trusted in planning and site-selection literature: it does not make your values right, but it proves they are internally coherent, and it produces a transparent audit trail from judgement to weight to map. That is the difference between a suitability map you can defend at a public hearing and one you cannot.

Now combine. Weighted overlay multiplies each reclassified criterion raster by its AHP weight and sums them cell by cell - a local map-algebra operation (Module 5.1). The output is a continuous suitability score for every cell, which you reclassify into ranked bands (most suitable to least) for a clean decision map, with constraint areas masked to zero.

Then do the step everyone skips: sensitivity analysis. Nudge the weights up and down and watch whether the top sites stay top. If a small change in one weight reshuffles the answer, your result is fragile and you should say so. A suitability map is a structured argument, not an oracle - its honesty comes from stating the criteria, the breaks, the weights, the CR and how sensitive the outcome is. Present all of it, and let the decision-maker own the value judgements.

One last discipline: watch for correlated criteria. If slope, drainage and soil all really encode the same underlying landform, giving each a full weight silently triple-counts that factor and skews the result. Pick criteria that are as independent as possible, and where two overlap, either merge them or split the weight. The method rewards clear thinking about what genuinely, separately matters - which is, in the end, the designer's own contribution to the arithmetic.