A site survey assembled entirely from public data

What can you know about a plot before visiting it?

Far more than most briefs assume — and most of it is free.

A conventional early-stage site analysis leans heavily on assumption: a rough sense of orientation, a regional climate impression, and whatever the client remembers about drainage. Meanwhile the measured record for that exact coordinate is sitting in public APIs, and much of it needs no key at all.

This is not a replacement for a site visit. It is a replacement for guessing before one.

Why sample elevation nine times instead of once?

One reading gives you a height. Nine give you a slope, a direction and a drainage problem.

A single elevation point tells you almost nothing useful for design. Sampling across the plot — corners, edges and centre — yields the level difference, the slope percentage and, critically, the direction of fall.

That last one is the design input. Fall direction determines where water goes, which determines plinth height, cut-and-fill strategy, foundation stepping and where you must not put a basement. A number without its direction is trivia.

Which climate figures actually change a drawing?

Design extremes and prevailing wind — not annual averages.

Mean annual temperature is nearly useless for design. What matters is the 1% extreme figures, the diurnal swing, the number of days above 35 and 40 degrees, the wettest month, the heaviest recorded day, and the longest wet and dry spells.

The diurnal swing in particular is a direct instruction about thermal mass: a large day–night difference makes heavy construction work in your favour, and a small one makes it a liability. The prevailing wind direction by season tells you which side the openable windows belong on — and it changes between seasons, which a single annual wind rose hides.

• Design temperatures at the 1% extremes, not the annual mean.
• Diurnal swing, because it decides whether thermal mass helps or hurts.
• Monthly rainfall, wettest month, and the heaviest single recorded day.
• Longest wet and dry spells, which drive storage and drainage sizing.
• Wind rose by season, because the useful direction is not the same in summer and monsoon.

Can you get real flood history for a specific point?

Yes — roughly three decades of daily reanalysis, which is enough to compute return levels.

This is the source most teams do not know exists, and it is the one that changes decisions most sharply. A daily flood reanalysis stretching back to 1998 gives you the actual largest events at that coordinate with their dates and magnitudes, recurrence intervals, and design return levels at the 1-in-2, 5, 10, 25 and 50-year marks.

It also tells you which months floods historically peak in, and whether peaks are rising or falling across the record — a trend that matters far more than a single worst-case figure when you are setting a plinth.

Worth noting what is not available: some flood-forecasting APIs are allowlisted and return permission errors, and at least one major weather API retains only 24 hours of history, which is useless for design work despite looking authoritative. Checking what a source actually returns before building on it saved us from both.

What do the neighbours have to do with it?

They decide your midwinter light and your overlooking, and they are mapped.

Neighbouring building footprints and heights, with distance and bearing, are enough to compute which sides of the plot are built up, what overshadows it at midwinter when the sun is lowest, and where overlooking comes from.

Midwinter is the case that matters. A courtyard that works beautifully in June can be in permanent shade in December, and that is knowable from open map data before a single line is drawn.

What is the actual hard part?

Not fetching the numbers. Attaching each one to its consequence.

A dashboard of site statistics is not useful to an architect. A 9.9 degree diurnal swing is not an insight; 'this swing makes thermal mass work for you, so consider heavier construction on the west' is. A river flashiness figure is not an insight; 'raise the plinth and do not put habitable space below grade' is.

Every figure needs to carry its architectural consequence or it is decoration. That translation — from measurement to instruction — is where the engineering effort actually goes, and it is the part a generic data integration will not do for you.

Where should assumption still be visible?

Everywhere it remains, and it should be impossible to miss.

Some inputs cannot be measured remotely: soil bearing capacity, statutory setbacks in an unpublished local by-law, the client's actual budget. These have to be assumed to proceed, and the assumption must remain visible until confirmed.

The interface convention we settled on is to reserve one colour exclusively for 'assumed, needs confirmation', so that the colour always carries exactly one meaning. Anything statutory that has been guessed rather than verified stays flagged until a human confirms it. An assumption that quietly becomes indistinguishable from a measurement is how a drawing set becomes dangerous.