What a drainage survey actually measures

This page sets out all four, how each is computed, and — the part that matters more — the limit that belongs beside each one. Use it to interrogate whatever you are about to pay for, ours included.

1. Ground slope, stated per wall

What it is. The gradient of the ground moving away from a wall, over the first few meters. The usual target is a 5% fall over 3 m — about 15 cm of drop.

Why per wall. A single figure for a property averages away the only interesting case. Each wall faces a different way and sits on different ground; a house can be textbook on three elevations and a bowl on the fourth.

How it should be computed. Along each wall's own outward normal. Measure measures the slope from 0.75 m to 3.25 m out, repeated at several stations along the wall and combined. Nothing is sampled nearer than 0.5 m to the wall, because inside a footprint the bare-earth surface is not ground — it is the interpolation across a building that was removed.

The limit. Nearby readings are not independent observations, so the error should be computed on the information actually present. And a building outline is a trace that can sit a meter from the wall it represents. On a real Pittsburgh footprint, shifting it north in 0.2 m steps moved one wall from +5.4% to −23.5%, crossing zero inside 1.4 m. A single-position reading printed to one decimal is a confident guess.

2. Contributing area, per wall

What it is. The square meters of upstream ground that drain to a given wall. It is the quantity that separates a nuisance from a scope of work.

Why it matters more than people expect. Two walls at an identical −6% can be different jobs by an order of magnitude: one collects a side yard, the other collects a hillside — frequently including land outside the property line.

How it should be reported. Sampled at several points beside the wall, re-sampled at every displaced footprint position, and published with its range and the distribution it came from — never as a bare maximum, and rounded to the precision it actually has. A single large number with no range is a sales figure.

3. Modeled flow paths

What it is. Where runoff concentrates as it crosses the site. Not steepness — direction and accumulation.

How it should be computed. Water routed downhill and accumulated, spreading the way overland flow actually behaves — near-single-channel in a gully and dispersive across a lawn. The cheap shortcut forces all the water into one channel and manufactures a hard line across ground where flow really spreads.

Buildings must stand as walls in the model, so water routes around structures rather than through them.

The independent check. Where a city digs its storm drains is a yardstick nobody in this building tuned. Against Pittsburgh Water's published inlets, the 73 inlets in the frame fall within 10 m of a modeled channel 3.99× as often as the frame's own base rate at the 260 m radius the product ships — the control being the share of the frame that sits that close to a channel anyway.

The limit on that check, which is not small. Pittsburgh is the only municipality that publishes its inlets in a form we can consume, so it is a one-city check. Published inlet positions carry one to two meters of georeferencing error of their own. And an inlet sits where a crew could dig, which is not always where the water goes. Agreement is evidence the routing is sane, not proof it is right.

4. Standing water: depth and extent

What it is. Where water collects with no downhill exit, and how deep it gets before it spills.

How it should be computed. Every closed depression filled to its spill point, so no water is left with nowhere to go, and the fill depth kept rather than discarded. Reported within 20 m of the footprint and outside every building. Under 3 cm is measurement noise and should not be counted.

Why the step is mandatory. On real ground, 282,572 m² of contributing area in Chicago and 201,312 m² in Pittsburgh drained into interior pits two to four millimeters deep and stopped, against largest surviving channels of 1,034 m² and 33,931 m². Water spills over a 2 mm lip. A model that has not been conditioned is reporting garbage in all four quantities, not just this one.

The arithmetic check. On synthetic ground where the answer is known, a bowl cut into a 2% slope fills to its spill point at 0.28 m predicted against 0.289 m measured.

The fifth thing, which is the one to ask for

A good survey names what it declined to answer. Walls whose direction did not survive the uncertainty pass. Walls with no readable profile. Addresses whose ground data failed the quality gate at all — Measure requires the median step between adjacent elevations to clear 0.004 m, and refuses below it rather than interpolating.

The quantity declined should travel in the payload beside the ones that resolved. A report with no gaps in it has either measured a very simple property or hidden something.

And before all four: which building?

The best drainage model in the world is worth nothing pointed at the neighbor's roof. Identity is decided by graded evidence, strongest first — and where the nearest outline is all there is, that is a guess and it is reported as one.

The grade of evidence that answered ships with the result, so an identification and a location are never mistaken for each other, and where nothing identifies a structure the address is refused rather than measured on the nearest roof. Where the underlying records are thin the evidence is thin too, and the survey says so rather than guessing harder.