About PipeHK

How the network is joined

What the Drainage Services Department publishes is a pile of individual conduits, each recording the id of the node it runs from and the one it runs to, but not a network. PipeHK joins the 135,206 raw records end to end, drops the ones that cannot be connected or that contradict themselves, and arrives at 132,654 usable conduits. Those are then topologically sorted — arranged so that every conduit comes before the one it drains into. Routing water from a hillside to an outfall is one pass along that order.

Flow direction is not inferred. Roughly 99% of conduits arrive with it already recorded, as a from-node and a to-node. That is the fact the whole project rests on: without direction these are lines on a map, not a network.

Where the rain comes from

The map follows the Hong Kong Observatory's rainfall nowcast, refreshed every 12 minutes and reaching about two hours ahead. Rain lands first on catchments — derived from the government's 5 m digital terrain model, which is what decides where the water falling on a given patch of ground ends up — and is then accumulated downstream through the network. Whatever an upstream conduit receives is added to the load of every conduit below it.

Besides following the forecast you can pick one of three rainstorm scenarios — amber, red or black — which lay a single uniform intensity over the whole territory, or use the rain brush to paint a storm wherever you like on the map and watch where the water goes.

Capacity and stress

Each conduit's capacity is estimated with Manning's equation, which needs three things: bore, gradient, and roughness. Bore is published for everything. Gradient comes from invert levels, and invert levels are not published for every conduit — where they exist the surveyed value is used, where they do not the gradient is interpolated from neighbouring conduits, and failing that it is assumed to be 1:200.

This matters enough to be printed on every conduit's panel: open one and it tells you whether that conduit's gradient was surveyed, interpolated, or assumed. Where the capacity is an assumption, so is the stress figure derived from it.

Tracing a conduit to the sea

Tap any conduit and the whole route from that point lights up, all the way to its outfall, listing the manholes, catchpits, sand traps and drop shafts it passes through and how long the water takes to reach the sea. Water splits at junctions: the lit route is the largest-bore line through each one, and the faint lines are everywhere else it can reach.

Travel time assumes every conduit runs full bore, which is the fastest case. Real conduits rarely run full, so the real journey is longer than the figure shown.

Where street water actually enters

The inlets the Drainage Services Department publishes are a small fraction of where water actually gets in. What collects street runoff is the Highways Department's road gullies — the metal grate beside your foot at a pedestrian crossing — and there are 280,918 of them, nineteen times the entry points DSD lists.

Zoom in to street level — level 15 and above — and the gullies are drawn as small hollow rings. They sit on the surface rather than joined to the network, because no published record says which drain any one of them discharges into: the short lateral from grating to pipe is not part of the record. A ring marks where a gully is, and claims nothing about what it connects to. Click one and the panel lists everything the record holds, which is not much.

Folding the gullies in makes the catchment model considerably better, and that work is still in progress — which is why the current catchment binding is marked interim.

What this map cannot tell you

There is no public telemetry inside Hong Kong's stormwater pipes — none at all. Every number on this map is therefore a modelled load derived from rainfall, not a measured flow. How much water is actually in a given pipe right now is not something anyone here knows.

It is also not a flood warning and must not be used for emergency planning. The original idea was to predict flooding blackspots, but only five are officially recognised — five examples will not train anything. So that part became hydraulic capacity screening instead: pointing out which conduits sit at a high load ratio under a given rainfall, rather than claiming where water will pool. The first follows from the data. The second does not.

The data has its own gaps. Of the 132,654 conduits, 21 have a recorded position that contradicts their own surveyed line — the two endpoints sit kilometres from the pipe drawn between them. Those are reused feature numbers in the records rather than real pipes, and a trace that reaches one stops and says so. Elsewhere, some dead ends were given a downstream link that the build inferred rather than found, and the route panel marks each inferred leg as it goes.

Sources

The network comes from the Drainage Services Department via the Common Spatial Data Infrastructure; the road gullies from the Highways Department via DATA.GOV.HK; the rainfall from the Hong Kong Observatory's open data; and the terrain from the government's 5 m digital terrain model. All used under their respective open data terms.

PipeHK is not a government website and is not affiliated with the Drainage Services Department, the Highways Department, or the Hong Kong Observatory.