Watershed Analysis Explained

A raw DEM is full of small closed depressions — cells with no lower neighbour. A few are real (quarries, karst, ponds), but most are artefacts of interpolation, noise or vegetation removal.

If you route water on the raw surface it gets trapped in these pits and the drainage network falls apart into disconnected fragments. So the first step fills or breaches them until every cell has a downhill path to the edge of the grid.

This matters more than it sounds: pit filling is the step that decides whether your channels connect. It also means the surface used for hydrology is not quite the surface you measured — deliberately.

Next, each cell is assigned a direction that water leaves it. The classic method is D8: water goes to whichever of the eight neighbours is steepest downhill — all of it, in one direction.

D8 is simple and gives crisp channels, but on flat and gently convex ground it forces water into artificial straight lines, because a single direction has to be chosen even when the real flow spreads.

Multiple-flow methods split the outflow between all downhill neighbours in proportion to slope. The result is more realistic on hillslopes and alluvial fans, and produces smoother, more believable accumulation patterns.

STREAM fills pits, then computes flow with a distributive multiple-flow method, and derives stream order from the result.

Flow accumulation counts, for every cell, how much upstream area drains through it. Ridge cells have almost none; valley cells have a lot.

This single grid is where channels come from. Pick a threshold — say, every cell with more than N upstream cells — and you have a drainage network. Lower the threshold and the network grows finer and reaches further up the slopes.

There is no universally correct threshold. It is a modelling choice about what counts as a channel in your landscape, and it should be stated alongside any map you hand over.

Once you have a network, Strahler ordering classifies it. Headwater segments with no tributaries are order 1. Where two segments of equal order meet, the order increases by one; where unequal orders meet, the larger is kept.

It is a compact way to say how far down the drainage hierarchy a given channel sits, which is why it is used for filtering ("show me order 3 and above") and for comparing catchments.

Finally, pick an outlet — a culvert, a gauge, a pit sump — and trace every cell whose water eventually passes through it. That set of cells is the watershed, and its outline is the catchment boundary.

Move the outlet a few cells and the boundary can change noticeably, especially in flat terrain. Snap the outlet onto the accumulated channel before delineating; an outlet sitting one cell off the channel drains a tiny sliver instead of the whole basin.

Resolution: a coarse grid smooths small divides and can merge neighbouring catchments; a very fine grid captures every kerb and windrow, including ones that will be gone next week.

Surface type: hydrology on a DSM routes water over building roofs and tree canopies. For drainage you almost always want bare-earth terrain.

Site works: on an active site, temporary bunds, haul roads and stockpiles genuinely redirect water. That is not error — it is why the analysis has to be re-run as the site changes.

Sizing drainage: knowing the contributing area upstream of a culvert or channel is the input to almost every runoff calculation.

Site water management: finding where water will pond on a graded platform, and which way a spill would travel.

Mining and quarrying: routing water away from working faces, planning sumps, and understanding what a pit intercepts.

Erosion and sediment: high-accumulation paths on steep ground are where you will lose material first.