Catchment

Watershed delineation online, from any outlet

Click an outlet on the map and Delta Hydro Studio returns the catchment boundary, area, Strahler-ordered stream network and longest flowpath, free with a Studio account. It covers land between 60°S and 85°N on MERIT-Hydro, runs at 30 m on a hydrologically corrected Copernicus DEM in South Africa, and can call USGS StreamStats or NLDI for US outlets.

How DEM-based watershed delineation works

A watershed (catchment) is every piece of ground that drains through one outlet, the pour point. Automated delineation finds it from a digital elevation model (DEM) in five steps.

  1. Condition the DEM. Raw elevation data has pits: single cells lower than all their neighbours, usually noise, which would trap flow. Conditioning fills them, resolves flat areas and carves channels so that every cell has a downhill path to the sea or a genuine closed basin. MERIT-Hydro also removes the tree-canopy and building bias that radar DEMs carry.
  2. Assign a flow direction to every cell. The D8 method (Jenson and Domingue, 1988) sends each cell’s water to the one of its eight neighbours with the steepest drop.
  3. Accumulate flow. Each cell counts how many cells drain through it. Channels are the cells where that count is large.
  4. Snap the outlet. Your click is moved onto the nearest high-accumulation cell, because an outlet one cell off the channel traces a hillslope instead of the river.
  5. Trace upstream. Every cell whose flow path passes through the outlet is collected and its outline becomes the watershed boundary.

The D8 rule in step 2 is:

Sk=z0−zkdk,flow goes to max⁡kSkS_k = \frac{z_0 - z_k}{d_k}, \qquad \text{flow goes to } \max_k S_k
SymbolMeaningUnit
z0z_0Elevation of the cell being processedm
zkz_kElevation of neighbour kk (eight neighbours)m
dkd_kDistance to neighbour kk: one cell width for the four sides, 2\sqrt{2} cell widths for the diagonalsm
SkS_kDrop per unit distance towards neighbour kkm/m

The Global Watersheds service that Studio calls for most of the world (Heberger, 2022) speeds this up for large basins: it unions precomputed MERIT-Basins unit catchments upstream and runs the raster trace only inside the unit catchment that holds your outlet. That split is what the precision setting controls.

Snapping the outlet: where most bad delineations start

A click “on” a stream is accurate to tens of metres at best. With the COP30 engine in South Africa, Studio looks up the COP30 channels within 120 m of your click and offers them, each labelled by the area already draining to it, its distance from the click and its stream order. Where several channels meet near the click, all of them are listed. Pick the one you are designing for and the outlet moves onto it, with the engine snap distance set to 0 m; or keep your point as placed. The outlet never moves without your choice. The Global Watersheds and USGS sources snap with their own fixed rules.

Confluences are the trap. An engine snaps to the highest flow accumulation within its snap distance, so near a junction it reaches across onto the bigger branch. Measured at one South African confluence, on a tributary whose true catchment is 57 km², clicks 15 m, 30 m, 50 m, 80 m and 150 m upstream of the junction all returned 114 km², which is the neighbouring river. The tributary only delineated as itself past 300 m. The answer looks plausible, which is why it gets into reports.

The COP30 engine lets you set its snap distance:

SettingUse when
0 mThe outlet is a surveyed structure position and must be honoured exactly
20–30 mThe outlet already sits on the channel you chose
150 m (default)The outlet was placed by eye, away from any confluence
Up to 500 mThe outlet is rough and the nearest channel is uncertain

A larger snap distance is not safer. It is the setting that lets the engine cross a confluence.

Choosing a delineation source

SourceCoverageTerrainNotes
Delta Hydro COP30South AfricaCopernicus GLO-30, 30 m, hydrologically correctedReturns the longest-flowpath profile, elevation statistics and the downstream flowpath with the boundary
Global WatershedsLand, 60°S to 85°NMERIT-Hydro, about 90 mThe usual choice outside South Africa; high and low precision
Esri ArcGIS WatershedWorldwideHydroSHEDSIndependent cross-check
USGS StreamStatsUnited StatesNational datasetsAdds basin characteristics: impervious and forested percentage, mean annual precipitation, curve number
USGS NLDIUnited StatesNHDPlus networkNavigation on the national stream network

US-only sources are disabled outside the US and COP30 is disabled outside South Africa. For a 1 km² culvert catchment in South Africa, COP30’s 30 m grid gives about nine times as many cells as MERIT-Hydro’s 90 m grid.

Worked example: six South African gauges

To check the engine against published areas, the Global Watersheds service was run on 8 October 2026 at the listed coordinates of six Department of Water and Sanitation (DWS) flow gauges, at both precision settings. The DWS catchment areas come from the station list used by the Stream Gauge Finder.

GaugeRiverDWS area (km²)High precision (km²)Low precision (km²)
U2H006Karkloof @ Shafton339337387
U2H013Mgeni @ Petrus Stroom299295300
C5H007Renoster Spruit @ Shannon Valley348338343
T5H004Mzimkulu @ FP 1609030545538605
K6H001Keurbooms @ M’Kama165162200
G4H007Palmiet @ Welgemoed464464470

High precision landed within 3 % of every published area, and five of six within 2 %. Low precision overstated three of them by 11 % to 21 %, because it returns the whole unit catchment containing the outlet, including ground downstream of the gauge. Use high precision for anything you will size a structure from. Basins over 50 000 km² revert to low precision automatically.

What comes back

What a run returns depends on the source. Every source gives the boundary and area. Global Watersheds, USGS NLDI and COP30 also give a longest watercourse and a stream network (COP30 where the service returns its channels); Esri and StreamStats give the boundary only, and the Esri longest watercourse is traced afterwards from a headwater pin. Across the sources and the follow-up analyses, the result can hold:

  • Area, perimeter and the snapped outlet coordinates, which can differ slightly from your click.
  • Stream network with Strahler order (Strahler, 1957). Order 1 is a headwater stream; two order-1 streams make an order 2, and a lower order joining a higher one does not raise it. Each exported segment carries its order.
  • Longest watercourse, the route from the most remote source to the outlet, with its length. Open it in Watercourse Profile for the long section and the 10-85, equal-area and Taylor-Schwarz slopes.
  • Downstream flowpath from the outlet to the sea, with its length.
  • Shape indices: maximum basin length, elongation ratio, form factor and circularity ratio, which indicate how fast runoff concentrates.
  • Elevation statistics: minimum, maximum, mean, relief and mean slope of the catchment.
  • Drainage density and stream frequency for South African catchments, measured on the 1:50 000 mapped channel network of the national topographic series.

Drainage density is total channel length divided by catchment area (Horton, 1932). The Studio documentation reads it as follows:

Drainage density (km/km²)TextureTypical setting
< 0.5CoarsePermeable substrate, low relief
0.5–2.0MediumMost South African catchments
2.0–4.0FineImpermeable substrate, steeper relief, sparse vegetation
> 4.0Very fineBadlands, highly dissected terrain

Outside South Africa the value is left blank, never reported as zero: a missing measurement is not a catchment without channels.

Limitations and when to use something else

  • Resolution. At about 90 m, MERIT-Hydro does not resolve catchments under roughly 1 km², narrow valleys or terraces. Use COP30 in South Africa or a local LiDAR DEM elsewhere.
  • Flat terrain. On coastal plains, pans and deltas, D8 directions follow DEM noise. Check the result against imagery.
  • Engineered drainage. Stormwater pipes, kerb inlets, canals and road cross-drainage are not in any DEM. In urban and peri-urban work, overlay the municipal stormwater layout and adjust the boundary by hand.
  • Endorheic basins. Conditioning can force a closed basin to drain to the sea. Check known closed systems.
  • Estuaries. The tidal channel on imagery may not match the flow grid. Move the outlet 100–500 m upstream to the freshwater channel.
  • Service availability. Global Watersheds is an external service. If it is down, switch to the Esri source.

If you already have a catchment boundary from survey or another GIS, upload it as GeoJSON, KML, KMZ, a zipped Shapefile, GeoPackage or DXF instead of delineating.

What the full Studio tool adds

Watershed delineation is free with a Studio account. Results save to a project and export as GeoJSON, Shapefile or KML, as a Word or Excel report, and as a SWMM 5 input file with every assumed parameter listed for review. The catchment feeds straight into the next steps:

FAQ

Common questions

How do I delineate a watershed from a point?
Click the outlet on the map or type its latitude and longitude in decimal degrees. Pick a delineation source, check where the outlet snapped, and run it. The tool traces every cell that drains to that point on a conditioned elevation model and returns the boundary, area, perimeter, stream network and longest flowpath. Move the outlet and re-run until the area matches the crossing or structure you are designing.
Is there a StreamStats alternative outside the United States?
USGS StreamStats only covers the United States. Outside it, Studio uses the Global Watersheds service built on MERIT-Hydro, which covers land from 60°S to 85°N at about 90 m, plus an Esri delineation built on HydroSHEDS as a cross-check. In South Africa the Delta Hydro COP30 engine delineates on a 30 m hydrologically corrected Copernicus GLO-30 DEM. StreamStats' regional regression flow statistics are US-only; in South Africa, Studio's regional flood methods are separate tools.
How accurate is automated watershed delineation?
For six South African flow gauges between 165 and 545 square kilometres, Global Watersheds at high precision returned areas within 3 percent of the published DWS catchment areas. Low precision was out by up to 21 percent on small catchments, because it uses the whole unit catchment that contains the outlet. Errors grow in flat terrain, small catchments under about 1 square kilometre and urban areas.
Why did my watershed come out twice the size I expected?
The usual cause is a confluence. A delineation engine snaps the outlet to the cell with the highest flow accumulation within its snap distance, so a click on a small tributary near a junction can jump onto the larger river. Put the outlet on the branch you mean, then cut the engine snap distance to 20 to 30 metres so it cannot move back.
What DEM resolution do I need for a small catchment?
MERIT-Hydro cells are about 90 m, so a 1 square kilometre catchment is only around 120 cells and its divide can move by a cell width. Use the 30 m COP30 engine in South Africa, or a local LiDAR DEM in your own GIS, for culvert catchments under a few square kilometres. For regional and river-scale work, 90 m conditioned data is adequate.
Can I export the watershed to GIS or SWMM?
Yes. The boundary, stream network and flowpaths export as GeoJSON, zipped Shapefile or KML in WGS84. A SWMM 5 input file writes the catchment as one subcatchment draining to an outfall, projected to the outlet's UTM zone. Every SWMM value the delineation cannot supply, such as imperviousness and curve number, is listed at the top of the file for you to replace.
Full documentation: Watershed Delineation
Every input, option and equation in the Studio tool, with its limitations.
Run it from your AI agent
This engine is open to Claude, ChatGPT, Cursor and other MCP clients as snap_to_channel and delineate_watershed. The agent passes the inputs, the engine does the calculation, and the record is saved to your project for you to review.

References

  1. Yamazaki, D., Ikeshima, D., Sosa, J., Bates, P.D., Allen, G.H. and Pavelsky, T.M. (2019) MERIT Hydro: a high-resolution global hydrography map based on latest topography datasets. Water Resources Research, 55(6), 5053-5073.
  2. Heberger, M. (2022) Delineator: fast, accurate global watershed delineation using hybrid vector/raster methods. Zenodo, doi:10.5281/zenodo.7314287.
  3. Jenson, S.K. and Domingue, J.O. (1988) Extracting topographic structure from digital elevation data for geographic information system analysis. Photogrammetric Engineering and Remote Sensing, 54(11), 1593-1600.
  4. Strahler, A.N. (1957) Quantitative analysis of watershed geomorphology. Transactions, American Geophysical Union, 38(6), 913-920.
  5. Horton, R.E. (1932) Drainage-basin characteristics. Transactions, American Geophysical Union, 13(1), 350-361.
  6. Lehner, B., Verdin, K. and Jarvis, A. (2008) New global hydrography derived from spaceborne elevation data. Eos, Transactions American Geophysical Union, 89(10), 93-94.
  7. European Space Agency and Airbus (2022) Copernicus DEM GLO-30: Global 30 m Digital Elevation Model.

Last reviewed by the Delta Hydro engineering team.

Keep the whole study in one place

Delta Hydro Studio puts this tool alongside catchment, rainfall, runoff and hydraulics tools, with projects, team sharing and Word and Excel reports. Run them by hand, or let your AI agent run them for you to review.