20. 2D overland flow and 1D-2D interfaces

SWMM routes water through a network of nodes and links. What happens when a manhole surcharges and water runs down the street, pools in a car park, and finds its way back into a downstream inlet is not something the 1D network can tell you: the engine reports the flooded volume at the node and loses it (or, with Aponded, stores it in a column of air). The 2D menu adds a surface to the model — a grid of cells on a digital elevation model (DEM) — solves the shallow-water equations on it, and couples the two so that surcharge leaves the network at the nodes and bank lines, flows over the ground, and is captured again where the surface meets an inlet.

The equations, the numerical scheme, the wet/dry treatment and the tests that show the solver reproduces known solutions are in 20b. This chapter is the user's side: what to load, what to set, how to run, and what the map shows afterwards.

Everything 2D lives outside the .inp. The settings are kept in a sidecar text file <model>.2d beside <model>.inp, and the results go to <model>.2d.out. The .inp stays EPA's format, so the model still opens in the EPA GUI and runs on any engine; the sidecar is simply ignored by software that does not know it. AquaSWMM never writes anything 2D into the .inp.

20.1 Workflow

DEM. Get a DEM of the site in the model's map units and coordinate system, as an ESRI ASCII grid (.asc) or a GeoTIFF (.tif). Chapter 19 covers coordinate systems and where DEMs come from.
Setup. 2D → 2D Setup…: point at the DEM, choose the cell size and window, roughness, rain on the grid, boundaries and time stepping. Save writes the sidecar.
Interfaces. 2D → Interfaces…: say which nodes exchange water with the surface (manhole, inlet, sealed), and draw bank lines along open channels. Sources… adds point inflows that are not nodes.
Run. 2D → Run 2D Only runs the surface by itself (rain on grid and sources only), or Run Coupled (1D-2D)… runs the network and the surface together.
Results. The depth, velocity and hazard grids are painted on the map under the network, driven by the same time slider as the SWMM results; the interface markers show which nodes are surcharging and which are capturing; the grids export as ASCII rasters for a GIS.

Item	What it does
2D Setup…	The setup window (§20.3).
Interfaces…	Node interfaces and bank lines (§20.4).
Sources…	Point sources (§20.5).
Run 2D Only	Runs the surface alone (§20.6).
Run Coupled (1D-2D)…	Picks the engine and the coupling mode, then runs both.
Stop 2D	Asks a running 2D or coupled run to stop at its next progress report.
Load 2D Results	Opens `<model>.2d.out` beside the model (a finished run loads it by itself).
Export Max Depth Grid…	Writes the run's maximum depth as an ESRI ASCII grid.
Export Frame Grid…	Writes the depth of the frame the overlay is showing.
Export Hazard Grid…	Writes the run's maximum depth × velocity.
Draw Profile Line	Click the vertices of a section line on the map; Enter or a double-click finishes and opens the profile of the ground and 2D water along it (§11.7). Escape cancels.
Clear Profile Line	Removes the section line; the profile goes back to the pipes.
2D Legend	Shows or hides the 2D legend in the map's corner.
Show Interfaces	Shows or hides the interface markers and bank lines on the map.

Every item needs a model open. The run items need the model saved to a file: the run reads <model>.inp from disk and writes <model>.2d.out beside it. A model with unsaved changes is saved first, exactly as the Run menu does before an ordinary run; an untitled model is refused with "Save the model first". A model whose validation finds errors is refused too, with the count, as Run → Run Model is.

20.3 2D Setup

The window's first line names the sidecar the settings go to. Nothing in this window touches the .inp.

DEM — the path to the elevation grid. Type it, drop the file on the window, or use … to browse. Load DEM reads it; once loaded the window reports its size and cell, its bounds, the elevation range, and how much of the model's drawn extent it covers ("covers the whole model extent", a percentage, or "does NOT overlap the model: check the coordinate system" — the usual first sign of a DEM in a different projection or different units from the .inp's coordinates). ESRI ASCII grids (.asc) and GeoTIFFs (.tif, read by the GIS chapter's reader) are accepted; the run reads the DEM with the same reader, so what loads here is what runs.

Grid

Resample to — a cell size in map units. Off: the DEM's own cell. A coarser cell runs faster; the time step scales with the cell size and the cell count with its square, so halving the cell costs about eight times the run time.
Refine to — a finer cell size, and the distance from a feature the fine cells reach. Off: one cell size everywhere. Section 20.9d covers what gets refined and what it costs.
Around 2D nodes — refine around every node that has a 2D interface. On by default, because the inlet is usually the thing worth resolving. Clear it to refine only the boxes you list.
max neighbour ratio — how abruptly the cell size may change between neighbours, 2.0 by default. Lower is smoother and larger; 1.0 allows a fine block to sit directly against a coarse one.
The box list takes one x0 y0 x1 y1 per line, in map units. Corners may be given in either order. A line that is not four numbers is refused on Save with its line number.
Window — restrict the grid to a rectangle of the DEM (x and y of two opposite corners). From map view fills the rectangle with the extent the map shows now. Off: the whole DEM.

Roughness (Manning's n) — one of:

Uniform — a single n for the whole grid (0.05 by default, a reasonable rough street-and-yard value; paved surfaces are nearer 0.015, grass 0.03–0.05, brush 0.1).
Raster of n — a grid of n values on the DEM's grid (or resampled to it).
Land-cover classes — a land-cover raster of class codes plus a table of class → n. Add class adds a row; − removes one. A row whose class is not a whole number or whose n is not a number is refused on Save with its row number.

Rain on grid — one of:

None — water reaches the surface only through the interfaces and sources.
Rain gage — one gage from the model's [RAINGAGES]; its time series falls on every cell.
Constant — one intensity in the model's rain units (in/hr or mm/hr) everywhere.
By subcatchment — each cell takes the gage of the subcatchment whose outline ([POLYGONS]) contains the cell's centre, so a model with several gages rains on the grid the way it rains on its subcatchments. A cell outside every outline takes the gage nearest to it on the map (the gages' [SYMBOLS] positions — Thiessen polygons, in effect). The Fallback gage (the model's first gage unless you choose one) covers what neither rule reaches: every cell when the model has a single gage, or has no subcatchment outlines and no placed gages, and the cells outside every outline when no gage is placed. Check rain assignment works the assignment out from the DEM and the model there and then and lists, for each gage that covers any of the grid, its cells, the area they cover, and how many came from subcatchment outlines (hover for which), from the nearest-gage rule and from the fallback. The same summary is among the run's notes.
Gridded — a sequence of rain rasters, such as radar frames, each an ESRI ASCII grid or a GeoTIFF in the model's map coordinates. Choose whether the values are an Intensity (in/hr or mm/hr) or a Depth per frame (in or mm fallen over the frame), and how long each frame lasts (Each frame lasts, seconds). List the frames one per line as a start time and a path — 0:05 radar/r0005.asc — the time as H:MM, H:MM:SS or a number of seconds from the model start. Add rasters… appends files in name order, one frame length after the last listed frame. A frame applies from its start for its length, or until the next frame starts if that is sooner; there is no rain outside every frame. Each grid cell takes the area-weighted mean of the raster cells it overlaps, so a coarse radar grid over a fine mesh puts the radar cell's value on every cell under it, and a cell over the raster's edge is not diluted by the ground outside it. No-data raster cells are left out; a grid cell with no raster data under it gets no rain, and the run's notes say how many there are.

Direct rainfall on the grid is the choice when the surface itself is the catchment; with subcatchments already draining to the nodes, leave it off or the rain is counted twice. Whatever the kind, the solver ends a time step wherever the rain changes — a new gage value, the end of a value's interval, a new frame — so the rain it books is exactly the input's.

Infiltration from the surface — None, a Constant loss rate in rain units, Horton (f0, fc, k in 1/hr), or Green-Ampt: Suction head (in or mm), saturated conductivity Ksat (rain units) and initial moisture deficit IMD (a fraction), the same three values and units as a Green-Ampt subcatchment's [INFILTRATION] row. Each cell keeps its own running total of what has infiltrated, so a cell that has been wet for an hour takes water more slowly than one the flood has just reached, and no cell ever loses more than the water standing on it. With From subcatchments ticked (the default), a cell inside a subcatchment's outline takes that subcatchment's own Green-Ampt row, and the three values here apply only outside every outline and in subcatchments that use another method; the run's notes say how many cells came from where. Chapter 20b §2.8 has the equations.

Boundary — how the edge of the grid behaves. Other edges sets every edge face that no boundary segment covers: Closed (a wall), Open (free outflow at the local slope), or Fixed head (a receiving water level; water flows out or in to hold it). Below it, boundary segments give chosen parts of the edge their own condition — a tide or river stage, an inflow hydrograph, normal depth, a wall or open outflow — with Add segment and Draw segment; section 20.9e covers them.

Exchange coefficients — the model-wide Weir C and Orifice C for every node and bank interface whose own coefficient is left on auto in the Interfaces window. Both are 0.6 by default, the usual literature value; the weir coefficient governs shallow capture, surcharge onto a dry lid and flow over bank crests, the orifice coefficient deep capture and surcharge through a submerged lid. Changing them here changes every interface still on auto, and none that has its own value. They are saved in the sidecar's [EXCHANGE] section only when they differ from 0.6.

Scheme

Local inertial (the default) — the momentum equation without its convective term (Bates et al. 2010). Fast and robust for flood spreading, ponding and rain on grid, where the flow is slow (Froude number well below 1).
Full dynamic (HLL) — the full shallow-water equations. Use it for dam breaks, steep streets and chutes, supercritical gutter flow and anything with a hydraulic jump, and where velocities matter, not just depths and extents. It takes more time steps than local inertial (about 1.4 times as many in slow water, more in fast) and more work in each.
Second order — shown once Full dynamic is chosen; on by default. The face values are reconstructed from each cell's neighbours and every step is taken in two stages, so fronts, bores and hydraulic jumps stay sharp (two cells instead of several) and smooth waves keep their height; a dam-break profile is about six times closer to the exact solution than at first order. Near shorelines and in water shallower than ten times the dry depth it falls back to first order on its own. It costs about twice the run time of first order. Turn it off to reproduce a run made before the option existed, or to trade accuracy for speed. The method and its tests are in 20b §2.4a and §6.3–6.3c.

The scheme and the order are saved in the sidecar's [RUN] section as SCHEME and ORDER. A sidecar saved before they existed runs as it did: local inertial (and first order if the full-dynamic scheme is chosen for it on the command line, aquaswmm-swmm twod run … --scheme hll; --order 1|2 overrides the order there too).

Time

Duration — hours; off: the model's own simulation length.
Output step — seconds between saved frames. Every frame stores depth and velocity for every cell, so 300 s on a 1000 × 1000 grid over six hours is about 900 MB; use a coarser step or a window for large grids.
Dry depth — below this depth (in the DEM's units) a cell is dry: it neither flows nor is drawn. 0.003 by default.
Courant — the Courant number of the adaptive time step (0.7 by default; lower is safer and slower).
Max dt — an explicit cap on the time step, seconds; off: the CFL condition alone sets it.
Threads — worker threads; 0 uses every core.

Save writes the sidecar and reports the path; Revert reads it again, discarding the edits. An "● unsaved" mark shows while the draft differs from the file. The same two buttons, acting on the same file, are in the Interfaces and Sources windows.

20.4 Interfaces

The 2D Interfaces window: where the 1D network and the 2D surface exchange water.

The table has every node with coordinates: its name and type, Ground (the DEM's elevation at the node once a DEM is loaded), Rim (the node's invert plus its maximum depth, from the .inp), and the interface:

Manhole — the default. Surcharge leaves the network when the node's head rises above ground; ponded water re-enters through the lid only when Lid open is ticked (a lid that has been lifted or a grated cover).
Inlet — captures surface water by a weir/orifice with the given Perimeter (ft or m) and clear opening area (ft² or m²); the perimeter governs shallow capture, the area deep capture.
Sealed — no exchange at all.

Weir C and Orifice C are the node's discharge coefficients: the weir one for shallow capture and for surcharge over a dry lid, the orifice one for deep capture and surcharge through a submerged lid (chapter 20b §4.1 has the formulas). auto uses the model-wide value from 2D Setup → Exchange coefficients (0.6 unless changed there); untick it to type the node's own.

A ground elevation well below the rim means the DEM and the model disagree about where the surface is, and the node will surcharge onto a surface lower than its own rim. Check the DEM's datum and units before running.

Selection — pick a kind and Apply to selection to set every node selected on the map at once. Seal all outfalls sets every outfall Sealed; an outfall is where water leaves the model, not somewhere it comes back onto the ground.

Bank lines couple an open channel to the cells beside it: water spills over the crest when the channel's head exceeds it and flows back when the surface is higher. Draw bank line starts the pick tool on the map: click near the conduit to choose it (or select the conduit first), then click the vertices along the bank; Enter or a double-click finishes, Escape cancels. Each line then has its Left/Right bank (as seen looking downstream), a Crest elevation (auto uses the DEM along the line), a Weir C (auto is the model-wide weir coefficient), its vertex count, and Remove. Spill from the channel onto the surface is limited, in a tight coupled run, by what the conduit's two end nodes hold and receive: each end gives at most what is left of its node's allowance for the interval after the node's own lid, so a bank cannot spill water the network does not have. While the pick tool is on, the ordinary map tools do not act; the wheel still zooms and the middle button still pans.

Everything here edits the sidecar draft; Save writes it. The .inp is untouched: the Unsaved mark in the status bar stays off.

20.5 Sources

The 2D Sources window: point inflows onto the surface that are not network nodes — a culvert outlet, a pump discharge, an upstream catchment represented by a hydrograph. Each has a Name, a place (X, Y), and a Series: a [TIMESERIES] in the model, in the model's flow units (Project → Time Series adds one). Type the name and choose the series, then Place source and click the map (Escape cancels); a name already in use gets the next free number. The table edits the place and series in place; Remove deletes a row.

20.6 Running

Run 2D Only runs the surface by itself: rain on the grid, infiltration and sources apply; the interfaces feed from nothing because there is no network run to feed them. It is the run for a pure overland-flow question (where does the rain go on this DEM?) and the quick check that a setup is sound before a coupled run.

Run Coupled (1D-2D)… opens a dialog with the Engine (the same registry as the Run menu) and the mode:

Tight — the engine is driven step by step through the engine bridge; every sync interval the node heads are read and the exchange flows set as lateral inflows, so surcharge and re-entry are resolved within the same time step. This is the mode to use when the surface feeds the network back — inlets capturing street flow, ponds draining into manholes. It needs the bridge helper aquaswmm-swmm-bridge32.exe (chapter 23 says where it is looked for, and scripts\build-bridge.ps1 builds it in a source tree) and a 32-bit engine with its swmm5.dll beside runswmm.exe, as EPA SWMM 5.2's Windows install has; without them the option is greyed with "Tight coupling needs the engine bridge; iterative is available". The engine runs to the end of its own simulation after the surface finishes, so the report is complete; its continuity errors, when above 5 %, are among the run's warnings.
Iterative — runswmm runs the whole 1D model, its node overflow hydrographs drive the surface, the surface's captured flows are written back as [INFLOWS] time series into a scratch copy of the model, and the pair is re-run until the exchanged volume changes by less than the tolerance or the pass limit is reached. It works with any engine and no bridge. It converges quickly when the surface mostly receives water (surcharge that ponds and drains away over the surface) and slowly, or not at all, when the surface and the network trade the same water back and forth, which is the tight mode's job. The scratch model, report and results are in the same per-model run folder as ordinary runs (chapter 9); the model's own .inp is never rewritten.

Run starts; Cancel closes the dialog.

Both runs open the Run 2D window and report as they go: simulation time against the duration as a progress bar, the current Time step, Wet cells, Surface volume, the running Mass error (coloured green, amber and red at the same 1 % and 10 % thresholds as the Run Status window's continuity errors), and the Elapsed wall clock. Stop 2D (also in the menu) raises a flag the solver checks at every progress report. When a run finishes the window shows the summary — results path, final mass error, steps and frames, elapsed, peak depth (over every step and every cell, so it agrees with the max-depth grid rather than with the frames), largest wet area, the inflow, outflow, infiltrated and stored volumes, and for a coupled run the iterations, the surcharged and captured volumes and the 1D run's files — a Boundary segments table of the volume in and out through each segment and through the edges no segment covers, when the run had segments — plus every warning; Load 2D Results brings the results onto the map (a finished run does this by itself). Anything that stops a run from starting — no DEM, no saved model, validation errors, a DEM that does not cover the model — is written on the window's status line; the window never disappears on an error. Close closes it.

The run happens on its own thread; the editor stays live and the map can be panned while it goes.

20.7 Results on the map

With results loaded, the map paints a grid over the DEM's extent under the network. The Layers tab's 2D overland section controls it:

2D results — the overlay on or off.
The mode: Depth, Velocity or Hazard (depth × velocity) in the current frame; Max depth, Max velocity, Max hazard and Arrival time over the whole run; Water table depth in the current frame, for a run with groundwater (section 20.9f).
opacity.
2D frame — a slider over the saved frames, shown when the frame is not already set by the SWMM time slider (below).
2D Legend and Show Interfaces, the same toggles as the menu.

Cells below the dry depth are transparent, so the DEM, backdrop and GIS layers show through the dry ground. The colour scale of a frame mode is the run's maximum of that quantity, so the same depth is the same blue in every frame; the legend in the map's lower-right corner gives the scale, the unit (ft or m, and the corresponding velocity and hazard units, after the model's flow units) and the frame's time. Move the pointer over the grid and the depth, velocity and hazard under it are written beside the cursor (the static modes read out their own grid).

The time slider. When the SWMM results are loaded and the Results view is showing an instant (the period slider, Play, ◀ ▶), the 2D overlay shows the 2D frame nearest that instant: SWMM's reporting period p is at (p + 1) × report step seconds, and the nearest saved 2D frame by time is chosen. Playback of the SWMM results therefore animates the surface as well. With the SWMM results showing the run's peaks, or with no SWMM results loaded, the 2D frame slider in the Layers tab chooses the frame instead.

Interface markers (Show Interfaces): a circle at every manhole interface (a dot inside it when the lid is open), a square at every inlet, a cross at a sealed node, and the bank lines as thick orange polylines labelled with their link and bank. With results loaded the node markers are coloured by the exchange in the current frame: red when water is leaving the network onto the surface, blue when the surface is being captured, grey when nothing is passing; the whole-run modes colour by the net exchange. Sources are purple triangles.

20.8 Exporting

The three export items write ESRI ASCII grids (.asc) with the results' geometry and -9999 for no-data, which any GIS opens directly: the run's maximum depth, the depth of the current frame (whichever frame the overlay is showing), and the maximum hazard. Arrival time and maximum velocity are on the map but not yet exported.

20.9 The sidecar file

<model>.2d is a plain text file beside the model holding everything in the three windows: the DEM path and grid, roughness, rain, infiltration, groundwater, boundary and boundary segments, time stepping, the node interfaces, the bank lines and their vertices, the sources and the sealed list. Its format (sections, keywords and an example) is in 20b, §5.1. Opening a model reads its sidecar; a model without one starts from the defaults (a manhole at every node, uniform n = 0.05, no rain on grid, closed boundary, local-inertial scheme). The file carries a format version ([FORMAT] VERSION 2 since the scheme options were added); an older AquaSWMM cannot open a version-2 sidecar, and this one reads every older file as it was written. A different model gets its own sidecar; the results and DEM loaded for one model are dropped when another is opened.

20.9b Buildings

Building footprints block the surface. Give the Buildings box in 2D Setup one polygon layer per line (.shp or .geojson); every feature in them removes the cells it covers from the flow field, and water goes around the building.

This is deliberately not the same as raising the DEM under the footprint. Raising the ground makes the roof part of the flow field: water runs up the wall, ponds on top, and the building appears as a puddle in the depth map. Blocking takes the cells out of the domain instead, which is what a footprint is actually telling you.

Details worth knowing:

A cell is blocked when its centre falls inside a footprint. Blocking on any overlap would swell every building by up to a cell on each side, and on a 1 m grid a terrace of houses would fuse into one block with the alleys between them sealed.
Holes in a polygon — courtyards — stay open.
A blocked cell is exactly a no-data cell to the solver, so buildings obey the same wall rule as the edge of the survey. Nothing enters them, and the mass balance is unaffected.
The DEM keeps its ground elevations underneath. Clearing the box puts the terrain back without reloading it.
A node whose cell is under a footprint gets no 2D interface, and says so by name in the run's warnings — distinct from the "no-data cell" warning, because a manhole in a courtyard or an alley is a mapping problem, not a survey gap.
The sidecar records this as a [BUILDINGS] section with one BLOCK line per layer, so it travels with the model.

20.9c Evaporation

Standing water on the surface evaporates according to the model's own [EVAPORATION] section. CONSTANT, MONTHLY and TIMESERIES are read; rates are a depth per day, which is SWMM's convention. SWMM's DRY_ONLY is honoured, so nothing evaporates off a surface that is being rained on.

TEMPERATURE and FILE derive evaporation from a climate record the 2D surface does not read. Rather than evaporate nothing and say nothing, the run warns that surface evaporation is off and names the three forms that work.

Evaporation only ever removes water that is there: it is applied after infiltration, bounded by the remaining depth, and booked in the mass balance as its own term, so a run with evaporation still closes to the same tolerance as one without. A MONTHLY rate uses the column for the month the run starts in — a 2D run covers a storm, not a season.

A uniform grid has to be as fine as its most demanding feature. Resolving a kerb line at 0.5 m means 0.5 m cells over the whole catchment, and since the cell count goes as the inverse square of the cell size and the time step goes with the cell size, that is roughly an eightfold cost for detail you wanted in one street.

Refine to lets the grid be fine where it matters and coarse elsewhere. Columns and rows are split — the mesh stays rectilinear, so a refined column is fine for its whole height — and a grading sweep keeps neighbours within the max neighbour ratio of each other, so the size change is gradual rather than a step from 10 m to 0.5 m in one face.

What gets refined:

every node with a 2D interface, within the radius you give, unless Around 2D nodes is cleared;
every box you list.

What it costs, and what it does not buy:

The time step follows the smallest cell anywhere on the mesh, because a wave crosses the narrowest cell first. Refining a single inlet to 0.5 m on a 10 m grid makes every step a 0.5 m step. The saving is in cell count, not in step size — which is still most of the saving, but it is worth knowing before refining to a tenth of the base cell.
Refinement splits columns and rows, not individual cells, so a box refined in the middle of the grid leaves a fine stripe running to each edge. The stripe is cheap (it is one column wide) but it is not free; a handful of boxes is the intended use, not dozens.
Refinement does not add information to the DEM. Cells are sampled from the elevation grid you loaded, so refining below the DEM's own cell size subdivides the same surface into more cells of it. A 0.3 m mesh on a 1 m LiDAR DEM resolves the flow, not the kerb.

The solver treats a graded mesh exactly as it treats a uniform one: each face uses the centre-to-centre distance across it and the length of the face itself, and continuity uses the true cell area. Lake at rest stays exact, mass closes over the real ground area, and steady uniform flow matches Manning straight through a refinement interface — those three are tests, run on every build, and section 6 of the methods chapter gives the numbers.

The sidecar records this as a [REFINE] section (CELL, RADIUS, RATIO, NODES, and a BOX line per box), so it travels with the model. A refined run writes a version 2 results file, which carries the mesh edges; a uniform run still writes version 1.

20.9e Boundary segments

The grid's edge is often not one thing. A model of a tidal outfall has the sea along one side, a creek coming in across another, and high ground everywhere else; a model of a reach has the upstream river entering at one end and leaving at the other. A boundary segment gives a part of the edge its own condition. The faces no segment covers keep Other edges (Closed, Open or Fixed head), so a model without segments runs exactly as it did before they existed.

Segments are listed in 2D Setup, under Boundary. Each row has:

Name — anything; it names the segment in the run's volumes and on the map.
Type:
- Stage — a water level at the edge: a tide, a downstream river or lake level. Water flows in while the level is above the water on the grid and out while it is below. The level is an elevation in the DEM's units (the same datum as the DEM).
- Inflow — a flow hydrograph entering across the segment, in the model's flow units. The flow is shared among the segment's faces by conveyance, so it enters where a channel crossing the edge would carry it — the deep, smooth part of the edge — rather than spread evenly over dry banks; on a dry edge the first water enters at the lowest cells. The volume that enters is the hydrograph's, exactly.
- Normal depth — free outflow at Manning normal depth for the friction slope you give (the slope of the channel or ground beyond the edge). Unlike Open, which uses the local slope at each face, normal depth does not draw a pond down against a flat edge and does not back water up at a sloping one.
- Closed — a wall on just this part of the edge.
- Open — the free outflow of Open, on just this part.
Value (stage and inflow) — Constant, a Series (any [TIMESERIES] in the model, dated or not, placed on the model's clock as for sources), or a File of time value rows (comma, tab or space separated; the time H:MM[:SS] or decimal hours from the start; a header row is skipped). A stage holds its first value before the series starts and its last after it ends; an inflow is zero outside its series, as a source is, and a negative flow counts as none.
Place — a side (West, East, North, South), either the Whole side or, with Whole side off, the part between two coordinates measured along it (y on the west and east sides, x on the north and south), or a line drawn on the map.

Add segment adds a stage on the whole west edge to edit. Draw segment draws one on the map: click points along the grid edge, then Enter or a double-click finishes (Escape cancels); the new segment is a stage, to be given its type and value in the list. A drawn line takes the edge faces it runs beside, within one cell of the edge, so it can be drawn along the edge or through the edge cells; at a corner it takes only the side it runs along. Where two segments claim the same face, the one listed first keeps it, and the run's notes say so; a segment that covers no face is a warning too. − removes a row.

On the map each segment is drawn along the edge in its type's colour — stage blue, inflow green, normal depth teal, closed grey, open orange — with its name, whenever the interfaces are shown or 2D Setup is open.

After a run, the Run 2D window's Boundary segments table gives the volume in and out through each segment and through the other edges; they add up to the boundary part of the mass balance.

In the sidecar each segment is a SEGMENT row in [BOUNDARY]:

[BOUNDARY]
CLOSED                                     ; every face no segment covers
SEGMENT  Tide   STAGE TideLevel EDGE WEST       ; a [TIMESERIES] of levels
SEGMENT  Creek  INFLOW FILE "creek q.csv" EDGE NORTH 120 180
SEGMENT  Outlet NORMAL_DEPTH 0.002 LINE 500 0 500 40

Chapter 20b §2.7 has the methods for each type and §6.7c the tests.

20.9f Groundwater

On flat ground with a high water table — coastal plains, Florida — floods are made by the ground filling up: once the soil is saturated the rain runs off, low spots pond from below, and groundwater leaks into the pipes. The Groundwater part of 2D Setup puts SWMM's own two-zone aquifer (the one behind [AQUIFERS], [GROUNDWATER] and [GWF]) under the 2D surface, so the surface, the groundwater and the sewer keep one water balance.

It is SWMM's lumped model: one water table per store, no groundwater flow between stores or cells. It is not a 2D or 3D groundwater model (a MODFLOW-type coupling would be a separate, later option). Chapter 20b §2.11 has the equations and §6.12 the comparison with EPA SWMM's own groundwater.

Groundwater under the surface — off by default. When on:

Parameters. A cell inside the outline of a subcatchment that has a [GROUNDWATER] row belongs to that subcatchment's store and uses its [GROUNDWATER] and [AQUIFERS] rows (and any [GWF] expressions) as SWMM would. Set them in the subcatchment's groundwater editor (chapter 21).
Default aquifer for other cells — one store for every other cell: the Aquifer (an [AQUIFERS] name), the Node it drains to, its ground elevation (Ground from the DEM: the mean DEM elevation of its cells; clear it to type one), and the flow coefficients A1, B1, A2, B2, A3 and the Fixed depth of surface water, as in a [GROUNDWATER] row. Without it, cells outside those subcatchments have no groundwater and their infiltration leaves the model as before.
Owner — who computes the groundwater of a subcatchment that has a [GROUNDWATER] row and lies under 2D cells, so the same water is not in two aquifers: 2D surface (the store does; coupled runs take that subcatchment's [GROUNDWATER] row out of the engine's copy of the model — your .inp is not changed) or SWMM subcatchment (SWMM does; the 2D cells inside it get no store). While it is Not chosen and such a subcatchment exists, the run stops with a message naming it.
Step — the groundwater time step, 60 s by default. The 1D-2D exchange happens at least this often when groundwater is on.
GWI coefficient — groundwater infiltration into sewers, per day: a conduit over a store and below its water table takes C × π × diameter × length × (water table − invert), half to each end node. 0 turns it off. 20b §2.11.4 converts a gpd/inch-diameter/mile allowance into it.
Seepage onto the surface — cells whose ground is below the water table receive groundwater (at the aquifer's conductivity), so low ground ponds from below. Water the store cannot hold always comes back this way.
Start state — a state file to start from (below). Empty: each aquifer's own initial water table and moisture.

What it does to a run:

Infiltration under a store goes into the store. A cell may infiltrate only what the store has room for (none where the water table is at the ground), whatever its own infiltration method allows, so a saturated store turns rain into runoff.
The store drains to its node by SWMM's groundwater flow equation and into pipes as GWI. In a coupled run that water reaches the network: as lateral inflow in tight coupling, in the write-back series in iterative coupling; the volume the network is given is exactly what the stores lost. In a 2D-only run it leaves the model and is reported.
The Run window's summary gains the groundwater balance: Groundwater error (round-off), GW recharge / exfiltrated (from the surface and back onto it), GW ET / deep (upper- plus lower-zone evapotranspiration, and deep percolation), GW lateral / GWI (to the nodes and into the pipes) and GW stored (start → end), all in cubic DEM units. The surface's own Mass error now counts the exfiltrated water as an inflow and the recharge as the infiltration it is.

Files beside the model:

<model>.2d.gw — the water table of every store at every saved frame; the map's Water table depth layer reads it (depth from the ground down to the water table, negative where the water table is above the ground).
<model>.2d.gwstate — the state at the end of the run. For a continuous or forecast run, copy it and name the copy in Start state: the next run starts with the water table where this one ended. From the command line, aquaswmm-swmm twod run model.inp --gw-state-in a --gw-state-out b does the same, and chapter 25 shows how a forecast chain carries it beside SWMM's hotstart files. Its format is in 20b §5.3.

20.10 Limitations

DEMs are read from ESRI ASCII grids and GeoTIFFs; other raster formats have to be converted first.
Boundary segments act on the outer edge of the grid rectangle only. The edge of a no-data region inside it (a modelled-area polygon, a building) is always a wall. A stage cannot control flow faster than critical: water leaving supercritically passes the edge at critical depth whatever the stage says. With Second order on, the cells along the edge are first order, as at every wall (section 20.9e; 20b §7).
The local-inertial scheme, and the full-dynamic one with Second order off, smear fronts and jumps over a few cells; with Second order on, the full-dynamic scheme is still first order at shorelines and in very shallow water (20b §7).
The mesh is rectilinear: a refined region splits whole columns and rows, so fine cells run in stripes to the edge of the grid rather than sitting in an isolated patch. A quadtree, which would confine them, is a planned change.
The time step follows the smallest cell on the mesh, so refinement saves cell count but not step size (section 20.9d).
The overlay is sampled down to at most 4096 cells along either side for drawing (the results themselves keep every cell); exports are full resolution.
Tight coupling needs the 32-bit engine bridge (EPA's Windows engine is 32-bit and AquaSWMM 64-bit; chapter 23 explains the bridge). Without it the coupled dialog offers iterative coupling only.
The results file stores depth and the two velocity components per cell per frame; there is no compression, so budget the output step for the grid size.
Rain on the grid needs TIMESERIES gages; a gage reading a rainfall FILE cannot be used. Rain by subcatchment and Green–Ampt from subcatchments assign a cell by where its centre falls, so a subcatchment narrower than a cell may get no cells of its own.
Green–Ampt on the surface does not recover the moisture deficit between rain bursts within a run, and Modified Green–Ampt is run as Green–Ampt; a 2D run covers one storm.
Buildings block the cells whose centres they cover; a building narrower than a cell is not seen (section 20.9b). Surface evaporation reads CONSTANT, MONTHLY and TIMESERIES only (section 20.9c).
The exchange coefficients default to 0.6 for every interface, a literature value rather than a calibrated one; set them per interface in the Interfaces table or model-wide in 2D Setup where a rating is known.
A node's Rim in the Interfaces table is the rim the engine uses: the invert plus MaxDepth raised to the crown of every pipe that meets the node (storage units keep their MaxDepth). A junction with MaxDepth 0 therefore shows the crown of its highest pipe, not its invert. Water leaves a node onto the surface when the engine floods it at that rim, or, for a node with a surcharge depth, when its head rises above both the rim and the water on the cell. Chapter 20b §4.2 has the formulas.
Groundwater is SWMM's lumped two-zone model (section 20.9f): one water table per store, no lateral groundwater flow between stores or cells, no flow from the network into the ground, and GWI is a single leakance for every pipe.
Iterative coupling cannot take water out of a network run that has already finished. The engine's own flooding goes onto the surface in the pass that reads it. A pressurised manhole's lid overflow (a node with a surcharge depth whose head rises above its rim) is worked out in one pass, written into the next run as a withdrawal from the node, and put on the surface by the pass after that, as exactly the volume that run withdrew. It therefore needs at least two passes; with one, the run's warnings say so. The passes stop early only once those overflows have settled. Bank spill from a channel onto the surface still needs tight mode; in iterative mode bank lines only return surface water to the channel.

20.11 Tutorial: rain on the Site Drainage model

This tutorial is the one the editor's end-to-end tests run, step by step, through the same menus and buttons (app/src/swmm_twod_e2e_tests.rs); the numbers quoted are from those runs with EPA SWMM 5.2.4.

The EPA Site Drainage sample (Site_Drainage_Model.inp, in the EPA install's Samples folder) is a small development drained by swales, culverts and street gutters to one outfall, with coordinates in feet and junction inverts from 4963 to 4973 ft. Copy it to a folder of its own as site.inp.

Make a DEM. The model has no DEM, so build one from the network: interpolate a surface through the ground at every node (the Rim the Interfaces table shows: the invert plus the node's full depth by SWMM's own rule, which for these MaxDepth 0 junctions is the crown of the highest conduit that meets them) at a 20 ft cell, with a cell of margin round the model's drawn extent. Any GIS does this: inverse-distance weighting from a point layer of the nodes (File → Export GIS Layers…), saved as .asc or .tif. Save it as ground.asc beside site.inp.
Setup. Open site.inp, 2D → 2D Setup…, type the DEM's path (or drop the file on the window) and Load DEM: it covers the whole model extent. Set rain on grid to Constant, 3 in/hr; boundary Open; tick Duration and set 0.333 h (20 minutes); output step 300 s. Leave roughness uniform 0.05 and infiltration None. Save: the status line reads "Saved …\site.2d".
Run 2D Only. 2D → Run 2D Only. The Run 2D window counts to 00:20:00 in about a second and reports 5 frames (0, 5, 10, 15 and 20 minutes), mass error +0.000 %, inflow 210,267 ft³ (3 in/hr for a third of an hour over the grid), outflow 32,993 ft³ over the open edge, and a peak depth of about 1.4 ft in the low spots between the nodes. The results load by themselves; the status bar says "2D run finished".
Look and export. The map shows the depth over the DEM. In the Layers tab's 2D overland section drag 2D frame to the last frame, or switch to Max depth. 2D → Export Max Depth Grid… writes site_max_depth.asc, a grid the size of the DEM with the deepest water of the run in each cell; Export Hazard Grid… writes the depth × velocity maximum. 2D → Load 2D Results re-reads the file at any time.
Inlets. Now let the network take the water. 2D → Interfaces…, select everything on the map (Ctrl+A), choose Inlet and Apply to selection (10 ft perimeter and 2 ft² opening unless you change them in the table), then Seal all outfalls. Back in 2D Setup, set the duration to 0.5 h and Save.
Coupled, iterative. 2D → Run Coupled (1D-2D)…, choose the engine, Iterative, up to 3 passes, tolerance 0.020, Run. The run takes 2 passes (the second changes the exchanged volume by less than 2 %), the surface mass error stays at +0.000 %, and the inlets capture 23,276 ft³ of the 315,400 ft³ that fell on the grid in the half hour; nothing surcharges. The summary's Surcharged / captured and 1D run rows show the volumes and the scratch model's files.
Coupled, tight. The same dialog, Tight, sync every 30 s, Run (it needs the bridge, see §20.6). The engine is stepped alongside the surface and the captured water enters the network as it is captured: 32,954 ft³ captured, nothing surcharged, surface mass error +0.000 %, no engine continuity warning. Tight captures more than iterative here because the network heads it sees are the ones the capture itself produces, not a previous pass's.

When the network surcharges. The sample's pipes carry the storm, so nothing leaves the network above. On a stressed copy (every conduit narrowed 2.5 times, the same kind of DEM through the rims, manholes at the junctions) the engine alone reports a flooding loss of 0.040 MG (about 5,350 ft³). Coupled, that water goes onto the surface instead: 5,229 ft³ iterative, 5,360 ft³ tight. With every lid ticked Lid open, tight coupling captured 1,454 ft³ of it back, which the engine booked as 0.011 MG of external inflow. The surface balance was 0.0000 % in every case.

Stop 2D ends a run early at its next progress report. Try it with a ten-hour duration: the summary then carries "2D run stopped by the caller at … s" among its warnings, and the frames written so far load as usual.