NeuraCities
All posts
StormwaterStormwaterFlood modelingDrainage engineering

AI stormwater analysis for engineers: how a drainage study works, and what AI can speed up

The NeuraCities Team 11 min read

A drainage study rarely gets stuck on the hydraulics. It gets stuck on everything around them: finding terrain, reconciling a pipe network, building the model, running storm after storm, working out what is actually at risk, then pricing alternatives and writing it up. The physics is often the quickest part.

This guide walks through a stormwater study as engineers run it, which modeling tool fits which question, and where AI helps. It uses a real HEC-RAS model of Houston's Sims sub-basin as the example. The aim is to speed up the assembly and comparison, and leave the engineering judgment with the engineer.

What a stormwater study has to answer

Whether it is a master plan, a capital project or a grant application, a study usually comes down to four questions:

  1. 1How much water arrives, and how fast, for a given storm?
  2. 2Where does the drainage system run out of capacity, and where does the water go?
  3. 3What is in the way: which buildings, roads and public facilities?
  4. 4What should we build, what does it cost, and is it worth it?

Each question depends on the previous one, so a change in rainfall or land cover flows all the way through to the benefit-cost table. That is why studies are slow, and why a repeatable workflow pays off.

SWMM, HEC-RAS or HEC-HMS: which model fits

The three most familiar public tools overlap, but each has a home:

ToolFromBest atTypical use
HEC-HMSUS Army Corps of EngineersHydrology: how watersheds turn rainfall into runoff, by event or continuouslyRunoff hydrographs for a watershed, often passed to a hydraulic model
EPA SWMMUS EPAUrban runoff and drainage networks, including pipes, pumps, storage and water qualityStorm sewer capacity, combined systems and green infrastructure
HEC-RASUS Army Corps of EngineersHydraulics: 1D and 2D unsteady flow, plus stormwater pipe networksFloodplains and detailed overland flooding maps

SWMM is free and open source, and it covers eight kinds of low impact development control, including bioretention, permeable pavement, green roofs and vegetative swales, so it is a natural fit for comparing green infrastructure.

The choice depends on where the flooding happens. When water leaves the pipes and spreads across streets and yards, a 2D surface matters. When the question is pipe capacity along a trunk line, a network model may be enough. Follow your reviewing agency's requirements first: many specify accepted models and methods.

Stage 1: build the model

Before anything runs, you need the study area's ground and pipes in one place:

InputCommon sourcesWhat to watch for
TerrainLiDAR-derived elevation models such as USGS 3DEPBridges and culverts can appear as dams in a bare-earth surface
Land cover and imperviousnessNational land cover data, local impervious layersOlder layers can miss recent development
Drainage networkThe city's stormwater GIS, record drawingsMissing inverts, unknown sizes and disconnected pipes
RainfallNOAA Atlas 14 or a local design manualChoice of duration, distribution and return period
SoilsSoil survey dataNeeded for infiltration losses

The drainage network usually takes the most effort. Public GIS layers often lack flow lines or have gaps, and some of the detail exists only in scanned drawings. For those, NeuraCities Digitization can return drainage sheets as a runnable SWMM network.

The example: Houston's Sims sub-basin

The example on the Stormwater product page is a real model: HEC-RAS 7.0, with 1D pipes and 2D overland flow, covering the Sims sub-basin in Houston under a 10-year event. The drainage data is Houston Public Works' own stormwater GIS, laid over the model's terrain.

21,448
storm drain mains
Houston Public Works GIS
7,750
inlets
in the study extent
4,911
manholes
in the study extent
39,501
building footprints
mapped in the sub-basin extent

Stage 2: choose and run the storms

In the US, design rainfall usually starts with NOAA Atlas 14, which NOAA describes as the current national standard. It gives depth by duration and return period at a location, and most local drainage criteria build on it or on a local equivalent.

10-year rainfall depth by duration, Sims sub-basin
inches of rain
NOAA Atlas 14 Volume 11 point estimates for the Sims sub-basin centroid in Houston (partial duration series). These are the figures behind the example study.
Show the data
30 minutes2.4 in
1 hour3.3 in
2 hours4.3 in
3 hours5 in
6 hours6.2 in
12 hours7.5 in
24 hours8.9 in

Depth is only part of a design storm. Duration, temporal distribution and areal reduction all change the peak, so state them with your results. A short, intense storm tends to stress small pipes, and a long one tends to stress storage and large trunk lines. That is why studies usually run several durations and return periods.

Atlas 14 and Atlas 15
Atlas 14 assumes the rainfall record is stationary, so it does not account for change in extremes over time. NOAA's Atlas 15 will supersede it and adds nonstationary methods and future projections. NOAA's page lists contiguous US publication as no earlier than September 2027, though parts of the same page give earlier dates, so check it directly. Until publication, Atlas 14 stays the authoritative source. Some jurisdictions also apply their own factors.

In the example, the 10-year, 12-hour storm drops 7.47 inches. The model steps through it hour by hour, tracking flood depth across the surface and through the pipes.

Stage 3: find what is at risk

A depth grid is not yet a finding. Decision makers ask how many homes are affected, and whether a fire station can still operate. To answer that:

  1. 1Collect building footprints and public assets, including emergency facilities.
  2. 2Sample the flood depth at each one, using the peak for each storm.
  3. 3Compare that against a damage threshold. Ideally, use finished floor elevations rather than ground level.
  4. 4Count and map what exceeds the threshold, by storm event.

Your choice of threshold changes the headline number, so report it alongside the count. The Houston example flags buildings where water goes over 1.6 feet. Footprints come from Microsoft's building footprints, and public assets come from OpenStreetMap. A real damage assessment would use surveyed floor elevations and depth-damage curves where they are available.

Sanity-check the map
A 2D model grid simplifies the ground. Deep values can sit in channels and detention basins rather than where people are, so peak depth alone can mislead. Look at where the depth occurs. Then check flow paths against known complaint locations, high-water marks or gauges before you present anything.

Stage 4: compare mitigation designs

This is where the most value is, and often where teams run out of budget. Each alternative has traditionally meant a manual model edit, a rerun and a cost estimate, so a study may compare only a handful of options. The common families are:

OptionWhat it doesTradeoffs
DetentionStores water and releases it slowly, reducing peak flow downstreamNeeds land, but can double as a park or open space
Bigger mainsAdds conveyance at a known pinch pointDisruptive to build, and can move the problem downstream
Green infrastructureSlows and absorbs water close to where it falls using bioswales, porous paving and similarSpread across many sites, and needs ongoing maintenance

A workable comparison reruns each option through the same storms, then measures what it takes out of the water: buildings removed from the flooded count, peak depth reduced, and cost. Check for side effects too, because a fix that lowers one street can raise another.

This table shows how a ranked result can look. The numbers are illustrative, from the product page example, and are not engineering estimates for the Sims sub-basin:

RankOption (illustrative)SizeCostBenefit-cost ratio
1Regional detention basin42 ac-ft, 11 ac footprint$8.4M2.7
2Upsized trunk main2.4 km, 96 in box$5.1M2.1
3Green infrastructure corridor9 blocks of bioswales and porous paving$1.8M1.5

Benefit-cost ratios and funding

For federal hazard mitigation funding, FEMA generally treats a project as cost-effective when its benefit-cost ratio is 1.0 or greater, using its approved methods and tools. Thresholds, discount rates and accepted tools can change by program, so confirm them in the current notice.

Funding programs also move. FEMA's Building Resilient Infrastructure and Communities (BRIC) program is a good example: FEMA announced it was ending the program in 2025, a court later ruled that unlawful, and a new funding notice followed in 2026. Check the current status with your state hazard mitigation officer before you build a plan around it.

Stage 5: report it

A good stormwater report lets a reviewer trace every number. Include the following:

  • Study area, data sources and their dates.
  • Model software and version, and the rainfall source, durations and return periods.
  • Assumptions, including roughness, infiltration, boundary conditions and any filled-in network data.
  • Calibration or validation evidence, or a plain statement that there is none.
  • Depth maps and affected counts for each storm, with the threshold stated.
  • Alternatives considered, their costs and benefit-cost ratios, and the reason for the recommendation.

Council members and grant reviewers need a shorter version with the same numbers. Keeping one source for both avoids the common problem of a summary drifting away from the model.

Where AI helps, and where it does not

AI is useful where the work is assembly, repetition and comparison. It should not replace the solver or the engineer.

TaskGood fit for AI?Why
Finding and cleaning input dataYesRepetitive work with clear checks
Setting up model inputsYes, with reviewMechanical, but assumptions need checking
Solving the hydraulicsNoUse established solvers. A chat model should not guess flood depths
Running many storms and designsYesThis is where the time savings add up
Drafting reports and chartsYes, with reviewEvery number should come from the model output
Choosing assumptions and sealing the workNoRequires professional judgment and accountability

The important distinction is between an AI that writes about flooding and one that runs a recognized engine and reports its output. Depths should come from SWMM, HEC-RAS or HEC-HMS, so they can be inspected and reproduced.

How NeuraCities Stormwater Engineering runs a study

Stormwater Engineering covers the whole study, from building the drainage model to recommending fixes:

  1. 1Model. Terrain, land cover, rainfall and the city's real drainage infrastructure are pulled in automatically, and the model is prepared to run.
  2. 2Storm. It simulates multiple storm events and returns flood depths for each, using EPA SWMM, HEC-RAS and HEC-HMS.
  3. 3Risk. It identifies every building footprint in the sub-basin, plus emergency facilities and city assets, and counts what stands in damaging depths.
  4. 4Design. It evaluates alternatives such as detention, bigger mains and green infrastructure, considering 100 or more possible designs, with cost and reasoning on every option.
  5. 5Report. One click generates a shareable link with charts built from the flood curve, a costed options table and benefit-cost ratios. You can also export for the GIS team.
50x
less time
for a city-wide study
$500k+
saved
per sub-basin
100+
designs considered
for mitigation

Those are the product's headline figures. Results will vary with the catchment and with the quality of the local drainage data. Stormwater Engineering is available by request: bring a catchment and the team will run a real storm across it.

The engineer stays in charge
NeuraCities can do the assembly and comparison, but the responsible engineer still reviews the assumptions, checks the results against observed flooding and signs the work. A modeling tool does not replace that role, or the requirements of your reviewing agency.

Who this is for

  • City and county drainage engineers prioritizing capital projects across many sub-basins.
  • Consulting engineers who want to spend less time on data preparation and more on design.
  • Public works and floodplain managers who need to show what floods first and why.
  • Planners and resilience teams preparing the evidence for a mitigation plan or a funding application.

Frequently asked questions

What is the best software for stormwater modeling?

It depends on the question. HEC-HMS suits watershed hydrology, EPA SWMM suits urban drainage networks and green infrastructure, and HEC-RAS suits river and 2D overland flooding. Many studies combine them, and your reviewing agency may specify what it accepts.

What is the difference between SWMM and HEC-RAS?

SWMM is an EPA model built around urban runoff and drainage networks, including pipes, storage, pumps and water quality. HEC-RAS is a US Army Corps of Engineers hydraulic model for 1D and 2D unsteady flow, with stormwater pipe network modeling. HEC-RAS is often chosen when detailed surface flooding matters.

Can AI do stormwater analysis?

AI can handle much of the assembly and comparison: gathering data, setting up inputs, running many storms and designs, and drafting reports. Flood depths should come from an established engine such as SWMM or HEC-RAS, and a qualified engineer should review the assumptions and results.

Is NOAA Atlas 14 still current?

Yes. NOAA describes Atlas 14 as the current national standard and says it stays authoritative until Atlas 15 is published. NOAA's Atlas 15 page lists contiguous US publication as no earlier than September 2027, so check that page for the latest schedule.

What is a good benefit-cost ratio for a drainage project?

For FEMA hazard mitigation programs, a ratio of 1.0 or greater generally counts as cost-effective, when calculated with FEMA-approved methods. Requirements vary by program and change over time, so check the current notice.

How do you identify buildings at risk of flooding?

Overlay building footprints and public assets on the modeled flood depth for each storm, compare depth at each location with a damage threshold, and count what exceeds it. Using finished floor elevations gives a better answer than ground level alone.

What does NeuraCities Stormwater Engineering do?

It runs the whole storm study: pulling in terrain, land cover, rainfall and the city's drainage network, running storm events with EPA SWMM, HEC-RAS and HEC-HMS, flagging buildings and assets in damaging depths, and comparing designs such as detention, bigger mains and green infrastructure with costs and benefit-cost ratios.

Does AI replace a drainage engineer?

No. It speeds up the data work and lets you compare more options. Engineers still choose assumptions, validate results against real flooding, make the recommendation and take professional responsibility for the work.

More guides on stormwater engineering

What floods first? See a real storm run across Houston's Sims sub-basin, then bring your own catchment.

See Stormwater Engineering

Have a question about your own data?

Talk to us, or read the common questions.