AI stormwater analysis for engineers: how a drainage study works, and what AI can speed up
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:
- 1How much water arrives, and how fast, for a given storm?
- 2Where does the drainage system run out of capacity, and where does the water go?
- 3What is in the way: which buildings, roads and public facilities?
- 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:
| Tool | From | Best at | Typical use |
|---|---|---|---|
| HEC-HMS | US Army Corps of Engineers | Hydrology: how watersheds turn rainfall into runoff, by event or continuously | Runoff hydrographs for a watershed, often passed to a hydraulic model |
| EPA SWMM | US EPA | Urban runoff and drainage networks, including pipes, pumps, storage and water quality | Storm sewer capacity, combined systems and green infrastructure |
| HEC-RAS | US Army Corps of Engineers | Hydraulics: 1D and 2D unsteady flow, plus stormwater pipe networks | Floodplains 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:
| Input | Common sources | What to watch for |
|---|---|---|
| Terrain | LiDAR-derived elevation models such as USGS 3DEP | Bridges and culverts can appear as dams in a bare-earth surface |
| Land cover and imperviousness | National land cover data, local impervious layers | Older layers can miss recent development |
| Drainage network | The city's stormwater GIS, record drawings | Missing inverts, unknown sizes and disconnected pipes |
| Rainfall | NOAA Atlas 14 or a local design manual | Choice of duration, distribution and return period |
| Soils | Soil survey data | Needed 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.
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.
Show the data
| 30 minutes | 2.4 in |
| 1 hour | 3.3 in |
| 2 hours | 4.3 in |
| 3 hours | 5 in |
| 6 hours | 6.2 in |
| 12 hours | 7.5 in |
| 24 hours | 8.9 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.
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:
- 1Collect building footprints and public assets, including emergency facilities.
- 2Sample the flood depth at each one, using the peak for each storm.
- 3Compare that against a damage threshold. Ideally, use finished floor elevations rather than ground level.
- 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.
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:
| Option | What it does | Tradeoffs |
|---|---|---|
| Detention | Stores water and releases it slowly, reducing peak flow downstream | Needs land, but can double as a park or open space |
| Bigger mains | Adds conveyance at a known pinch point | Disruptive to build, and can move the problem downstream |
| Green infrastructure | Slows and absorbs water close to where it falls using bioswales, porous paving and similar | Spread 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:
| Rank | Option (illustrative) | Size | Cost | Benefit-cost ratio |
|---|---|---|---|---|
| 1 | Regional detention basin | 42 ac-ft, 11 ac footprint | $8.4M | 2.7 |
| 2 | Upsized trunk main | 2.4 km, 96 in box | $5.1M | 2.1 |
| 3 | Green infrastructure corridor | 9 blocks of bioswales and porous paving | $1.8M | 1.5 |
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.
| Task | Good fit for AI? | Why |
|---|---|---|
| Finding and cleaning input data | Yes | Repetitive work with clear checks |
| Setting up model inputs | Yes, with review | Mechanical, but assumptions need checking |
| Solving the hydraulics | No | Use established solvers. A chat model should not guess flood depths |
| Running many storms and designs | Yes | This is where the time savings add up |
| Drafting reports and charts | Yes, with review | Every number should come from the model output |
| Choosing assumptions and sealing the work | No | Requires 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:
- 1Model. Terrain, land cover, rainfall and the city's real drainage infrastructure are pulled in automatically, and the model is prepared to run.
- 2Storm. It simulates multiple storm events and returns flood depths for each, using EPA SWMM, HEC-RAS and HEC-HMS.
- 3Risk. It identifies every building footprint in the sub-basin, plus emergency facilities and city assets, and counts what stands in damaging depths.
- 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.
- 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.
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.
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