Industries / Energy, water and waste / Environmental services / Stormwater management
Ponds and holding areas
Building ponds, basins, tanks and tunnels that hold storm runoff for hours or days and let it out slowly, so the pipes and streams below them are not overwhelmed.
5
Milestones achieved, from c. 1970 to 2013.
2
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
c. 1970
Ponds required on new sites
In the late 1960s and early 1970s US towns began to require on-site detention: each new development must hold back its runoff in a pond or tank so that the peak flow leaving the site is no higher than before it was built. Ben Urbonas of Denver's Urban Drainage and Flood Control District later wrote that it was new at the time and that developers resisted it because they had to set aside land, and by the mid-1970s every city and county around Denver was designing drainage with the district's 1969 Urban Storm Drainage Criteria Manual.
LessonRichard McCuen warned in 1974 that ponds sized one site at a time can raise peak flows downstream, a warning later measured (see 2005).
Credit
early adopting local governments; Denver's Urban Drainage and Flood Control District (now the Mile High Flood District)
2005
Many ponds, little effect
Valley Creek, a 62 km² (24 square mile) watershed in Chester County, Pennsylvania, west of Philadelphia, had more than 100 detention basins built since the late 1970s, each designed site by site to keep peak flows at pre-development levels for 2- to 100-year storms. Clay Emerson, Claire Welty and Robert Traver surveyed every basin and built a model calibrated to a USGS stream gauge; for six measured storms the basins cut peak flows across the watershed by only 0.3% on average, and could raise them. Managing the volume of runoff, not only its peak rate, worked best in their tests.
LessonPonds that each pass their own site's test do not add up to flood control for the stream they drain to.
Credit
Clay Emerson, Claire Welty and Robert Traver (Journal of Hydrologic Engineering, 2005)
2006
A river under the suburbs
The Metropolitan Area Outer Underground Discharge Channel in Kasukabe, north of Tokyo, takes floodwater from small rivers such as the Naka and Kuramatsu, whose low, bowl-shaped basin floods often and whose new suburbs send more rain straight into the rivers. It carries the water 6.3 km through a tunnel 10 m across and 50 m underground to a surge tank, where four pumps together lift 200 m³ a second into the Edo River. It was built from March 1993 to June 2006, and its first 3.3 km began working in June 2002.
Credit
Japan's Ministry of Land, Infrastructure, Transport and Tourism (MLIT), Edogawa River Office
2007
A road tunnel holds floods
Kuala Lumpur's SMART (Stormwater Management and Road Tunnel) is a motorway tunnel that also diverts floodwater from the Klang River around the city center, through a holding pond, a bypass tunnel and a storage reservoir. In moderate storms only the tunnel's lowest channel carries water; when flow of 150 m³ a second or more is forecast, traffic is cleared, which takes about an hour, and the road decks can be filled with floodwater, reopening within 2 to 8 hours or, if flooded, within 4 days. Planners expected to close the road about once every one or two years; a 2017 paper by the consultants who run its flood forecasts counted 82 closures and 212 smaller diversions from 2007 to 2017, about 8 and 21 a year.
LessonStorms needing the storage came several times more often than designed; the authors suggest the flood statistics used at design time did not match recent years.
Credit
MMC Corp and Gamuda (joint venture), with Malaysia's Department of Irrigation and Drainage and Highway Authority; flood forecasting by Entura/Hydro Tasmania
2013
A square that floods
Rotterdam's Benthemplein water square, designed in 2011 and 2012 by De Urbanisten with the city's engineering bureau and finished in 2013, is a public square that doubles as storm storage. Two shallow basins take water whenever it rains, a deeper one fills only in long rain and collects water from the surrounding area, and steel gutters carry the water across the square; most of the time it is dry and used for sport and play. Water management departments and innovation grants paid for most of it.
LessonStorage built as public space can be paid for partly from water budgets and partly as a place people use every day.
Credit
De Urbanisten; City of Rotterdam; Schieland and Krimpenerwaard water board
Watch for
Lowering a lake before storms
In Statesboro, Georgia, an outlet valve linked to weather forecasts now lowers Lake Sal by about 2 feet (0.6 m) when a major storm is predicted, making room for about 19.5 acre-feet (about 24,000 m³) of runoff, against an estimated 1.5 acre-feet from a conventional fix, at roughly 60% less cost than a regional detention pond. Independent work on such valves is mixed: a University of Tennessee study of four retrofitted basins in Knoxville (Novatech 2026) saw the effects in the stream, but its modeling suggested the largest effects of upstream development may not be reduced, partly because of where the ponds sat.
Achieved whenthe city or an independent party publishes measured lake levels and downstream flows for several storms before and after the retrofit.
Claimed by
Oldcastle Infrastructure, which built the outlet structure, April 2026
Every pond mapped from above
The US Geological Survey says it is training an AI model to map stormwater facilities, including swales, dry and wet detention ponds, bioretention and sand filters, from lidar elevation data and aerial images in 30 small watersheds of the Chesapeake Bay's Piedmont, with an open-source method to estimate each one's storage volume, because managers "often have incomplete information about the location and storage volume" of these facilities.
Achieved whenthe promised USGS data release and journal article appear, with accuracy checked against town records or field visits.
Claimed by
USGS, March 2026
Open
01
An exam for drainage reviewers
A public test for AI tools that check the drainage reports developers submit to towns. It holds real reports with the comments town reviewers sent back, covering runoff, pipe sizes and pond volumes, plus a scoring script and a table of each tool's results.
The briefCredit · unclaimed02
Testing forecasts for ponds
A published table, for several US cities, of how often a pond that lowers its water ahead of forecast storms would have done so for nothing (a false alarm), done so too late, or been caught full, scored against the rain that actually fell.
The briefCredit · unclaimed03
Ponds checked against permits
For one county, a comparison of each dry detention pond's storage today, measured from public lidar elevation data, with the volume its permit required, to show how many ponds still hold what they were built to hold. Lidar cannot see below water, so wet ponds are left out.
The briefCredit · unclaimed
Open 01 · the brief
An exam for drainage reviewers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test for AI tools that check the drainage reports developers submit to towns. It holds real reports with the comments town reviewers sent back, covering runoff, pipe sizes and pond volumes, plus a scoring script and a table of each tool's results.
Why it is still open
Companies now sell AI reviewers of drainage studies; Nomic, for one, says its tool flags missing detention content and mismatched drainage areas and storm events. Searches found no independent accuracy test of any such tool, only the sellers' own figures. Public material for a test exists: El Paso County, Colorado, posts submittals online with reviewers' notes marked on them, such as a 2024 pond certification.
Built with what exists
Drainage reports and review letters from county permit portals; a current AI model given each town's drainage manual, as in Enveda's test of Claude on drug-making records (see the Note "Seven more flagged"); hydro-tools to recompute runoff and pond volumes; Python for scoring.
First step
Collect 10 reports with their review comments from one county's permit portal and list each comment as a test item.
Done looks like
50 reports from at least five towns, with the reviewers' comments as the answer key; for each tool tested, the share of reviewer comments it caught and the number of false flags, published with the test.
Open 02 · the brief
Testing forecasts for ponds
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A published table, for several US cities, of how often a pond that lowers its water ahead of forecast storms would have done so for nothing (a false alarm), done so too late, or been caught full, scored against the rain that actually fell.
Why it is still open
Valves that draw ponds down on a forecast are now sold and installed (see Lake Sal above), but the forecast tests a search found cover reservoirs, such as a 2026 Karlsruhe Institute of Technology study of 30 small-to-midsize German flood reservoirs, up to 5 million m³, which found false alarms and misses explained the largest losses in performance. Searches found no test of forecast drawdown for town stormwater ponds against archived forecasts.
Built with what exists
NOAA's National Blend of Models forecasts, archived on Amazon Web Services' open data registry since May 2020; observed rain from NOAA rain gauges; pystorms, the University of Michigan's open set of stormwater control test cases built on EPA SWMM; Python.
First step
For one city, download a year of archived 24-hour rain forecasts and match each one to the rain measured at the nearest airport gauge.
Done looks like
For 10 cities and every forecast of 1 inch (25 mm) of rain or more in 24 hours since May 2020, the share that came true, the number of drawdowns that would have been wasted or late, and the storage gained or lost, published with code.
Open 03 · the brief
Ponds checked against permits
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
For one county, a comparison of each dry detention pond's storage today, measured from public lidar elevation data, with the volume its permit required, to show how many ponds still hold what they were built to hold. Lidar cannot see below water, so wet ponds are left out.
Why it is still open
Towns check volume once, when a pond is finished: El Paso County, Colorado, files engineers' certifications such as a 2024 pond whose as-built survey showed 0.453 acre-feet (559 m³) against 0.416 acre-feet required. USGS is building an open method to estimate storage from lidar for ponds it maps in 30 Chesapeake watersheds (see Watch for), and a 2021 USGS check of two older flood basins at Fort Gordon, Georgia, found that neither had the storage Georgia's rules require for 25- and 100-year storms, and that sediment could shrink it further. Searches found no study that re-measures many ponds years after certification against their permits.
epcdevplanstorage.blob.core.windows.netusgs.govpubs.usgs.gov
Built with what exists
USGS 3DEP lidar; county permit and certification records; QGIS or WhiteboxTools to compute the volume below each pond's overflow level; the USGS method once released.
First step
Download 20 pond certifications from El Paso County's records and check which are covered by 3DEP lidar.
Done looks like
100 dry ponds in one county, each with its permitted volume, its as-built volume where one was filed, and its lidar volume, with the method first checked on 10 ponds certified within a year of a lidar flight and agreeing within 10%.