Riprap, open-source flood and heat evidence for New York City
The flood and heat record for any New York City block, cited line by line.
Riprap joins street sensors, 311 complaints, flood maps and storm records for an address or community district into one page. For heat it reads satellite surface temperature, tree canopy, the Health Department's Heat Vulnerability Index and the Weather Service forecast. Rules or an open Granite model read your question and choose the evidence. Every sentence of evidence is written by code from a public record and cites its source and date.
This site has saved briefings. Run it yourself to ask your own question. Runs in seconds on a laptop: no GPU, no API keys.
Try a real question or place
Has the block around 90-01 183rd Street, Queens flooded since Hurricane Ida?
Yes.
The record this rests on is a FloodNet sensor with a verified flood event 22 m from this address. 2 FloodNet sensors within 600 m have recorded 6 flood events (each a series of depth readings above 10 mm at the sensor, FloodNet's definition) on 3 separate days since they were installed on 2023-10-26, the most recent starting 2026-08-20 22:57 UTC (every date here is a UTC day). By UTC calendar year: 2 in 2025 and 4 in 2026. FloodNet's API lists 8 more events labelled flood here that are still to be verified by a person; Riprap counts verified events only. The highest depth in FloodNet's record for these sensors in that period is 1172 mm (46.1 in) on 2026-05-20, in an event the API marks as verified by a person, at a sensor listed as "noisy" in FloodNet's API when this was read (2026-10-05); that is the sensor's status now, which the API does not give for the day of the event. 1 of the 2 sensors is listed with a status other than "good" in FloodNet's API when this was read ("noisy") and recorded 3 of the 6 events.
USGS surveyed 2 Hurricane Ida high-water marks within 800 m of this address; the highest stood 0.76 ft above ground. Nearest mark: Intersection of 182nd St. and 90th Ave., Jamaica, Queens (174 m away, 0.7 ft above ground).
88 NYC 311 complaints about flooding and sewer backups filed within 200 m of this location in the last 5 years (since 2021-10-06; eleven descriptors; other sewer complaints, such as odors or missing covers, are not counted): 48 sewer backup, 27 catch basin, 10 street flooding, 3 manhole overflow. A count of complaints is a count of reports filed, not of floods: a low count can mean under-reporting and not the absence of flooding, because the propensity to file a 311 request varies with income, language and demographics (studies of other 311 complaint types: Kontokosta, Hong and Korsberg, arXiv:1710.02452; Boxer, Hong, Kontokosta and Neill, Annals of Applied Statistics 19(2), 2025, doi:10.1214/24-AOAS2003). For the days of Hurricane Ida, which the briefing's 311 window does not reach: 5 NYC 311 complaints about flooding and sewer backups were filed within 200 m of this location from 2021-09-01 to 2021-09-03: 4 sewer backup, 1 manhole overflow. A count of complaints is a count of reports filed, not of floods, and a low count can mean under-reporting.
Saved briefing of . Each sentence has its source and date on the full page.
Real questions, answered from the record
Each answer below is a saved briefing. Open one and check any sentence against its source.
A whole district
17 of 36 QN12 subway entrances in a rainfall flooding category of the city's modelled stormwater map "Extreme Flood (3.66 inches/hr) with 2080 Sea Level Rise"
Community district QN12, Queens
36 MTA subway entrances in this area (each entrance read at its own point on the maps, with no buffer): 0 inside the 2012 Sandy inundation extent and 17 inside the DEP stormwater map "Extreme Flood (3.66 inches/hr) with 2080 Sea Level Rise", a modelled scenario (17 in a rainfall flooding category and 0 in its future high tides category, which is coastal tidal inundation projected for 2080 and not rainfall flooding).
Jamaica Center-Parsons/Archer (E J Z), Jamaica-179 St (F), Parsons Blvd (F), Sutphin Blvd (F), Sutphin Blvd-Archer Av-JFK Airport (E J Z)
In a rainfall flooding category of the modelled "Extreme Flood (3.66 inches/hr) with 2080 Sea Level Rise" map: I.S. 059 Springfield Gardens
Public housing in a district
2 BK06 public housing developments inside the 2012 Sandy inundation extent
Which NYCHA developments in Brooklyn Community District 6 lie inside the 2012 Sandy inundation area?
2 inside the 2012 Sandy inundation extent and 0 inside the DEP stormwater map "Extreme Flood (3.66 inches/hr) with 2080 Sea Level Rise", a modelled scenario
Inside the 2012 Sandy extent: RED HOOK EAST, RED HOOK WEST
A No about reports, not about flooding
No. No complaint was filed. That is not a finding that the place stays dry.
Have people near 355 Food Center Drive, Bronx reported flooding to 311 in the last five years?
0 NYC 311 complaints about flooding and sewer backups filed within 200 m of this location in the last 5 years (since 2021-10-06; eleven descriptors; other sewer complaints, such as odors or missing covers, are not counted; the 311 service answered and none matched).
A count of complaints is a count of reports filed, not of floods: a low count can mean under-reporting and not the absence of flooding, because the propensity to file a 311 request varies with income, language and demographics
The address is in the Hunts Point wholesale food markets, not on a residential block.
A city scenario for 2050
What does the 2050 stormwater scenario show at 400 Carroll Street, Brooklyn?
The city's stormwater flood map "Moderate Flood (2.13 inches/hr) with 2050 Sea Level Rise" shows the category "Future High Tides 2050" (coastal tidal inundation, not rainfall flooding; at the edge of the mapped flooding (within about 3 m)) at the point mapped for this address.
When the record is empty
Have the FloodNet sensors near 189 Atlantic Avenue, Brooklyn recorded flooding?
No FloodNet sensors deployed within 600 m of this address.
Heat at a wholesale market address
12.4°F warmer than the city's land average within 150 m of 355 Food Center Drive: the surface of the ground, not the air
Hunts Point, Bronx
The surface of the ground within 150 m of this address ran 12.4°F warmer than the city's land average over 18 clear summer Landsat images (surface temperature, not air temperature).
Its neighbourhood, Hunts Point, scores 5 out of 5 on the Health Department's Heat Vulnerability Index, a rank among neighbourhoods and not a measurement.
The address is in the Hunts Point wholesale food markets, where most of the ground is paved or roofed. The figure is for that ground, not for the neighbourhood where people live.
A forecast, quoted as the Weather Service's
Will it be dangerously hot this week at 90-01 183rd Street, Queens?
From the National Weather Service, as issued for the next 7 days; Riprap predicts nothing itself
The Weather Service wrote this for an area; it is not a prediction for a building.
Made for people who have to cite it: reporters quoting a block's record on deadline, community boards and council offices taking a district to the meeting in one printable briefing, planners checking which schools, subway entrances and public housing sit in a flood extent, and researchers who want the same evidence in code.
Open models, open problems
I fine-tuned open models for New York and tested each against a simple baseline. One, which maps paved, green and tree-covered land, is in briefings, labelled experimental, with its tested accuracy in every sentence. One, a surge forecast, lost to its baseline and was taken out. The results are published either way.
Paved, green and tree canopy
ExperimentalEstimates how much of each 10 m pixel is tree canopy, grass, paving or roof, in 2018, 2021, 2024 and 2026. A TerraMind fine-tune, trained on the city's 2017 six-inch map. A briefing quotes that map first, and the model's estimate follows it.
- Measured
- On squares it never trained on, it reads a typical district's paved share 1.7 points above The Nature Conservancy's 2021 map. Its citywide tree canopy reads 4.2 to 5.0 points below the city's 2017 map in three of its four yearly maps.
- The open problem
- Maps of different summers differ by more than two images of one summer, so it cannot yet show change between years. Telling real change from the images is the next result to earn.
Storm surge at the Battery
Negative result, off by defaultForecasts how far the water may run above the predicted tide over the next four days. A Granite TTM r2 fine-tune. It is out of default briefings since 2026-10-05.
- Measured
- On 639 held-out four-day windows from January 2025 to October 2026, its mean error was 0.115 m. Damped persistence, a one-line rule, scored 0.108 m and beats it. The tide table alone scored 0.167 m.
- The open problem
- It foresaw 1 of the 23 windows that reached flood stage, which are 5 distinct events. A forecast that a simple rule beats adds nothing to the Weather Service forecast a briefing already quotes. The model card on Hugging Face was corrected to say so on 2026-10-05.
A third model, a satellite water layer, was retired on 2026-10-02 after two tests, one on Hurricane Ida and one on coastal and tidal floods that FloodNet sensors recorded at the moment of a satellite pass: in both it found recorded floods no more often than chance.
Built on 34 public sources
City, state and federal records, read live or from their published files and sorted into five Stones: 25 for a flood briefing and 11 for a heat briefing, two of them shared. No commercial data and no scores. Each record has its own portal; Riprap reads them together for one place and links back to each.
- Hazard maps and storm records Cornerstone
- FEMA's effective and preliminary flood maps, the city's four modelled stormwater flood maps for today, 2050 and 2080, the 2012 Sandy inundation extent, USGS Hurricane Ida high-water marks, USGS terrain For heat: Landsat surface temperature, the NYC Health Department's Heat Vulnerability Index and heat illness emergency visits, the city's 2017 land cover map (tree canopy)
- Places at risk Keystone
- NYC public schools, MTA subway entrances, NYCHA developments, hospitals, NYC Planning's floodplain counts; DOB construction permits only when a question asks For heat: NYC Parks spray showers and pools
- Live sensors and complaints Touchstone
- FloodNet street sensors, NYC 311 complaints about flooding and sewer backups, NOAA tide gauges, USGS stream gauges, National Weather Service observations For heat: Weather station records of 90 degree days, the latest air temperature at the nearest station
- Forecasts and projections Lodestone
- National Weather Service alerts and water-level forecasts, NPCC4 sea-level projections For heat: The National Weather Service forecast and heat alerts, NPCC4 heat projections
- How answers are written Capstone
- Code turns each record into one cited sentence, and rules set the answer's lead (Yes, No, a count). An optional open Granite model may route a question the rules do not match and choose among those sentences; it never writes one, and code checks its choice.
FloodNet data is licensed CC-BY-NC-SA 4.0 and the NPCC4 projections CC-BY-NC 4.0; the other sources are public domain or open. See every source and licence.
Where a language model can run, what the data cannot say, and how Riprap was built
For developers and AI agents
Every briefing is also data. Over HTTP you get the full result as JSON with its trace. Over MCP, seven tools return each finding with its figures, its source URL and the date of its data, plus the list of sources that did not answer. No language model is needed.
The MCP tools
- get_evidence
- get_district_summary
- get_briefing
- nyc311_flood_requests
- plan_query
- list_sources
- get_citation
# Get the code and its data (needs uv and Git LFS)
git clone https://github.com/msradam/riprap && cd riprap
git lfs install && git lfs pull
# Start the server: no GPU, no API keys
uv run uvicorn web.main:app --port 7860
# A district briefing as JSON
curl -s "http://localhost:7860/api/agent?q=QN12"
# The same evidence for an AI agent, over MCP
uv run riprap-mcpRiprap reports evidence, not advice. For flooding right now, FloodNet sensor dashboard, National Weather Service, Notify NYC; an official flood zone, FEMA Flood Map Service Center; insurance and what to do at home, FloodHelpNY; people and housing in the floodplain, NYC Planning Community District Profiles; what to do in extreme heat, NYC Health Department, National Weather Service heat safety, NYC Cooling Center Finder; maps to explore, Rainproof NYC Flood Map.
Riprap is not an alert or emergency service. For emergency alerts, sign up with Notify NYC; to report flooding or ask the city for help, use 311; for flood insurance questions, see FloodHelpNY.
Riprap is independent and open source, and it is not a City product. It is not endorsed by or affiliated with FloodNet, New York University, the City University of New York, FEMA, NOAA, USGS or the City of New York. The City publishes its open data for information only and does not warrant its completeness, accuracy, content or fitness for any use.
Build it with us
Riprap is built by Adam Munawar Rahman, open source under Apache-2.0. I'd like help with three things: a pilot in a studio course, council office or engineering firm; a second city; and better ground truth for the surge and land-cover models.
Run a pilot
A newsroom, council office, university lab, engineering firm or port can run its own copy on its own places and asset list. Tell us what you would ask it.
Bring your city
New York City runs in production. Chicago, Seattle and Albany run as experimental ports. A city with a Socrata portal needs only manifests.
Bring your data
Add your own sites, such as a list of firehouses, to every briefing with one YAML manifest and no fork.
Improve a model
Each model has a backtest and a baseline. When a test reruns, the briefing's accuracy sentence updates from its results.