The network keeps growing
Two sensors went up in November 2020. There are now, spread across all five boroughs. Each new one starts watching a corner nobody was measuring before.
Since November 2020 a growing mesh of sensors has been measuring exactly how deep New York City streets flood, how long the water stays and where it keeps coming back. This is the whole record: every flood the network has caught, what it can tell you and what it cannot.
Any reading of this record has to account for both of these, and neither one on its own answers the question people ask.
Two sensors went up in November 2020. There are now, spread across all five boroughs. Each new one starts watching a corner nobody was measuring before.
Measured floods climbed from in 2020 to at the peak. The network now catches water that a smaller one would have missed.
Those two curves climb together because the second is largely produced by the first. To ask a weather question you have to hold the corners fixed, which is what the next section does.
corners have been in the ground since before 2023 and are still reporting today. Holding that set fixed removes the growth of the network from the picture. Here is the record at those corners, on two different measures, with nothing smoothed.
Counting events alone treats a film of curb water the same as water over a car's hood. Every measured flood here is graded by how deep it got, and by how many hours the water stood above a foot, the point where a street turns impassable.
Every sensor, sized by how many floods it has recorded in its lifetime. Colour says what drives the water there: rain that cannot drain, or tide that pushes up through it. The gray dots are sensors that have never recorded a flood at all.
These are measured peaks. Depths above about two feet mean water over the wheels of a parked car.
| Corner | Borough | Date | Peak depth | Lasted |
|---|
Sources. Two NYC Open Data sets published by FloodNet, the research collaboration of CUNY, NYU and city agencies that builds and runs the sensors: flood events (aq7i-eu5q) and sensor deployment metadata (kb2e-tjy3). Everything on this page is computed by build_data.py straight from those two files. Nothing is modelled, estimated or smoothed. The record runs from .
Why the stable-corner section exists. The network grew from 2 sensors to , so a citywide yearly total mostly tracks deployment rather than weather. The fix is to hold the sensor set fixed. That is the only comparison on this page that attempts a like-for-like reading across years.
Survivorship. An earlier version of this page divided by all 34 sensors installed before 2023. Eleven of those have since been retired, and a sensor that is gone records no floods, so leaving it in the denominator manufactures a decline. The cohort here is the sensors that were in the ground before 2023 and are still in FloodNet's live deployment list, which raises the per-corner figures by roughly 40 percent against the flawed version. This is a proxy for uptime, not proof of it: a surviving sensor could still have been offline for stretches, and the published event record cannot show that.
What the stable-corner numbers do and do not support. They measure what those corners experienced. They are not a citywide climate trend, for three reasons. The sample is small, and in any given year only of the corners recorded a flood at all, so a handful of chronic spots move the totals. Three years is weather, not climate. And these particular corners were chosen early precisely because they were known trouble spots, often right after a bad flood, so some regression toward calmer years is expected regardless of the weather. The two measures also disagree about which year was worst.
Placement is not random, anywhere on this page. FloodNet puts sensors where flooding is expected. A corner with a big circle is a place the city chose to watch and where water duly arrived. A block with no sensor is unmeasured, and the absence of a dot is never evidence of absence of flooding.
Definitions. A flood event is FloodNet's own detection, with an observed floor of 0.43 inches held for 5 minutes. Depth is the peak of the event. Deep-water hours count time above 12 inches. Tidal versus rainfall follows the tidally_influenced flag on the sensor, so it describes the driver at that location rather than the cause of a particular storm.
Confidence. High for every raw measurement: depth, duration, location, counts. Medium for the stable-corner comparison, which depends on the survivorship proxy above and on a small sample. This page makes no claim about a citywide trend in flooding.
build_data.py directly from the two NYC Open Data sets named above; none are AI-generated or estimated. This is an independent civic-data project, not affiliated with FloodNet or the City of New York. FloodNet's own dashboard is at dataviz.floodnet.nyc.This project also ran a live companion page reading FloodNet's real-time endpoint. It was retired in August 2026 after FloodNet confirmed that endpoint is deprecated, carries no quality control and is being withdrawn. Why the live page was retired.