Flood warnings Ireland: what 461 State gauges say, checked
Flood warnings for Ireland come from three State bodies, each publishing on its own website. We built a free watchdog that reads all three and emails you when water moves near an address you choose. It covers 461 river gauges and 159 flood-zone datasets, and it reads the live Met Eireann warning feed. In 15 days it has run 1,227 checks, logged 50 warnings, and delivered one email. Below is the working: what the data covers, what it misses, what breaks, and how to use it this week for nothing.
What we actually measured
Every figure below was read off the running copy of the project on our build machine, at /data/work/floodwatch, and off the public instance at floodwatch.rennwick.com. Both were checked on 26 September 2026. When a number comes from reading the code instead of running a query, we flag it, since the two are different grades of evidence.
We took the measurements between 02:40 and 02:56 UTC on 26 September 2026. The local database held 1,227 completed checks, the last at 02:45:47 UTC. The public instance runs its own timer and its own database, and it reported a check five minutes earlier. Figures like these drift. Re-run the queries in the How to check us section a week from now and the counts will be higher, while the per-day averages should hold.
Here is the shape of it, in numbers.
| Measure | Value on 26 September 2026 | Where it comes from |
|---|---|---|
| Checks completed | 1,227 | ticks table, first at 00:36:13 on 12 September, last at 02:45:47 on 26 September |
| Median gap between checks | 15.0 minutes | calculated from the started column of every tick |
| Median checks per day | 95 | count of ticks per calendar day |
| Gauges stored and scored | 461 | gauges table and gauge_latest table, one row each |
| Raw readings per poll | 2,008 | highest gauges_n in the tick log; the poll returns every sensor, we keep the level sensor |
| Datasets found on the open data portal | 159 | data.gov.ie flood query, cached in data/flood_zones.json |
| Map layers wired for a point check | 17 (16 with a bounding box) | data/flood_zones.json and the coverage command |
| Distinct warnings logged | 50 | warnings table; 8 were active at extraction |
| Alert emails composed | 26, plus 11 suppressed by the daily cap | alerts_log table, 37 rows |
| Alert emails delivered | 1 | outbox holds 26 queued files and one .eml.sent |
| Subscribers | 1 | subscribers table; it is the person who built it |
| Unit tests passing | 34, in 0.895 seconds | python3 -m unittest discover -s tests |
Three rows deserve a closer look. First: the service holds 461 gauges, but a poll returns about 2,008 readings. That is not a defect. A station can publish temperature, rainfall and several water-level sensors under a single reference; we keep the level sensor and throw the rest away. Second: 26 alert emails were composed and one went out. That gap runs through everything below. Third: there is one subscriber, and that subscriber wrote the code. None of this has met a public audience, and we will not claim otherwise.
Flood warnings Ireland: what this watchdog actually reads
The service watches; it does not forecast. Four upstream sources feed it, every one of them free and keyless, so anybody can pull the same feeds with no account. One file declares them, floodwatch/config.py, and you can read that file in a minute. Here is what it pulls.
- Met Eireann weather warnings, from the live JSON at
prodapi.metweb.ie/warnings. It is the same warning text carried on Met Eireann's warnings page, handed over as data: a national section, a marine one, an environmental one and a Northern Ireland one, each with a level, a type, the regions it touches and a window of validity. We refetch it no more often than every four minutes. - OPW real-time water levels, from waterlevel.ie, the Office of Public Works gauging network. From it we take every station's reference number, name and coordinates, the latest reading on each sensor, and a downloadable CSV series for each station. The station list is refetched daily, the latest readings every ten minutes.
- Council and OPW flood-zone maps, found through the flood datasets on data.gov.ie, queried as ArcGIS REST services, which are the map servers the councils run. For a given address we ask whether it sits inside a mapped flood zone, and we use the polygon count at that point to say.
- OpenStreetMap geocoding, through Nominatim, which turns a town or an eircode lookup into coordinates. Their usage policy sets a modest request rate, so we cache the results to disk after the first lookup.
Two absences stand out, and both were chosen. The Environmental Protection Agency is not on the list. We went looking for a live EPA feed that would add flood signal and found none better than what we already read. The EPA's water work is mainly monitoring, assessment and catchment characterisation, published through catchments.ie and the EPA freshwater and marine monitoring pages. That material helps you understand a river and what pressures it carries. It is not a live gauge feed, and Ireland's live level gauges belong to the OPW. Better to say that plainly than to pad the list.
The second absence is larger: this system carries no forecast. We read warnings that forecasters have already pushed out, and we read water levels that have already occurred. No rainfall forecast sits here, no rainfall-runoff model, and nothing from the OPW's National Flood Forecasting Service is connected. So when we call a gauge rising, we mean it rose. Unless Met Eireann has told everybody, we cannot say where the water will be in six hours; by then the warning is the news and we are just the messenger with a mailbox.
The numbers on the board: 159 datasets, 17 layers, 461 gauges
The public board carries three panels. Panel one is the national picture: the highest warning level in force, the count of warnings, and how many of those are marine warnings, which count as warnings but rarely as flood warnings. At extraction the national panel showed one active national warning and six marine warnings.
Panel two is 26 county cards, one for each county in the Republic. Each card shows the highest warning that names that county, and any advisories that are not warnings. Advisories cover things like a blight warning, which has a weather cause and no flood effect. On 26 September 2026 four counties carried a yellow rain warning and 22 carried nothing. The rain warning covered Cork, Kerry, Waterford and Wexford, valid from 04:00 to 11:00 on Sunday 27 September. Six marine warnings sat beside it, four of them small-craft warnings and one a gale warning. The board counts marine warnings so a reader can see the weather moving, but a small-craft warning never enters a county's flood score.
Panel three is the gauges: 461 river stations, each with its latest level, the change over six hours, the change over 24 hours, and a trend word. On our build copy at extraction, 446 of the 461 readings were under an hour old, and the median reading age was 15 minutes. That is a healthy feed. The three that were older are listed in the section on what breaks.
The flood-zone layer is the part that sounds simple. It is not. A search of data.gov.ie for flood datasets returns 159 datasets, published by the Office of Public Works, the Marine Institute, the Geological Survey, the Central Statistics Office and 12 local authorities. We turned 17 of them into queryable point-in-polygon endpoints. Of the 17, 16 declare a bounding box, which lets us skip a layer fast when an address is nowhere near it. Eight layers name a county in their metadata: six from Galway and two from Roscommon. The county tag decides ranking, because the ranking prefers a layer whose county matches yours, and nine layers can never be preferred because they say nothing about where they are.
In practice, a point check on an address in Leitrim queried five candidate layers and reported that the address was not inside a mapped council flood zone. A second query through the public instance reported six layers checked. Both answers are honest, and both are narrower than they sound. Councils have published far more flood mapping than 17 layers, much of it as PDFs, shapefiles and drawings that never turned into a queryable map service. If your house sits on a council's paper flood map, and nobody ever digitised that map, our point check will miss it. For that, the reference is still the OPW flood maps on floodinfo.ie, and you should read them alongside anything we say.
How often it polls, and how it notices a stale reading
A systemd timer does the polling, firing every 15 minutes, with a 30-second accuracy allowance and a first run three minutes after boot. The unit file runs to about ten lines. Every run fetches the sources, updates the database, scores each subscriber, drops any alert email into an outbox directory, and writes one row to the tick log: warnings seen, readings seen, alerts produced, and any error text.
That last column is the whole monitoring story, and we should be exact about how thin it is. There is no separate uptime monitor, no alert on the monitor, no paging. What you get is a table you can read, and a status page that renders the last ten ticks and the last ten alerts from it. Stop the timer and the tick log stops with it, and nothing shouts. The 31.1-hour gap described below happened that way, and a human reading a table noticed it.
Freshness is tracked per reading, not per station. Each time we store a level we work out its age in hours from the timestamp the gauge itself published, and that age lands in the database. The board shows the reading and its age, so a station that stopped reporting yesterday shows yesterday's number, plainly dated, instead of being repeated as today's. Three gauges were in that state at extraction. The oldest, Killykeen Forest Park, station reference 0000036083, last published a level at 12:15 on 13 March 2025 and still sat on the board, carrying a reading age of 13,478 hours, which is 561 days. The other two, Moneycashen and Ballynary, were 152 hours and 51 hours stale.
That is a design decision, and it carries a cost. We would rather show a 561-day-old level, clearly dated, than drop the station and leave a hole a reader cannot see. A stale reading is still a reading, though, and any score that uses it is moving on old information. Taking stale stations out of the score sits on the list of unfinished work.
Two limits protect the upstream services from us, since we are a guest on free infrastructure. The station list is cached for 24 hours, the latest readings for 10 minutes, warnings for 4 minutes, and each station's CSV series for 15 minutes, so a burst of page views never turns into a burst of requests to the OPW. When a trend needs a series, we fetch at most 60 series in one check, and we look at no more than four gauges per subscriber, ranked by distance. A point-in-zone check queries at most ten layers. The point of those limits is laid out on the waterlevel.ie API page, which sets the republication rules for OPW data, and we follow them.
How the score is built, and what the thresholds are
Each area gets a single number from 0 to 100, and a single colour. The bands run 0 to 19 green, 20 to 44 yellow, 45 to 69 orange, 70 and above red. Three things build the number, and the weights are fixed constants in floodwatch/risk.py, not learned and not tuned.
- Warnings. A warning is worth 12 points, multiplied by its level step (green 0, yellow 1, orange 2, red 3) and by a flood-relevance weight for its type. Rain carries the highest weight at 3.0, thunderstorm and storm at 2.0, wind and snow-ice at 1.0, fog at 0.2, and heat, drought and blight at zero. A warning whose weight falls below 1.0 is dropped from the flood calculation altogether, which is why a blight advisory never lifts a flood score.
- Gauges. A gauge that rose past the sharp-rise threshold is worth 18 points. A gauge sitting above its own 24-hour median adds 8 more. The gauge block is capped at 45, so one river on its own can never carry an area into the red.
- Flood zones. An address inside a mapped council flood zone adds a flat 15 points. That is a baseline, not a risk assessment. It says the address sits where the council drew a flood-prone area, which is a planning fact and not a live condition.
Two caps in that list are worth stating out loud, because they shape what the service can say. The warning contribution is capped at 60 points, so one warning at any level cannot, by itself, put an area into the red band. On our copy, a lone orange rain warning scored 60 out of 100, a lone yellow rain warning scored 36, and a lone red rain warning also scored 60, capped. Reaching red takes a warning plus something moving in the water, or a very active river over a flood zone. The second cap is the gauge block at 45 points, so three rising gauges score the same as one.
The sharp-rise test is the heart of the gauge work. A rise counts as sharp when it reaches at least 0.15 m inside the window, or at least 10 per cent of the day's range, whichever bar is higher. When the window holds fewer than two points, the code refuses to report a trend instead of widening the window until the answer looks like something. That line is there because the alternative, quietly opening a wider window, invents movement that never happened in the period the reader thinks they are watching. A six-hour rise of 0.30 m or more gets the label orange river alert, which is a different label from a Met Eireann orange warning, and the email subject line never mixes the two.
Alerts fire on a change of state, not on the state itself. A yellow area that stays yellow sends you nothing. You get an email when the level changes, up or down, and a clear email when the last alert lifts. No subscriber receives more than four alert emails in a day; past that the transition is logged as suppressed, not sent. That cap produced 11 of the 37 rows in the alert log, which is a lot of suppression for two weeks, and it says the score is jumpy along the green-yellow boundary.
What an alert actually said, from the log
Instead of describing the email in the abstract, take one of the 26, pulled from the queue file written at 04:27 on 12 September 2026. The subject line read "FloodWatch Ireland: Orange river alert - Leitrim". The plain-text body said the river level was rising near New Bridge in Manorhamilton, that the gauge was up 0.394 m over six hours to 0.901 m, that it sat 11.4 km from the subscriber's area, and that the 24-hour change was 0.335 m. The flood-risk score was 77.0 out of 100, the red band. The email closed with the source list, a one-click unsubscribe link, and one line saying the service is free and made by Rennwick.
That is the whole product. A gauge reference, a place name, a rise in metres, a distance, a score, and a way out. It will not tell you whether your kitchen floods. It tells you a river 11 km upstream rose 39 cm in six hours, which someone in Manorhamilton might want to know at half four in the morning, and which otherwise sits buried in a table on a State website.
The same log shows what a wet fortnight looks like. Three orange river alerts landed, on 12, 14 and 18 September, all for the one area. Eight yellow-level updates arrived, and 15 green ones; a green update is either an all-clear after a warning ended or a low-level arrangement of warnings that never reached yellow. Then comes the mail rail, the unfinished part, which gets its own section.
What is running today, and what is not switched on
The reading and scoring half is finished and running. It has sat on a 15-minute timer since 12 September 2026, completed 1,227 checks, logged 50 warnings, kept 461 gauges current, and served a public site that answered every page we asked for on 26 September. The public address is floodwatch.rennwick.com, and the pages live there: the home board, a live risk view, a data sources page, a status page, a sign-up form, a one-click unsubscribe, a privacy page, and three machine-readable endpoints, /api/risk.json, /api/warnings.json and a small /widget.js you can drop into another website.
The delivery half is not finished, and the status page says so in as many words: mail mode, dry_run. Each alert email is composed, written to disk, logged, and then nothing collects it. One exception exists. On 12 September a single email was copied to a separate host that holds mail credentials and sent through that host's mail relay, the only alert a human has ever received from this service. The queue holds 26 unsent files right now. Sign up today and you join a test list, not a live alerting list, which the website also states.
The hold is not technical vanity. Sending mail to strangers from a small firm's domain needs a sending reputation, a bounce address, a complaint path and a privacy position, and we would rather build those on a list of one than find the gaps on a list of 1,000. The double opt-in is in place: a sign-up creates an unconfirmed row, a confirmation link sets the confirmation timestamp, and only confirmed rows get scored. The unsubscribe link carries a token and works in one click. What is missing is the sending rail and the routine that would run it.
What breaks, with counts
The service is a watchdog. A watchdog that hides its own faults is a press release, so here is the fault log, biggest first.
The mail rail, already covered. 26 emails composed, one delivered. The half that works is the half nobody can feel yet.
Name resolution failures, 73 checks. Of the 1,227 checks, 73 logged an error, or 5.9 per cent. All of them share two causes: the request to the Met Eireann warnings endpoint failed after three attempts with "Temporary failure in name resolution", and the request to the OPW latest-readings endpoint failed the same way. That error text means the machine could not resolve a hostname, which is a network or resolver problem on the box, not a problem at either State service. Of those 73, 71 also stored zero gauge readings, so on those runs the board worked from whatever it last knew.
One very bad day, 19 September. Of the 71 empty checks, 49 landed on a single day. That is roughly half of that day's scheduled runs with no gauge data at all. No retry queue survives the run, no exponential backoff, no notification, so the first anyone knew was a gap in a table. A three-try retry inside one run does nothing when the resolver itself is the problem, which is the lesson from that day.
One 31-hour hole. The longest gap between checks ran 1,865.8 minutes, from 05:12 on 12 September to 12:17 on 13 September, the first day of the service's life. Five gaps passed 20 minutes in total, and one passed an hour. The other four are ordinary timer jitter of 20 to 32 minutes. The 31-hour hole was a scheduling failure, not a data failure.
Stale gauges still scored, three of them. Killykeen Forest Park at 561 days, Moneycashen at 152 hours, Ballynary at 51 hours. A reader who looks at the age column can see them, and they still feed the distance calculation for anyone whose nearest gauge is one of the three.
Nine flood-zone layers with no location. Nine of the 17 wired layers carry no county in their metadata, and one of the 17 has no bounding box, so every point check has to try it. The ranking that decides which layers get queried first, once more than ten are candidates, has nothing to work with for those nine.
The score is jumpy near the green-yellow boundary. 11 suppressions in a fortnight, against a cap of four emails a day, means the score crossed its bands often enough to hit the cap. A subscriber never sees this, because the extra emails never arrive. Anyone trying to understand the score should read a yellow band as closer to noise than the label suggests.
One subscriber, one area. Every alerting number in this article describes one address in Leitrim and one mailbox. No second user, no second county, no rural address whose nearest gauge is 40 km away has exercised any of it. Until one does, treat the alerting behaviour as demonstrated, not proved.
What the system does not cover
A free watchdog over State data should mark its borders, because the failure mode of a service like this is not being wrong. It is being quietly partial. Below is what sits outside.
Drainage and flood relief schemes. The OPW runs Ireland's arterial drainage and flood relief works, and no feed we read carries the state of a scheme's channels, sluices and pumps. Work on a scheme near you, whether maintenance, desilting or repair, changes your flood behaviour, and FloodWatch Ireland will never mention it. The Office of Public Works publishes scheme information directly.
Insurance. This system says nothing about insurability, premiums or claims. Flood-zone membership and insurability are related but not identical, the mapping insurers use is commercial, and no part of our board replaces a call to an insurer. If a zone polygon we report changes how a broker prices your cover, take it up with the broker, not with a website.
Planning. We read flood-zone map layers. We do not read planning applications, planning decisions or any local authority's development plan text, and a point-in-zone answer is not a planning opinion. Buyers and builders should consult the county development plan and the planning register, and ask the council. Misleading someone who is not paying attention is the commonest way a tool like ours goes wrong.
Small catchments and short rivers. The OPW network is large, but it is a network, and not every stream has a gauge on it. A town flooded by a small, fast-responding tributary may have no gauge nearby, in which case the nearest gauge on a page could sit 20 km away on a different river, and its rise says something different from what the bridge in town sees. We show distance for exactly that reason, and four gauges per area is a limit, not a claim of coverage.
Groundwater, turloughs and coastal flooding. Winter flooding from high groundwater and from turloughs behaves nothing like a rising river, and it shows on our board only where a council drew it as a flood zone. Coastal flooding, storm surge and wave overtopping are not modelled here either. We count marine warnings, small-craft and gale warnings among them, but a marine warning never enters a flood score, so the coastal risk a reader cares about is not what the number describes.
Northern Ireland. The county list is the 26 counties of the Republic, and the board renders 26 county cards. The warning feed does carry a Northern Ireland section and we read it, but it scores into no county card and there is no cross-border view. From this board, a reader in Derry, Enniskillen or Newry gets nothing.
Everything between the gauge and your door. We do not know your ground level, your floor level, your drainage, your access road, or where water reaches your property. A river rising 11 km away is information. Whether it matters to you is a judgement about a specific place, and the system deliberately leaves that judgement to you.
Small print on the score itself. The 0 to 100 number is a house construction out of three weighted parts, with two caps that change what it can say. It is not an official measure, it does not compare to Met Eireann's warning colours, and an orange score is not an orange warning. Met Eireann issues warnings. The OPW measures water. We add a number so a person can sort a list.
How to use it this week, for free
Smallest useful step first. Stop at any line.
- Look at your county card on the board. Open the FloodWatch Ireland board and find your county. It shows the highest warning in force that names your county, and whether anything is happening at all. That takes 20 seconds, needs no sign-up, and on most days the honest answer is nothing is happening.
- Check a specific place. Enter a town name in the place check on the same site, and it returns the flood-risk score, the nearest gauges and their levels, six-hour and 24-hour changes, and whether the point falls inside a mapped council flood zone. If you are weighing up a house or a yard, this answers the question in five minutes, and it gives you the distance to each gauge so you can judge whether that gauge is about your river.
- Open the source pages alongside it. The Met Eireann warnings page is the authority on warnings, waterlevel.ie on levels, and the OPW flood maps on floodinfo.ie on flood extents. Ours is a convenience layer over those three, and you should read the originals yourself for anything that matters.
- Sign up for alerts if you want the email side. Enter a town or eircode, an email address, and a radius. The default radius is 20 km. You get a confirmation email first, and nothing is scored until you click it. Right now that puts you on a test list, not a live alerting list, and the site will say when that changes.
- For a business, watch a fixed set of sites. A depot, a yard, a farm, a shop with a basement, a distribution route. One address will not do; the useful pattern is the one we use internally. Watch the specific places you would have to act on, and check the board once each morning in a wet week. Water rises in hours, so for most small operations a daily look plus an email when a nearby gauge moves will do.
- For a developer or an agency, take the data.
/api/risk.jsonis the whole board as JSON, with per-county levels, the warning list and every gauge and its changes, and/widget.jsis a small embed for a site or a dashboard. Both are free, both hold a stable shape, and both carry the OPW attribution string the licence requires.
One rule covers all six: a flood warning from Met Eireann, or an instruction from your local authority, outranks anything on our board. Where the two disagree, the State is right and we are a badly written table. As an example of what a council publishes, Leitrim County Council's flooding page carries local advice and contact routes that no aggregator can replace.
How to check us
Anyone with the project directory can reproduce every number in this article, and two of them need only a browser. Here is how, in the order we would do it.
Check the live service. Fetch the machine-readable board and read the generation timestamp, the national summary and the count of counties. Fetch the status page and read the last ten checks and the last ten alerts, with the mail mode stated at the top. If the last check is more than half an hour old, the service is behind, and everything below it should be discounted.
Check the upstream sources directly. Two commands matter here: a fetch of https://waterlevel.ie/geojson/latest/, which returns the current reading for every OPW sensor, and a fetch of https://prodapi.metweb.ie/warnings, which returns the live warning set. Compare their counts with ours. If the OPW returns 900 readings and our board says 461, remember that we keep one level sensor per station, so the raw count will always beat the station count.
Check the counts from the database. In the project directory, one query answers most of this article: the number of ticks, the first and last tick timestamp, the number of gauges, the number of gauges whose reading is older than a day, and the number of alerts by level and mode. That file holds ten tables: subscribers, alert_state, alerts_log, warnings, gauges, gauge_latest, geocode_cache, zone_cache, ticks and kv.
Check the logic against the code. The weights, bands and caps are constants in floodwatch/risk.py, and the source URLs and thresholds sit in floodwatch/config.py. Both files are short. If you think a threshold is wrong, read the number yourself instead of taking ours, and run the scoring function on a case you invent to see what score it returns.
Run the tests. The project ships 34 unit tests, covering the warning parser, the gauge parser, the county mapping, the score, the store, the email renderer and the flood-zone registry. They ran in 0.895 seconds on 26 September 2026. They exercise the pure functions and not the network, so a green suite means the logic works, not that the sources are up.
Check the claims yourself. We keep a register of factual claims, with the command behind each one, at the fact provenance register. Every figure in this article entered it with the query that produced it, so a reader who wants to argue with a number can start from the command and not the sentence.
What this means for you
If you live or work near a river in the Republic of Ireland, you can now be told when it moves, free, instead of checking a State website and interpreting a table. The information was always public and always free. What was missing was a mailbox and a threshold. That is the whole contribution, and it is worth having on the mornings when it matters.
If you run a business with a fixed site that floods, the pattern is mechanical: name the sites you would have to act on, watch those, and keep a manual morning look during wet spells. Do not expect the score to tell you whether to move stock. Expect it to tell you a gauge upstream rose 39 cm in six hours, and then use your own knowledge of the yard.
If you are the kind of reader who cares how public data gets used, here is the pattern we sell: take a State dataset that is genuinely good, watch it on a timer, publish what it says and what it does not, and keep a log of your own faults where a reader can find it. The board is not the interesting part. A small team in Leitrim wired 159 open datasets, 461 gauges and one warning feed together in a fortnight, ran it on a box that costs a few euro a month, and now answers to anyone who reads the source files.
What we do not know
Whether the alerts would fire correctly at scale. Every alerting number in this article comes from one address in Leitrim over 14 days. We have not seen how the score behaves for a coastal town, a city address with no river within 20 km, or a site whose nearest gauges all sit upstream on a different river. We do not know the false-positive rate, because until the mail rail is live nobody is receiving the emails to complain about them.
Whether the sharp-rise threshold is right. The 0.15 m floor and the 10 per cent of daily range rule were chosen by reading series, not by testing against known flood events. Nobody has taken the five or six floods that caused real damage in Ireland in recent years and asked whether this system would have fired early, on time, or late. That backtest is the single most valuable piece of work still missing, and it needs historical series, not live data.
How much of the country the flood-zone layer really covers. We know 159 datasets exist and 17 became queryable layers, and we know nine of those carry no county metadata. We do not know what share of Irish flood mapping sits trapped in PDFs and drawings that no point query can reach. Our guess is most of it, and it is only a guess.
What the 561-day-old reading should be doing. We chose to keep stale gauges visible with their age. Whether they should instead be excluded from scoring, flagged on the board, or dropped from the distance calculation entirely is a judgement we have not tested against a reader's expectations.
Whether the resolver failures are local or typical. The 5.9 per cent error rate sits on our box, on our network, and could be a local DNS arrangement with nothing to do with the State services. The counters here cannot tell "this box has a flaky resolver" from "these two endpoints are unreliable under load".
Whether anyone wants the email. One subscriber, who built it. The honest answer to "is this useful" is that a free service with one user has not answered the question, and we will not pretend that 1,227 checks is the same as demand.
Sources
- The Met Eireann live warnings JSON was checked 26 September 2026. It feeds every county card and we read it every four minutes.
- The Met Eireann warnings page was checked 26 September 2026. It holds the same warnings in human-readable form and is the authority on them.
- The OPW water levels on waterlevel.ie was checked 26 September 2026. It carries the station list, the latest readings and a per-station CSV series, and it is the source of all 461 gauges.
- The waterlevel.ie API and licence page was checked 26 September 2026. It sets out the republication policy that limits us to station references 1 to 41,000 and requires attribution.
- The data.gov.ie flood datasets was checked 26 September 2026. This is the open data search that returned the 159 flood datasets behind the zone layer.
- The OPW flood maps on floodinfo.ie was checked 26 September 2026. It is the State's own flood map viewer, and the reference for the flood extents we cannot digitise.
- The Office of Public Works was checked 26 September 2026. It owns the gauging network and the arterial drainage and flood relief schemes we do not cover.
- The EPA catchments.ie and the EPA freshwater and marine monitoring pages were checked 26 September 2026. They are named because they are the obvious candidate sources we decided do not add live flood signal, and a reader should see that decision.
- The Nominatim usage policy was checked 26 September 2026. It sets the rate limits and attribution rules for the geocoding that turns a town name into coordinates.
- The Leitrim County Council flooding page was checked 26 September 2026. It is an example of the local-authority advice and contact routes that outrank anything on our board.
Revision log
v1.0, 26 September 2026. Author: Chris Churchill, Rennwick. First publication. Every figure was extracted from the project directory at /data/work/floodwatch between 02:40 and 02:56 UTC on 26 September 2026, and from the public instance at floodwatch.rennwick.com, checked the same day. The provenance register for these figures is published alongside this article.
Supersedes. An earlier short note, FloodWatch Ireland: a free watch on the State's flood data, built in public, described the service at 1,221 checks and carried neither the fault log, nor the thresholds, nor the coverage limits. That page stays up as a record; this one replaces it as the reference.
Related reading. Irish public contracts 2023 to 2025 is the closest thing we have published to this study: the same method, applied to what the State buys and not to what its rivers do. Our insights index lists the rest, what we do explains the work this came out of, and you can contact us about a dataset you would like watched. If you want the same pattern applied to your own industry data, our shop lists the packaged versions.
This article is general information about public data. It is not planning, insurance, legal or engineering advice, and it is not a flood warning. For a decision about a specific property, use the OPW flood maps, your local authority and a qualified professional, and treat a Met Eireann warning or a local authority instruction as overriding anything published here.