Files

T

Alby96andClaude Opus 5 b4b8d089e6 Add municipality and coordinate search, and a map target selector

Puts the controls that change what the map is looking at over the map itself,
where they act, and gives them something to search.

Search is one field, not two. A string either parses as coordinates or it does
not, and the answer is obvious from the text, so making the user declare up
front which kind of thing they are looking for would be asking them to do the
program's job. Coordinates accept decimal and degrees-minutes-seconds, comma or
space separated, with hemisphere letters — including the Italian O for ovest,
because someone reading an Italian map will type it and silently reading it as
east would put them the wrong side of Greenwich.

The 1180 Piedmont municipalities are bundled rather than geocoded online. The
app is region-scoped, so a list of one region's towns is small enough to ship
(100 KB) and beats a geocoder on every axis that matters: instant, offline, no
API key, no rate limit, and it cannot return a result somewhere the app has no
radar for. Matching folds accents, so "aglie" finds "Agliè", and prefix matches
outrank substring ones — typing "tor" should surface Torino, not the first
alphabetical name that happens to contain those letters.

tool/generate_places.py derives the list from Istat boundary shapefiles
(CC BY 4.0). Two properties of that file cost time and are now written down:
the geometry is UTM 32N rather than degrees, and the DBF is UTF-8 despite one
bilingual Friulian record that makes strict cp1252 fail. A terminal renders
utf-8 and latin-1 output identically, so the encoding cannot be settled by
looking at printed text — it took dumping codepoints. A guard in the generator
and a test against the shipped asset both check for mojibake now, and the guard
caught a real mistake the moment it was written.

The target selector switches between following the device, a saved place, and
the whole region, and the selection doubles as what the app reopens on: "the
place I marked" and "what I see when I open the app" are one idea to the person
using it. Dragging the map while it is following stops the camera chasing them
but leaves that preference alone, because looking somewhere else now is not the
same as changing their mind about next launch.

The position marker and the following are MapLibre's own, driven by
onCameraTrackingDismissed, so there is no second location stream to keep in step
with the map.

A test asserts that all 1180 municipalities fall inside the region bounds, which
is what actually validates the UTM-to-degrees conversion end to end.

Verified on the emulator: the dropdown lists follow, region and both saved
places; "aglie" finds Agliè; "44.3841 7.5426" offers the coordinate jump and
lands on Cuneo at town-reading zoom; selecting follow moves the map to the
device position with the blue dot on it and changes the recentre button to
match.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

2026-09-10 21:47:20 +02:00

16 KiB

Raw Blame History

Data sources

Every endpoint below was verified on 2026-09-10. Anything not verified is marked TO VERIFY and must not be relied on until checked.

1. Radar — Radar-DPC (the active source)

Public platform of the Italian Dipartimento della Protezione Civile.

Base URL: https://radar-api.protezionecivile.it/
Transitional test base URL: https://wagiqofvnk.execute-api.eu-south-1.amazonaws.com/prod

`GET /findLastProductByType?type=<PRODUCT>`

Returns the timestamp of the most recent sample of a product.

{
  "total": 1,
  "lastProducts": [
    { "productType": "VMI", "time": 1758706200000, "period": "PT5M" }
  ]
}

time is epoch milliseconds, UTC.
period is the ISO-8601 cadence of the product.
400 when type is missing, 404 when no product is available.

`POST /downloadProduct`

Returns a pre-signed S3 URL for the raw raster file.

{ "productType": "VMI", "productDate": 1758706200000 }

{
  "bucket": "dpc-radar",
  "key": "VMI/22-09-2025-14-20.tif",
  "url": "https://dpc-radar.s3.eu-south-1.amazonaws.com/...",
  "expiresSeconds": 900
}

productDate must be rounded down to the product step, otherwise 404.
Pre-signed URLs expire after 300–900 s: download immediately, never cache the URL.
400 when productDate is missing or non-numeric.

Mandatory header

origin: https://radar.protezionecivile.it

Required on every request to the platform for security reasons; requests without it are rejected. Content-Type: application/json is also required on downloadProduct.

Products and cadence

Cadence	Products
5 min	`VMI`, `SRI`, `SRT1`, `IR_108`
30–60 min	`TEMP`, `CUM3`, `CUM6`, `CUM12`, `CUM24`, `CAPPI_1`…`CAPPI_10`, `VIL`, `ETM`, `POH`
—	`SITES` (GeoJSON, fetched directly from S3 with `GET`)

The MVP uses VMI (Vertical Maximum Intensity) as the precipitation layer. History retention is 14 days. Files are GeoTIFF; CRS is read from the file metadata rather than assumed — the worker reprojects to EPSG:3857 in all cases.

Raster format (read from a live VMI file, 2026-09-10)

Property	Value
Size	1200 × 1400 px
Pixel size	1000 × 1000 m — a 1 km grid
Bands / type	1 band, Float32, LZW compressed
Tie point	raster (0,0) → model (-600000, 650000)
Model type	projected, datum WGS84 (GeographicTypeGeoKey 4326)
Projection centre	12.5° E, 42.0° N (`ProjCenterLong`/`ProjCenterLat`)
Linear units	metre
Nodata tag	absent — expect NaN

The rasters are not in EPSG:4326 and not in EPSG:3857. They are on a custom projection centred on Italy, so reprojection to Web Mercator is mandatory, not an optimisation. The extent works out to 1200 km east-west and 1400 km north-south around that centre, which covers the whole country — the worker crops to the region bounding box before doing anything else, so it never carries the full mosaic through the pipeline.

The GeoKeys as written are internally inconsistent (ProjCoordTransGeoKey = 1, which is transverse Mercator, combined with the centre keys an azimuthal projection uses). Do not hardcode a source CRS. The worker reads the CRS from the file with rasterio and warps from whatever it finds, and logs the detected CRS on every run so a change at the source is visible instead of silently shifting the imagery.

Values are physical units (dBZ for VMI), not palette indices — the colour mapping is ours, and the legend that goes with it is published in the manifest.

Push channel

wss://radar-wss.protezionecivile.it pushes a notification whenever a new sample is published. The worker uses this as its primary trigger, with a 5-minute cron as a fallback, so we never poll DPC aggressively.

License

CC BY-SA. Attribution to "Radar-DPC" is mandatory and derived products must be released under the same license. The rendered PNG frames we publish are a derived product and inherit CC BY-SA.

Docs: https://dpc-radar.readthedocs.io/it/latest/

2. Radar — ARPA Piemonte (future adapter, disabled)

C-band polarimetric radars at Bric della Croce and Monte Settepani, scanning every 5 minutes. Real-time volumes of reflectivity, Doppler velocity and differential reflectivity are published in OPERA HDF5 (ODIM) format, covering the last hour.

Listed as open data at https://www.arpa.piemonte.it/dato/open-data, described there as constantly updated, machine-readable and free of charge.

Two things block adoption, and both belong to the project owner, not to the code:

The access link is not public. The page states that to obtain the real-time link for radar and radiosonde data "è necessario inviare una e-mail all'indirizzo info.meteo@arpa.piemonte.it". There is no documented endpoint to call without it.
The licence is stated site-wide, not on that page. The open-data page itself says only that the data is gratuiti. ARPA's legal notice at https://www.arpa.piemonte.it/note-legali does state a licence for environmental and open data: CC BY 4.0, commercial use permitted, crediting "Fonte: Arpa Piemonte - www.arpa.piemonte.it". Logos, emblems and third-party content are excluded from it.

Both ARPA REST APIs link that same notice from their OpenAPI description, so for them the licence is unambiguous. The radar page does not repeat it, so the safe move is to ask ARPA to confirm the terms apply to the radar volumes in the same email that requests the access link.

Until the link exists, ArpaRadarSource stays a stub whose every method throws. It is named by the region config as unavailable so the seam is visible, not because it is nearly ready.

3. ARPA Piemonte REST APIs — surveyed 2026-09-10, not yet adopted

Two public APIs, no authentication, no key, no registration. Both link https://www.arpa.piemonte.it/note-legali from their OpenAPI description, so both are CC BY 4.0 with commercial use permitted.

`api_realtime` — https://utility.arpa.piemonte.it/api_realtime

Spec at /api_realtime/openapi.json (the /docs page is JS-rendered and cannot be scraped; fetch the spec instead).

Endpoint	Content
`GET /pie_anag`	Station registry: 374 stations, `lat`/`lng`, elevation, municipality, province, basin, sensor letters
`GET /data_pie`	Hourly observations, last 3 days: `air_temperature`, `humidity`, `cum_rain_1h/3h/6h/12h/24h`, `wind`, `gust_of_wind` + directions, `snow_height`, `hydrometric_level`. Paginated
`GET /pie_neve`	Daily validated snow height, 08:00
`GET /ggd`	Heating and cooling degree days
`GET /status`	Health check

Sensor coverage across the 374 stations: 286 with a rain gauge (P), 277 temperature, 143 hydrometer (H), 90 wind, 79 snow. Rain gauges span 74 m to 2820 m elevation.

Measured latency: ~4.5 hours, uniform across every station sampled at 16:30 on 2026-09-10, on an hourly cadence. This is the single most important property of the feed: it is an observation archive with a publishing delay, not a real-time source, and it cannot be presented alongside 5-minute radar as if it were current.

Hydrometric thresholds are effectively absent: only 4 of 143 hydrometer stations carry guard_threshold / danger_threshold, and only 5 carry a river name. A river-level warning feature built on this would be blank for 97% of stations.

Meteoweb — https://utility.arpa.piemonte.it/docs/

Spec at /schema/ (YAML). Django REST Framework, hyperlinked, paginated.

Historical archive under /meteoidro/: daily and monthly meteo (3.27 M daily records), hydrological (931 k), snow, century-long series, annual rainfall maxima, flow-duration curves and rating curves, plus the station and measurement-point registries. Latest daily data was one day behind (2026-09-09).

Also /sismica/: 35 910 seismic events since 2000 with position, depth and magnitude, and the seismic classification of municipalities.

Verdict

Only one thing here is on-scope: rain-gauge accumulations as ground truth for the radar. Radar infers rainfall from reflectivity aloft; gauges measure what actually reached the ground, and the network is dense exactly where Alpine terrain blocks the radar beam. Worth adding as a clearly-timestamped observations layer once the worker exists — never as a live reading.

Everything else is out of scope for this app: snow depth, degree days, the historical archive and the seismic catalogue all belong to a different product. Hydrometric levels would be valuable during heavy rain but the thresholds that would make them meaningful are not published.

4. Forecast — out of scope

The app shows radar, not forecasts. No forecast provider is integrated, and MET Norway, ItaliaMeteo ICON-2I and Open-Meteo are all out of scope. See CLAUDE.md for the current scope boundary.

5. Alerts — ARPA Piemonte XML-CAP

Feed: https://www.arpa.piemonte.it/export/xmlcap/allerta.xml
Landing page: https://www.arpa.piemonte.it/rischi_naturali/snippets_arpa/allerta/index.html
Official PDF bulletin: https://www.arpa.piemonte.it/rischi_naturali/boll/bollettino_allerta.pdf

Issued daily by 13:00 by the Centro Funzionale Regionale, 36 hours of validity.

Alert zones (11)

Piem-A … Piem-M (A, B, C, D, E, F, G, H, I, L, M), covering: Toce; Val Sesia/Cervo/ Chiusella; Valli Orco/Lanzo/bassa val Susa/Sangone; Alta val Susa/Chisone/Pellice/Po; Valli Varaita/Maira/Stura; Valle Tanaro; Belbo e Bormida; Scrivia; Pianura settentrionale; Pianura e colline torinesi; Pianura cuneese. The exact code-to-name mapping is stored in app/assets/regions/piemonte.json.

CAP document structure (verified against the live feed, 2026-09-10)

One <alert> contains one <info> block per zone; the zone is in info/area/areaDesc and holds the bare code (Piem-A), not a name.

<alert xmlns="urn:oasis:names:tc:emergency:cap:1.2">
  <identifier>261/2026</identifier>          <!-- bulletin number / year -->
  <sender>centro.funzionale@arpa.piemonte.it</sender>
  <sent>2026-09-09T10:00:02+00:00</sent>
  <status>Actual</status><msgType>Alert</msgType><scope>Public</scope>
  <note>AVVISO DI CONDIZIONI METEOROLOGICHE AVVERSE ...</note>
  <info>
    <event>GIALLO</event>                    <!-- overall level for this zone -->
    <onset>...</onset><expires>...</expires>
    <parameter><valueName>EFFETTI SUL TERRITORIO</valueName><value>...</value></parameter>
    <parameter><valueName>TEMPORALI_1224</valueName><value>GIALLO</value></parameter>
    <parameter><valueName>TEMPORALI_2436</valueName><value>VERDE</value></parameter>
    ...
    <area><areaDesc>Piem-A</areaDesc></area>
  </info>
  <!-- 10 more info blocks, one per zone -->
</alert>

Parameters are named <CATEGORY>_<WINDOW>:

Categories — IDRAULICO, IDROGEOLOGICO, TEMPORALI, NEVE, VALANGHE.
Windows — 1224 (hours 12–24) and 2436 (hours 24–36), matching the 36 hours of validity.

So each zone carries 10 level values plus one overall <event> level and a free-text EFFETTI SUL TERRITORIO.

Levels — six values, not four

Value	Meaning
`VERDE`	no significant phenomena
`GIALLO`	localised phenomena
`ARANCIONE`	widespread phenomena
`ROSSO`	numerous or extensive phenomena
`BIANCO`	avalanche scale only — outside the season / not assessed
`-`	no value published for that category and window

BIANCO and - are not "green with a different name" and must not be collapsed into it. The parser models the level as an enum with explicit bianco and notAvailable members, and the UI renders them as their own state rather than as "no alert".

Levels are republished verbatim, never reinterpreted, always alongside a link to the official bulletin. No license is stated on the ARPA page, so we credit ARPA Piemonte and link the source rather than claiming any reuse right.

Official bulletin PDF, linked from every alert view: https://www.arpa.piemonte.it/rischi_naturali/boll/bollettino_allerta.pdf

Zone codes

Piem-A … Piem-M, eleven zones: the Italian alphabet is used, so J and K do not exist — the sequence is A B C D E F G H I L M. The code-to-name mapping lives in app/assets/regions/piemonte.json; the human-readable names are not in the feed and come from the ARPA zone documentation.

A captured copy of the live feed is kept as a test fixture so the parser is verified against the real document rather than a hand-written approximation.

6. Municipalities — Istat boundaries

The 1180 municipalities of Piedmont with a representative point each, bundled as app/assets/regions/piemonte_comuni.json (98 KB) and generated by tool/generate_places.py.

Source: Istat, Confini delle unita amministrative a fini statistici, generalised boundaries at 1 January 2025, CC BY 4.0. https://www.istat.it/storage/cartografia/confini_amministrativi/generalizzati/2025/Limiti01012025_g.zip

Bundled rather than geocoded online because the app is region-scoped: the search is instant, works with no network, needs no API key, has no rate limit to respect, and cannot return a result somewhere the app has no radar for.

Two properties of the source file are worth knowing before touching the generator:

The shapefile is in WGS84 / UTM zone 32N, not in degrees. The generator runs an inverse transverse Mercator to get latitude and longitude.
The DBF is mixed-encoding. It is UTF-8 — "Aglie" ends in a single U+00E8 — but at least one bilingual Friulian record makes strict cp1252 fail, and a terminal renders utf-8 and latin-1 output identically, so this cannot be settled by looking at printed text. The generator asserts that no output name contains the tell-tale mojibake sequences, and a test asserts it again against the shipped asset.

Points are the centroid of each municipality's largest part, not of the whole shape: several are multipart, and averaging a main body with its exclaves can land the point on a neighbour. Expect 1-4 km between the point and the town centre, which is well inside what framing a map needs.

7. Base map — OpenFreeMap

OpenStreetMap vector tiles served by OpenFreeMap (https://openfreemap.org).

Style in use: https://tiles.openfreemap.org/styles/positron
Tiles: https://tiles.openfreemap.org/planet
Glyphs: https://tiles.openfreemap.org/fonts/{fontstack}/{range}.pbf

No API key, no registration, no request limits, commercial use permitted. That is what makes it the right starting point: every other free tier — MapTiler, Stadia, Jawg, Thunderforest — requires a key, which means a secret to manage and a quota to outgrow.

Positron rather than Liberty or Bright: a radar overlay has to be the loudest thing on screen, and Positron is a desaturated grey base designed to sit under data. On Liberty the precipitation colours compete with road casings and landuse fills.

Mandatory credits

Credit	Required?
`© OpenStreetMap contributors` (ODbL)	yes
`© OpenMapTiles`	yes
`OpenFreeMap`	optional, and appreciated

The style JSON carries no attribution field, so MapLibre will not display these on its own. The app renders them itself from the region config: the two mandatory credits go in the always-visible attribution bar, and all three are listed on the Sources screen.

When there is no network

With MAP_STYLE_URL=offline the app generates a self-contained style from the region bounding box — flat background plus the extent outline, no network sources at all. It is deliberately plain so it is never mistaken for a finished map, and it is what the widget tests run against.

8. Lightning — out of scope

Not part of the app. If it is ever revisited: Blitzortung prohibits commercial use, and the DPC LTG product is the source to reach for instead.

16 KiB Raw Blame History Unescape Escape