Hurakai

A native macOS tracker for hurricanes and tropical disturbances in the Eastern and Central Pacific — live positions, forecast tracks, uncertainty cones, watches and warnings, and full ATCF model guidance on a clickable MapKit chart, including Google DeepMind's cyclone ensemble.

---

Contents

• What it is
• Basin scope
• Screens
• Data sources
• Model guidance comes from the a-deck
• Third-party sources
• Appearance
• Requirements
• Install
• Build from source
• Packaging
• Code signing status
• Repository layout
• Not a forecast product

---

What it is

Hurakai is a single-window macOS application that consolidates the operational picture for a Pacific tropical cyclone: where the storm is, where it is forecast to go, how confident that forecast is, what every guidance model says about its track and intensity, and what the forecaster wrote in the discussion.

Everything is fetched live from NOAA and NWS endpoints. There are no third-party libraries, no package manager, and no build step beyond swiftc — the app links only against system frameworks (SwiftUI, MapKit, Swift Charts, WebKit, Compression).

Basin scope

Eastern and Central Pacific. Central Pacific is the CPHC domain — 140°W to the dateline, the Hawai'i basin — and the default map view spans the Mexican coast west past Hawai'i to the dateline.

Both Central Pacific (CPHC) and Eastern Pacific (NHC) systems arrive through the same feeds; the GIS service and CurrentStorms.json cover the EP, CP, and AL basins.

Basin strings from the outlook layers are normalised (EP/CP/AL plus spelled-out and *PAC variants). An outlook area whose basin cannot be identified is dropped rather than guessed at, so the failure mode is a missing Pacific area rather than an Atlantic one appearing in a Pacific-scoped application. The Atlantic can be enabled explicitly from the Layers menu; it is off by default.

Screens

• Map — every active system on one chart. Selecting a storm zooms to its cone and opens an inspector with current intensity, motion, the forecast point table, the full public advisory, and the forecast discussion. Layers — cone, forecast track, past track, forecast points, watches and warnings, disturbance areas, model guidance, live GOES-West satellite imagery, labels — toggle individually. The base map switches between hybrid, satellite, and standard.
• Satellite overlay — GOES-West GeoColor draped over the map as Web Mercator tiles from NASA GIBS, refreshed with the satellite (roughly every 10 minutes), so the cloud field and the forecast geometry are readable together. Toggled from the Layers menu; off by default.
• Models (inspector tab) — full ATCF model guidance for the selected storm, drawn as a spaghetti plot with labelled endpoints. Each technique toggles individually; featured models are enabled by default and "Every technique" reveals the ensemble members. Includes GDMN, the Google DeepMind ensemble mean.
• Intensity Models — the intensity counterpart to the track spaghetti: each technique's wind forecast against forecast hour, with Saffir-Simpson thresholds marked and the official forecast drawn heavier. One tab per active Pacific system. Shares model toggles with the map, so disabling a technique in one view disables it in both. Includes the intensity-only aids (SHIP, LGEM, DSHP, IVCN) that have no track to plot.
• Satellite & Outlook — NHC/CPHC 2-day and 7-day graphical outlooks, and GOES-West GeoColor and Air Mass imagery (full disk and Hawai'i sector), scaled to the window.
• DeepMind Weather Lab — Google's experimental AI cyclone predictions.
• Outlook Text — the Eastern and Central Pacific Tropical Weather Outlook discussions.

Data refreshes on launch, on ⌘R, and every 10 minutes.

Data sources

Source	What it provides	Transport
NHC CurrentStorms.json	Active systems: position, intensity, pressure, motion, advisory links	JSON
NOAA Tropical GIS MapServer	Forecast points, forecast track, uncertainty cone, past track, watches and warnings, TWO disturbance areas	ArcGIS REST → GeoJSON
NHC ATCF a-decks	Every technique's track and intensity guidance, including Google DeepMind (GDMN)	Gzipped flat file
NHC / CPHC text products	Public advisories, forecast discussions, Tropical Weather Outlooks	HTML, <pre> extracted
NHC Graphical TWO	2-day and 7-day outlook graphics	PNG
NOAA NESDIS/STAR	GOES-18 (GOES-West) GeoColor and Air Mass imagery	JPEG
NASA GIBS	GOES-West GeoColor as Web Mercator tiles, for the map overlay	WMTS
Google DeepMind Weather Lab	Experimental AI cyclone predictions	Embedded web view

Model guidance comes from the a-deck

Model track and intensity guidance is read from the NHC ATCF a-deck (aid_public/a{basin}{nn}{year}.dat.gz), which NHC publishes openly. This is the same dataset behind the model plots on third-party cyclone sites, so the underlying numbers are available natively rather than by scraping rendered images.

GDMN is the Google DeepMind ensemble mean, contributed to NHC's guidance suite. It arrives through the a-deck with no API key, no Google Cloud project, and no sign-in. GDMI (interpolated) and GDM2 appear alongside it.

Three conventions in the format are handled explicitly:

• Per-technique cycles. Each technique keeps its own most recent initialisation cycle rather than a single global "latest". Models run at 00/06/12/18Z while the official forecast runs at 03/09/15/21Z, so a global cutoff would silently drop OFCL.
• Repeated rows. Rows repeat per wind-radius threshold (34/50/64 kt), so positions are de-duplicated by forecast hour.
• Position-less aids. Intensity-only aids — SHIP, LGEM, DSHP, IVCN — publish 0N/0W rather than omitting the position. Taken literally that draws a track from the storm to 0°N 0°E, so those points are treated as carrying intensity without position: absent from the map, present in the intensity chart.

Third-party sources

Google DeepMind Weather Lab is presented as a web view. It has no public API and requires a Google sign-in, so the application renders the site for the user to authenticate with directly and never handles credentials itself.

WeatherNext via BigQuery / Earth Engine is not implemented. Google's WeatherNext datasets are reachable through BigQuery and Earth Engine, but for cyclone track and intensity guidance they are a longer route to GDMN, which the a-deck already provides. That path would add value only for the gridded fields — global forecast grids of wind, pressure, temperature and precipitation, or per-member ensemble output. It also requires a GCP project with billing, the datasets subscribed through Analytics Hub, and the gcloud/bq tooling. The natural shape, should it be added, is a helper that queries BigQuery with existing gcloud credentials and hands the application JSON, keeping OAuth out of the app entirely.

Appearance

Light by default, with a moon/sun button in the toolbar to switch to dark. The choice persists between launches.

The Saffir-Simpson palette is tuned for dark satellite imagery — pale yellow Cat 1 and a white official-forecast line disappear against white — so application chrome (sidebar, tabs, inspector, intensity chart) uses light-mode counterparts derived from the same hues, while map overlays keep the bright originals because the map always draws on imagery. A single transform derives the light variant rather than maintaining two hand-tuned palettes; achromatic colours resolve to near-black rather than grey, which is what keeps OFCL legible.

Requirements

macOS 14 or later, and a network connection — every forecast, model and image is fetched live. Building additionally requires the Xcode Command Line Tools; full Xcode is not needed.

Install

Download the disk image from Releases and drag Hurakai to Applications. See Code signing status for the first-launch step.

Build from source

./build.sh && open build/Hurakai.app

Self-checks over the parsing, geometry, and scoping logic:

./test.sh

The test target compiles Sources/Hurakai/Data.swift against Tests/main.swift and runs assert-based checks over GeoJSON decoding, a-deck parsing, gzip handling, region math, basin scoping, and the advisory text extractor. No test framework is involved.

There is no Package.swift. build.sh invokes swiftc directly and assembles the bundle: MapKit and WebKit require a real .app (bundle identifier plus Info.plist) regardless, so a SwiftPM manifest adds nothing, and the PackageDescription library shipped with the Command Line Tools fails to link at any tools-version.

Packaging

./package.sh

Builds a universal (arm64 + x86_64) application and wraps it in dist/Hurakai-<version>.dmg — drag to Applications, roughly 1.3 MB. hdiutil ships with macOS, so there is no packaging dependency to install. A .dmg rather than a .pkg: a .pkg installer earns its keep when install scripts, receipts, or files outside /Applications are needed, and none apply here.

./build.sh --universal performs the two-architecture build on its own; plain ./build.sh stays single-architecture so development builds remain fast.

Code signing status

The bundle is ad-hoc signed with no Apple Developer ID and is not notarized, so macOS blocks the first launch on any Mac that did not build it, reporting that it "could not verify that this app is free of malware".

On macOS 15 and later, Control-clicking the app and choosing Open no longer bypasses this — Apple removed that path. Use one of:

• System Settings. Try to open the app once and let it be blocked, then go to System Settings → Privacy & Security, scroll to the Security section, and click Open Anyway next to the message about Hurakai.

• Terminal. Clear the quarantine flag the browser attached on download:

  xattr -dr com.apple.quarantine /Applications/Hurakai.app

Either way it is a one-time step; the app launches normally afterwards.

On macOS 14, Control-click → Open → Open still works.

Building from source sidesteps all of this: locally compiled bundles are never quarantined, so ./build.sh && open build/Hurakai.app just runs. Removing the friction for downloaded copies requires a paid Apple Developer account to sign with a Developer ID and notarize the disk image.

Repository layout

Sources/Hurakai/Data.swift     models, GeoJSON and ATCF decoding, feeds, networking
Sources/Hurakai/App.swift      app entry, refresh loop, shell, layer menu
Sources/Hurakai/MapPane.swift  map, overlays, pins, Saffir-Simpson palette
Sources/Hurakai/Panels.swift   sidebar, inspector, intensity chart, imagery, web and text panes
Tests/main.swift               assert-based self-checks
build.sh                       compiles and assembles Hurakai.app
package.sh                     universal build wrapped in a drag-to-install disk image
test.sh                        runs the self-checks
docs/                          README assets

The map is an MKMapView behind an NSViewRepresentable rather than SwiftUI's Map. SwiftUI's Map has no tile-overlay API at any deployment target, and a georeferenced satellite layer has to be tiles. The SwiftUI pin views are reused unchanged by hosting them inside the annotation views, so dropping down costs nothing visually.

Not a forecast product

Hurakai displays official NHC and CPHC data but is not itself an official source. For warnings and protective decisions, use hurricanes.gov and the local NWS office.