Last Larch · Open source tool

ARIA
by Last Larch

A free, browser-based tool that helps national authorities query their own GIS data within an impact polygon, then generate sector-specific, library-grounded <description> text for Common Alerting Protocol warnings.

Early Warning Systems Common Alerting Protocol No server required Your data stays local Any AI model
LIBRARY CONTEXTUALISED SYNTHESISED

The challenge

Early warnings fail when language doesn't reach the right people

Countries implementing Early Warning Systems (EWS) under the UN EWEA initiative face a common bottleneck: converting meteorological data into actionable, sector-specific alert language — fast, consistently, and at scale.

01
Generic language doesn't move people

A flood warning that doesn't mention the school, the clinic, or the road your community depends on doesn't produce action. Specificity saves lives.

02
Drafting under pressure is error-prone

Warning officers work under extreme time pressure. Without structured support, alert language is inconsistent, incomplete, or misaligned with CAP standards.

03
Data is there — connecting it isn't easy

National authorities often have GIS data on infrastructure, schools and health facilities — but no fast way to connect that data to the impact footprint of an incoming hazard.


How it works

Three steps from polygon to publish-ready alert language

The tool runs entirely in your browser. No account, no server, no data leaving your machine unless you choose to use an AI API.

1
Query your impact area

Load your own GIS data sources — local files, ArcGIS REST services, WFS endpoints, or shapefiles. Draw a polygon on the map (or paste coordinates from a met service warning), and the tool instantly queries every layer to find what's inside: schools, health facilities, roads, population centres, critical infrastructure.

Supports: GeoJSON · Shapefile (.zip) · ArcGIS REST API · ArcGIS Web Map URLs · WFS endpoints · Paste polygon (lat/lng pairs, GeoJSON, WKT)
2
Review and configure

Select which layers to include in the alert. Set the hazard type, severity, and authority level — or let the tool infer the appropriate level from your data. Optionally load your organisation's pre-validated early action library (.db file) to ground the output in approved language.

Configure: hazard type · severity (Extreme / Severe / Moderate / Minor) · authority level (National / Regional / Local / Individual, or auto-inferred) · AI model of your choice
3
Generate sector-specific CAP descriptions

The tool queries your library for matching pre-validated messages, then passes them — alongside your GIS impact data — to the AI model of your choice. It produces sector-specific <headline> and <description> blocks ready to paste into your CAP editor, each tagged by source.

Output: copy individual sectors · download all as .txt · each block tagged Library / Contextualised / Synthesised

The early action library

Your validated language. The AI fills in the specifics.

Load a SQLite library of pre-validated early action messages — organised by hazard, sector, severity, and authority level. The AI uses these as its foundation rather than generating from scratch, then layers in the specific names, counts, and conditions from your GIS data.

Every output block is tagged so you always know what you're working with:

  • Library — your validated message, used as written, with local details slotted in
  • Contextualised — your validated message, meaningfully adapted with specific local data from the polygon
  • Synthesised — no library match existed; generated from your GIS data alone

The library file is a standard SQLite .db — you own it, you control it, and it never leaves your machine. Last Larch provides a seed library covering 8 hazard types across 8 sectors.

Example output tags
Library
"Activate school flood emergency plan; brief all staff on roles, flood evacuation procedures, and parent communication protocol."
Contextualised
"Activate flood emergency plans at all 14 schools in the Kigali polygon, including Remera Primary and GS Kimironko (boarding). Brief staff on evacuation routes and parent communication — 3,240 students are at risk."
Synthesised
"No library message matched this sector for this hazard and severity. This description was generated directly from GIS data and should be reviewed before publication."

Privacy & security

Your data never leaves your machine

For national authorities handling sensitive infrastructure data, data sovereignty is non-negotiable. ARIA is built around that principle from the ground up.

📁
Local files stay local

Shapefiles, GeoJSON files, and your .db library are loaded directly into your browser's memory. They are never uploaded to any server. Close the tab and they're gone.

🔑
You choose your AI provider

Connect any OpenAI-compatible model — Claude, GPT-4o, Gemini, or a self-hosted model via Ollama. Your API key is stored only in your browser's localStorage. Last Larch never sees it.

📄
One file. No install.

The entire tool is a single HTML file. No server, no dependencies to install, no account to create. Open it in any modern browser and it works. Share it by attaching it to an email.

⚠ Note: When you generate descriptions using an AI API, a summary of your GIS data (feature counts, attribute names, and up to 40 sample rows per layer) is sent to the AI provider you configure. No raw geometry is transmitted. Review your AI provider's data policy if this is a concern.


Who it's for

Built for practitioners, not just technologists

You don't need to be a developer to use this tool. If you can open a browser and load a file, you can produce publish-ready alert language.

🌦
National meteorological services

Translate forecast data into structured CAP alerts with sector-specific language — without a developer in the room. Compatible with any CAP editor.

🏛
National disaster management authorities

Query your own infrastructure data against any incoming hazard polygon. Generate authority-level appropriate messaging for education, health, logistics, and more.

🌍
UN agencies & regional bodies

Support country offices in building and maintaining validated early action libraries. The .db format is portable and version-controlled.

🔬
EWS researchers & implementers

Prototype, evaluate, and iterate on impact-based forecasting workflows. All outputs are traceable — you always know whether language came from the library or was generated.


Under the hood

Built on open standards

No proprietary dependencies. Every component is open source or an open standard.

Leaflet + Draw
Interactive map and polygon drawing
Turf.js
Spatial operations and polygon intersection
shpjs
In-browser shapefile parsing
sql.js
WebAssembly SQLite — reads your .db library in the browser
ArcGIS REST
Direct query of FeatureServer and web map layers
CAP v1.2
Output structured for the Common Alerting Protocol standard
OpenAI-compatible
Works with Claude, GPT-4o, Gemini, or any compatible API
SQLite
Portable, version-controlled early action library format

Seed library

ARIA-Lex — download the early action library

ARIA-Lex is the seed library that ships with ARIA. It contains pre-structured early action messages covering 8 hazard types across 8 sectors, organised by severity and authority level. It's a starting point — designed to be extended, localised, and validated by your organisation.

What's inside
  • 8 hazard types: Cyclone, Flood, Heatwave, Wildfire, Earthquake, Tsunami, Cold wave, Disease outbreak
  • 8 sectors: Government, Education, Health, Logistics, Transportation, Telecommunications, Food & Agriculture, Community
  • 4 severity levels × 4 authority levels per combination
  • Standard SQLite format — open in any DB browser or load directly into ARIA
How to use it
  1. Download ARIA-Lex.db below
  2. Open ARIA in your browser
  3. Click No library loaded in the top-right header
  4. Drop or browse to ARIA-Lex.db
  5. The library loads instantly — no server, no account
Free download
ARIA-Lex.db
SQLite · ~924 KB · Seed library v1
↓ Download ARIA-Lex Provisional seed data · validate before operational use
Free. Open source. Ready to use.

Download the single HTML file, open it in your browser, and start querying your data. No account, no server, no installation. Questions or custom implementations — Last Larch is available for consultancy engagements.