HarvestGuard 🌾

AI-Powered Food Insecurity Early Warning System

HarvestGuard is an open-source platform that fuses real satellite vegetation data, rainfall anomalies, and WFP food security indicators to generate AI-powered 30-60 day early warnings of crop failure and food insecurity — accessible to NGOs, governments, and humanitarian responders worldwide.

The Problem

828 million people face food insecurity. Smallholder farmers (who produce 70% of the world's food) have zero access to advance crop failure warnings. WFP's HungerMap shows current status; FEWS NET publishes monthly expert reports. No open, accessible platform provides real-time predictive intelligence at country level.

HarvestGuard fills this gap.

How It Works

NASA MODIS NDVI  ──┐
CHIRPS Rainfall  ──┤──► Risk Engine ──► Claude AI ──► Risk Assessment + Alerts
WFP HungerMap    ──┘                              ──► Conversational Analysis
Open-Meteo       ──┘

1. Real satellite data — NASA MODIS vegetation indices updated every 16 days
2. Rainfall anomalies — CHIRPS precipitation vs 1981-2010 historical baseline
3. Ground truth — WFP HungerMap IPC phases and affected population counts
4. AI analysis — Claude generates plain-language risk assessments from real data
5. Early warnings — Alerts fire when composite risk scores deteriorate significantly

Data Sources (100% Real, No Synthetic Data)

Source	Data	Update Frequency
NASA MODIS MOD13A2	NDVI/EVI vegetation indices	16 days
CHIRPS	Precipitation anomalies (1981-present)	Monthly
WFP HungerMap	IPC phases, affected populations	6 hours
Open-Meteo	Historical & forecast weather	Daily

Features

• Interactive World Map — 60+ countries color-coded by IPC risk phase
• Real-time NDVI & Rainfall — Satellite vegetation health and precipitation anomaly trends
• AI Risk Assessments — Claude-powered streaming analysis with real data context
• Early Warning Alerts — Automatic detection of deteriorating food security
• Trend Charts — 90-day NDVI + rainfall history per country
• AI Chat — Ask "What's happening in Ethiopia?" and get data-grounded answers

Quick Start

Prerequisites

• Docker & Docker Compose
• Anthropic API key (console.anthropic.com)

1. Clone and configure

git clone https://github.com/rishavsunny12/claude-code-tp
cd claude-code-tp
cp .env.example .env
# Edit .env — set ANTHROPIC_API_KEY

2. Launch

docker compose up --build

3. Seed real data

# Trigger the first data refresh (fetches WFP + weather data)
curl -X POST http://localhost:8000/api/v1/admin/refresh

4. Open the app

Navigate to http://localhost:3000

Click any country on the map to see:

• Current IPC phase and affected population
• NDVI and rainfall anomalies vs historical baseline
• 90-day trend chart
• Live-streaming Claude AI risk assessment

API Reference

Endpoint	Description
GET /api/v1/regions	All monitored countries with current risk scores
GET /api/v1/regions/{iso}	Full country detail with 90-day history
GET /api/v1/forecast/{iso}	Stream AI risk assessment (SSE)
GET /api/v1/alerts	Active food security alerts
POST /api/v1/chat	Conversational AI query (SSE)
GET /health	Health check

Architecture

harvestguard/
├── backend/          Python FastAPI + SQLAlchemy + APScheduler
│   └── app/
│       ├── services/data/    NASA NDVI, CHIRPS, WFP, Open-Meteo integrations
│       ├── services/analysis/ Risk scoring + alert engine
│       ├── services/ai/       Claude API streaming
│       └── api/v1/           REST endpoints
├── frontend/         React + TypeScript + MapLibre GL + Tailwind
│   └── src/
│       ├── components/map/    Interactive choropleth world map
│       ├── components/sidebar/ Country detail panel
│       └── components/chat/   AI conversation interface
└── docker-compose.yml

Optional: NASA Earthdata (Enhanced NDVI)

For higher-accuracy NDVI from actual MODIS raster data:

1. Register free at urs.earthaccess.nasa.gov
2. Add EARTHDATA_USERNAME and EARTHDATA_PASSWORD to .env

Without credentials, NDVI is estimated from Open-Meteo weather correlates (still useful as a proxy).

Application live screenshots

Contributing

This project is designed to serve humanitarian organizations, governments, and researchers. Contributions welcome — especially:

• Additional data source integrations (ACLED conflict data, FEWS NET)
• Sub-national resolution (admin level 1/2)
• SMS/WhatsApp alert delivery for rural communities
• Multi-language support

License

MIT — free for humanitarian and research use.

Data Attribution

• Vegetation data: NASA/USGS MODIS Land Processes DAAC
• Precipitation: CHIRPS, Climate Hazards Center, UC Santa Barbara
• Food security: World Food Programme HungerMap LIVE
• Weather: Open-Meteo (CC BY 4.0)