IoT-Enabled Real-Time Air Quality Intelligence Platform Ward-Level Monitoring | ML Source Detection | Wind-Aware Attribution | Smart Alerts | Policy Recommendations
Aeronyx is a full-stack air quality monitoring platform covering all 250 municipal wards across 12 MCD zones of Delhi. It combines IoT sensor hardware (ESP32), cloud data processing (Next.js API routes), ML models, wind-aware atmospheric context, and an interactive React dashboard to provide real-time pollution intelligence.
Key Metrics: 250 wards | 12 zones | 7 configurable alert rules | 72-hour AQI forecast | Wind + plume + source attribution APIs
Imagine a city where every breath is monitored, every pollutant traced to its source, and every citizen empowered with actionable insights. Aeronyx turns this vision into reality by deploying a network of low‑cost IoT nodes across Delhi’s wards, streaming data to a serverless Next.js backend that runs machine‑learning models in real time. The platform not only shows live AQI numbers but also tells you why the air is polluted—whether it’s traffic, construction, or biomass burning—and forecasts how it will change over the next three days. With wind‑aware attribution, city officials can see exactly where pollution is coming from and target interventions where they matter most. Residents receive personalized health alerts, while planners get policy recommendations backed by data. In short, Aeronyx doesn’t just measure air quality—it helps improve it.
ESP32 + Sensors (PM2.5, CO, NO2, TVOC, DHT22) | every 30s via WiFi v ThingSpeak Cloud (Channel 2697383) | REST API polling (30s) v Next.js Backend (API Routes)
- ThingSpeak Data Fetcher (live + demo mode)
- ML Pipeline (RF source, XGBoost forecast, IF anomaly)
- Wind Service (station cache + interpolation)
- Atmospheric Layer (trajectory + flow chain)
- Bayesian-Style Source Attribution
- Alert Engine (7 rules, 5-min debounce)
- SQLite Storage (async via Prisma or similar)
- WebSocket Broadcast (via Socket.io or similar) | REST + WebSocket v Next.js Frontend (React)
- 9 Interactive Pages
- Leaflet Ward Map (250 wards)
- Wind Analysis Page
- Recharts Visualization
- Client-side ML Fallback
| Feature | Description |
|---|---|
| Real-Time Dashboard | Live AQI gauge, 6-metric grid, time-series chart, health advisory |
| 250-Ward Map | Interactive Leaflet map with zone/ward toggle, color-coded AQI, search |
| ML Source Detection | Random Forest classifier identifying vehicle, industrial, construction, biomass, mixed sources |
| AQI Forecasting | XGBoost model predicting AQI 1-72 hours ahead with diurnal patterns |
| Anomaly Detection | Isolation Forest flagging unusual sensor patterns |
| Wind Intelligence | Wind station cache, interpolated wind field, seasonal pattern fallback |
| Plume + Trajectory APIs | Backward trajectory, upwind lookup, and zone flow chain endpoints |
| Source Attribution (Bayesian-style) | Probabilistic source mix using pollutant fingerprints + temporal + wind context |
| Smart Alerts | 7 configurable rules with severity levels and 5-min debounce |
| Health Advisory | Population-specific guidance for general public and vulnerable groups |
| Admin Policy Panel | Source-specific intervention recommendations for municipal officials |
| Layer | Technology | Version |
|---|---|---|
| Backend | Next.js API Routes | 15.x |
| Database | SQLite + Prisma (async) | Latest |
| ML | scikit-learn (RF, IF) + XGBoost | >=1.3, >=2.0 |
| Frontend | React + Next.js | 19.2.0 / 15.x |
| Maps | Leaflet + react-leaflet | 1.9.4 / 5.0.0 |
| Charts | Recharts | 3.8.0 |
| Animations | Framer Motion | 12.35.1 |
| IoT | ESP32 + ThingSpeak | Channel 2697383 |
| Deployment | Vercel | - |
Aeronyx/
app/
admin/ # Admin page
advisory/ # Health advisory page
alerts/ # Alerts page
analytics/ # Analytics page
api/ # Next.js API routes (backend)
live.js # /api/live, /api/advisory
history.js # /api/history
wards.js # /api/wards (250 wards + 12 zones)
ml.js # /api/ml/* (source, forecast, anomaly)
alerts.js # /api/alerts/* (rules, stats)
policy.js # /api/policy
wind.js # /api/wind/*
plume.js # /api/plume/*
attribution.js # /api/attribution/*
components/ # Reusable React components
hooks/ # Custom React hooks (e.g., useData)
map/ # Map-related pages/components
ml/ # ML utilities and model loading
plume/ # Plume map page
services/ # Service classes (if any)
wind/ # Wind analysis page
AppShell.tsx # App layout wrapper
globals.css # Global styles
layout.tsx # Root layout
page.tsx # Home page
lib/
alerts/ # Alert logic
attribution/ # Source attribution
db/ # Database connection and models
ml/ # ML utilities and model loading
thingspeak/ # ThingSpeak API wrapper
utils/ # Utility functions
weather/ # Weather API integration
public/
favicon.ico # Favicon
aeronyx.png # Logo
wards.json # GeoJSON (12 zones + 246 wards) for frontend
data/
alert_history.json # Historical alert data
alert_rules.json # Alert rule configurations
industrial_sources.csv # Industrial source data for ML
sensor_readings.json # Sensor data storage
wind_history.json # Historical wind data
AGENTS.md # Agent instructions for this project
CLAUDE.md # Project-specific instructions
DESIGN.md # Design specifications
README.md # This file
eslint.config.mjs # ESLint configuration
next-env.d.ts # Next.js TypeScript types
next.config.ts # Next.js configuration
package-lock.json # Locked dependencies
package.json # Dependencies and scripts
postcss.config.mjs # PostCSS configuration
skills-lock.json # Locked skills
tsconfig.json # TypeScript configuration
| Component | Model | Measurement |
|---|---|---|
| Microcontroller | ESP32-WROOM-32 | WiFi/BLE, Dual-core 240MHz |
| PM2.5 Sensor | WINSEN ZPH02 (laser) | 0-1000 ug/m3 |
| CO Sensor | MQ-7 | 20-2000 ppm |
| NO2 Sensor | DFRobot MEMS | 0-5 ppm |
| TVOC Sensor | WINSEN ZP07-MP503 | 0-50 ppm |
| Temp/Humidity | DHT22 | -40 to 80C, 0-100% |
| Display | SSD1306 OLED 0.96" | 128x64 px |
| Power | 3.7V LiPo + MT3608 boost | ~8h runtime |
Estimated cost per node: INR 5,200
- Algorithm: Random Forest (150 trees, max_depth=12)
- Classes: vehicle, industrial, construction, biomass, mixed
- Features: pm25, co, no2, tvoc, temperature, humidity, hour, pm25/co ratio, tvoc/no2 ratio
- Fallback: Rule-based detection (client-side + server-side) when model unavailable
- Algorithm: XGBoost Regressor (200 rounds, lr=0.08)
- Output: Hourly AQI predictions for 1-72 hours
- Features: hour, day_of_week, pollutants, temperature, humidity, lag features (1h/3h/6h)
- Algorithm: Isolation Forest (150 trees, contamination=0.05)
- Output: is_anomaly flag + anomaly_score (0-1)
The plume model predicts the dispersion of pollutants from sources using wind data and atmospheric stability. It provides backward trajectory analysis to identify upwind sources contributing to a ward's pollution, and forward plume simulation to estimate impact areas. The model uses Gaussian plume approximation adjusted for urban canopy layer, integrating real-time wind fields from the Wind Service. Outputs include: plume concentration maps, source contribution percentages, and travel time estimates.
All API endpoints are under /api in the Next.js app.
| Method | Endpoint | Description |
|---|---|---|
| GET | /api/live |
Latest sensor reading |
| GET | /api/advisory |
Health advisory for current AQI |
| GET | /api/history?hours=24 |
Historical readings |
| GET | /api/wards |
All 258 zone+ward readings |
| GET | /api/wards/{ward_id} |
Single ward reading |
| GET | /api/ml/source |
ML source classification |
| GET | /api/ml/forecast?horizon=24 |
AQI forecast |
| GET | /api/ml/anomaly |
Anomaly detection |
| GET | /api/ml/summary |
Combined ML analysis |
| GET | /api/alerts |
Alert history |
| GET | /api/alerts/rules |
Alert rules |
| GET | /api/policy?source=vehicle |
Policy recommendations |
| GET | /api/wind/current |
Station wind snapshot |
| GET | /api/wind/field?grid_size=12 |
Interpolated wind vector field |
| GET | /api/wind/history?hours=24 |
Wind history snapshots |
| GET | /api/wind/at?lat=...&lon=... |
Wind at a specific point |
| GET | /api/wind/upwind/{ward_id} |
Upwind wards for target ward |
| GET | /api/wind/seasonal |
Seasonal wind profile metadata |
| GET | /api/plume/trajectory/{ward_id}?hours=3 |
Backward trajectory from ward |
| GET | /api/plume/upwind/{ward_id} |
Upwind wards via plume context |
| GET | /api/plume/flow-chain/{zone_id} |
Wind flow order within zone |
| GET | /api/attribution/ward/{ward_id} |
Full ward source attribution |
| GET | /api/attribution/zone/{zone_id} |
Zone-level source attribution |
| GET | /api/attribution/city |
City-wide source contribution summary |
| GET | /api/health |
Backend health check |
| WS | /ws/live |
Real-time WebSocket feed |
Aeronyx unifies its air quality reporting under the Indian CPCB (Central Pollution Control Board) standard to maintain local regulatory consistency for Delhi municipal wards.
While global web platforms (like WAQI API) and commercial apps (like aqi.in, Dyson Link, or Apple Weather) default to the US EPA (Environmental Protection Agency) standard by default, Aeronyx calculates CPCB AQI dynamically from raw sensor PM2.5, CO, and NO2 concentrations.
Both standards use the same piecewise linear interpolation formula:
- Indian CPCB: Good category covers up to
30 µg/m³and Satisfactory up to60 µg/m³. An instantaneous PM2.5 of87 µg/m³translates to190 AQI(CPCB Moderate). - US EPA: Stricter at lower concentrations. Good category caps at
12 µg/m³and Moderate at35.4 µg/m³. An instantaneous PM2.5 of87 µg/m³translates to167 AQI(US EPA Unhealthy).
All telemetry gauges, sidebar indicators, and ward heatmaps in Aeronyx are fully synchronized using the Indian CPCB standard.
- Clone the repository
- Install dependencies:
npm install
- Create a
.env.localfile in the root with:THINGSPEAK_CHANNEL_ID=2697383 THINGSPEAK_READ_API_KEY=your_thingspeak_read_key # Optional: OWM_API_KEY for real wind data - Run the development server:
npm run dev
- Open http://localhost:3000 in your browser.
The easiest way to deploy your Next.js app is to use the Vercel Platform from the creators of Next.js.
- Push the repository to GitHub
- Import the project on Vercel
- Set the environment variables in Vercel dashboard:
THINGSPEAK_CHANNEL_IDTHINGSPEAK_READ_API_KEYOWM_API_KEY(optional)
- Vercel will automatically build and deploy the application.
Note: Replace the placeholder screenshot URLs with actual screenshots of your deployed application.
Built with ESP32, Next.js, React, scikit-learn, XGBoost, Leaflet, and Recharts