StreamDash

Real-time SuDS (Sustainable Drainage Systems) monitoring dashboard for the University of Hull campus, funded by the UPP Foundation.

Live site: streamdash.org

---

Overview

StreamDash is a student-driven living lab that collects and visualises environmental sensor data from 22 WeatherLink nodes deployed across the UoH campus. It integrates live sensor networks, interactive maps, geospatial analytics, and machine learning to support sustainability research and education.

---

Features

Page	Description
Home	Hero, live stats, feature cards with network health pills, live weather conditions, campus map
Dashboard	Filter by location → site → sensor, date range picker, time-series chart, sensor hardware tooltips
Network	Health status of all 22 campus nodes — last data timestamp, installation date, sensor categories
Digital Twin	ML surrogate model — adjust rainfall and temperature scenarios and predict soil moisture response at 6 depths
Map	Interactive Leaflet map of campus nodes with category filters, linked directly to dashboard
Parameters	Catalogue of monitored environmental parameters
Gallery	Project team

---

Tech Stack

Frontend

• React 18 (Create React App)
• React Router v6
• Recharts (time-series charts, radar chart, bar chart)
• react-leaflet v4 (interactive map)

Backend

• FastAPI + Uvicorn/Gunicorn
• MariaDB (mariadb Python connector)
• scikit-learn + joblib (ML surrogate models)
• requests (WeatherLink live conditions proxy)

Infrastructure

• System nginx with Certbot SSL (Let's Encrypt)
• Systemd service (fastapi.service) for the backend — uses conda env at miniforge3/envs/suds/
• Static frontend served from /var/www/streamdash/

---

Project Structure

suds-lab/
├── backend/
│   ├── main.py              # FastAPI app — all API endpoints
│   ├── db.py                # MariaDB connection (credentials from env)
│   ├── train_model.py       # ML training script — run once to build models
│   ├── models/
│   │   ├── index.json                # List of available model slugs
│   │   ├── planter_006_meta.json     # Feature importance + accuracy metrics
│   │   ├── planter_007a_meta.json
│   │   └── *.joblib                  # Trained model binaries (gitignored)
│   ├── requirements.txt
│   └── .env                 # DB credentials (gitignored)
├── frontend/
│   ├── src/
│   │   ├── pages/
│   │   │   ├── About.js          # Home page
│   │   │   ├── Home.js           # Sensor dashboard
│   │   │   ├── Map.js            # Leaflet campus map
│   │   │   ├── NetworkStatus.js  # Node health table
│   │   │   ├── DigitalTwin.js    # ML scenario simulator
│   │   │   ├── Gallery.js
│   │   │   └── ...
│   │   ├── components/
│   │   │   ├── MeasurementChart.js
│   │   │   ├── LiveWeather.js
│   │   │   ├── SensorSelector.js  # Shows hardware info chip
│   │   │   ├── LocationSelector.js
│   │   │   └── SiteSelector.js
│   │   ├── data/
│   │   │   ├── uohNodes.js        # 22 UoH campus nodes with GPS, categories, DB links
│   │   │   └── sensorHardware.js  # Maps sensor name patterns to hardware info
│   │   ├── About.css              # Shared badge/pill/chip CSS used across pages
│   │   └── App.js
│   └── .env                 # REACT_APP_API_URL (gitignored)
└── sudslab_data_loc.txt     # WeatherLink sensor catalogue (254 sensors, 68 nodes)

---

API Endpoints

All endpoints are proxied through nginx at /api/* → localhost:8001/*.

Method	Endpoint	Description
GET	/locations	All monitoring locations
GET	/sites?location_id=	Sites for a given location
GET	/sensors?site_id=	Sensors for a given site
GET	/measurements?sensor_id=&start_date=&end_date=	Time-series measurements
GET	/stats	Aggregate counts (locations, sensors, readings)
GET	/live	Live weather from WeatherLink API (proxied)
GET	/network-status	Last measurement timestamp per site (5-min cache)
GET	/twin/nodes	Available ML models with feature importance and accuracy
GET	/twin/predict	Run a soil moisture scenario prediction
GET	/twin/current	Latest actual sensor readings for a node

---

Digital Twin — ML Surrogate Model

The Digital Twin page uses Random Forest models trained on historical sensor data to simulate how soil moisture responds to different weather scenarios.

Nodes: Planter 006 and Planter 007A (the two best-instrumented nodes with co-located rain, temp/humidity, flow, and soil moisture sensors at 6 depths).

Features: Cumulative rainfall over 1h / 6h / 24h windows, temperature, humidity, antecedent soil moisture at each depth, cyclic hour-of-day and day-of-year encoding.

Targets: Soil moisture at 10cm, 20cm, 30cm, 40cm, 50cm, 60cm depth (one step ahead).

Accuracy: R² = 0.87–0.99 across all depths on held-out test data.

Retraining:

cd backend
python3 train_model.py

---

Local Development

Backend

cd backend
pip install -r requirements.txt
cp .env.example .env        # fill in DB credentials
uvicorn main:app --reload --port 8001

Frontend

cd frontend
npm install
# create .env with: REACT_APP_API_URL=http://localhost:8001
npm start

---

Production Deployment

Backend — restart the systemd service after any changes to main.py or db.py:

sudo systemctl restart fastapi.service
sudo systemctl status fastapi.service

Note: the service runs inside the miniforge3/envs/suds conda environment. Install new Python packages with /home/streamdash/miniforge3/envs/suds/bin/pip install <package>.

Frontend — rebuild and copy the static bundle:

cd frontend
npm run build
sudo cp -r build/* /var/www/streamdash/

The systemd service loads credentials from backend/.env via EnvironmentFile=.

---

Data Sources

• Sensor measurements — MariaDB database (suds_database), populated from the WeatherLink network. ~15 million readings across 296 sensors.
• Live conditions — WeatherLink embeddable API (campus weather station).
• Node metadata — sudslab_data_loc.txt, a JSON export of the WeatherLink sensor catalogue including GPS coordinates, installation dates, and sensor categories.

---

Funding

This project is funded by the UPP Foundation as part of the University of Hull's commitment to campus sustainability and student-led research.