Low-Cost IoT Flood Early Warning System: A Community-Level Monitoring Solution Designed for the Himalayan Region

This project is a community-level flood early warning system designed for the Himalayan region (Teesta and Rangit river basins in Sikkim, India), built using Arduino and Raspberry Pi with a total cost of approximately 4000 Indian Rupees (350 Chinese Yuan). The system integrates 7 environmental sensors and uses a random forest model to classify flood risks in real time, filling the coverage gaps of official monitoring networks. Original author: Priyanti08; the project is open-sourced on GitHub (link: https://github.com/Priyanti08/Flood-early-warning-system) and is under continuous maintenance.

In October 2023, a South Lhonak Glacial Lake Outburst Flood (GLOF) occurred in Sikkim, India, destroying the Teesta-III hydropower dam and causing over 55 deaths. Downstream communities received only 18 minutes of warning. This incident exposed the lack of a national monitoring network in the upper Himalayan basins and the near-zero community-level early warning capabilities. This project aims to provide a low-cost, easily deployable local warning solution for the Teesta and Rangit river basins.

Hardware Configuration: Total cost within 4000 Indian Rupees. Core components include Arduino Uno (for sensor data collection), Raspberry Pi4 (for real-time risk classification), 7 types of sensors (water level, temperature and humidity, rainfall, flow velocity, soil moisture, air pressure, turbidity), and alarm devices (LED, buzzer, servo motor).

Software Architecture: Core files include app.py (Flask dashboard + real-time inference), train_model.py (training the random forest model), simulator.py (generating synthetic datasets), etc. The Flask dashboard provides functions such as real-time readings, water level prediction, and threshold ETA estimation.

The system uses a random forest classifier to categorize flood risks into 4 levels: Normal, Watch, Warning, and Flood.

Dataset: Due to the scarcity of real data, 6000 synthetic data entries were generated based on physical laws, referencing 115 years of rainfall data from the India Meteorological Department, 61 years of meteorological records from Gangtok, and hydropower station flow data.

Standard Alignment: The warning thresholds are consistent with the standards of India's Central Water Commission (CWC) (Warning: 65cm, Danger:89cm, Extreme:120cm).

Environment Requirements: Raspberry Pi requires Python3.11+, and dependencies include flask, pyserial, joblib, etc. (Installation command: pip install flask pyserial joblib scikit-learn numpy pandas).

Startup Steps: 1. Upload the Arduino program; 2. Connect Arduino and RPi via USB;3. Run python app.py;4. Visit http://raspberrypi.local:5000.

Retraining: Run simulator.py to generate datasets, then run train_model.py to train the model.

Current Limitations:1. The model is trained on synthetic data and has not been validated in the real world;2. Short-term forecasting uses linear extrapolation, which may underestimate the risk of glacial lake outburst floods;3. Only supports single-point monitoring;4. Consumer-grade sensors have only undergone indoor testing.

Future Directions:1. Validate with real CWC monitoring data;2. Introduce complex hydrological models to replace linear extrapolation;3. Expand to multi-node networks;4. Upgrade to industrial-grade sensors;5. Integrate satellite remote sensing data.

This project embodies the concept of "Technology for the People":

1. Low-Cost Accessibility: The hardware cost is only 350 Chinese Yuan, far lower than professional monitoring stations;
2. Local Autonomous Monitoring: Communities can deploy and maintain the system independently, filling gaps in official networks;
3. Open-Source Sharing: Open-sourced under the MIT license, promoting technology dissemination and local adaptation;
4. Democratization of Disaster Prevention: Combining AI and IoT to protect lives from natural disasters.