There are likely over 1 billion farmers in this world without access to the internet, how can their lives be improved? Every single year, thousands of tonnes of produce are lost to unforeseen weather by these farmers. This has a profound effect on the environment, but also those farmers' livelihoods.

This industry has the potential to be a multibillion dollar industry in no time.

What it does

Making use of multiple sensors, such as; Humidity, Cloud Cover Classification, Sunlight Level, Air Pressure, Temperature and Elevation, the device then predicts the weather using multiple Machine Learning models. The ML models were trained on over 150,000 rows of data on weather prediction.

The device is placed around the farms of rural farmers where it records the environmental factors and uses the ML models to predict the weather for the day, and the proceeding 5 days with 90% accuracy rate in testing. These predictions require just the readings from sensors, no internet.

How we built it

Due to the lack of adequate sensors, a Raspberry Pi 3b and Arduino Uno Rev 3 were required to run the system. Two sensors are connected to the Arduino; one for Temperature, Pressure & Altitude and the other for Humidity & Temperature. These readings are transferred to the Raspberry Pi and fed into an XGBoost model, and a speaker announces the forecast to the user (as it was found a huge amount of rural Asian and African farmers are illiterate).

Cloud cover and sunshine are judged using a PiCamera and cloud classifier image processor and is taken at multiple points during the day to further improve model performance.

Challenges we ran into

The lack of computing power in the Raspberry Pi has been a constant hurdle. Even installing simple Python packages can take hours on end, only to fail and require restarting.

Hardware inclusion has been difficult as there are numerous variables required to make a dependable prediction.

Accomplishments that we're proud of

We are proud of producing a cheap and beyond trustworthy weather predicting device that will allow those whose lives depend on the weather to make more informed decisions about their work and plans.

What we learned

We learned a huge amount about pin mappings and electrical requirement of hardware. We also learned how to make a brilliant classifier quickly but carefully at the same time.

From the model we learned what appear to be the most important factors in weather prediction, which are; Humidity, Pressure, Cloud Cover and Temperature.

What's next for WeatherStation

Add a solar panel and more sensors to further improve estimation and predictions!

Share this project: