Data From the Beehive

In 2019, the we4bee project at Julius-Maximilians-Universität Würzburg (JMU) set out with an ambitious goal: It wanted to digitise beehives and use the collected data - for the benefit of bees, beekeeping and ultimately society as a whole. After all, as pollinators, bees make a major contribution to securing agricultural yields and thus the basis of humanity's food supply.

we4bee was launched by a research group at the JMU Chair of Computer Science X (Data Science) headed by Professor Andreas Hotho, by Würzburg bee researcher Professor Jürgen Tautz and by the non-profit business corporation we4bee. The project was initially funded by the Audi Foundation for the Environment and later by the Bavarian State Ministry for Digital Affairs.

Successes of the we4bee Team

Spread across Germany, a good 100 beehives have been "made intelligent" so far. They are each equipped with 14 different sensors that measure temperature, humidity, weight and other parameters. Sound and vibrations are also recorded at a selected beehive. The beehives are mainly located in schools, where they also fulfil teaching purposes. One hive is located at the Ministry of Digitalisation in Munich.

Thanks to financial support from the Free State of Bavaria, it has so far been possible to store 101 billion sensor data, more than 10.5 terabytes of images and audio files on the JMU servers, including more than 20,000 hours of audio recordings. The recorded sensor data is also available to other research institutions on request. The data has also been made available to the public via a web application on we4bee.org and the specially developed we4bee app.

Recognising the Activity Phases of Bee Colonies

One result of the project: the five seasonal activity phases of the bee colonies can be automatically recognised from the sensor data: Spring brood rearing, summer brood rearing, autumn brood rearing, brood-free autumn period and winter brood rearing. This can be achieved by combining conventional sensor data such as temperature and humidity. But it also works with audio data.

The change point detection method was used here. This algorithm searches for distinctive points in the data that indicate changes in development, so-called change points. One example is the start of the brood in spring: the temperatures in the hive rise significantly above the outside temperatures and the weight of the hive increases continuously. In addition, change points are always accompanied by a change in volume in the hive.

Bee Calls Announce Swarming

A neural network was used to detect anomalies in order to be able to read the behaviour of the bee colony from audio data alone. The focus was on recognising the swarming of a colony and the period shortly before. Result: Both points in time can be recognised by AI with a high degree of reliability.

The period before swarming is particularly relevant for beekeeping, as an automatic early detection system can indicate this event in good time. Beekeepers are then prepared to capture the swarming colony. In the audio recordings, the pre-swarming period can be recognised by a distinctive sound (wav.file), a kind of "tooting", which presumably comes from the queen. This finding was submitted as a publication to the ^30th International Conference on Systems, Signals and Image Processing, a European conference on machine learning.

Outlook

The we4bee team considers the audio data to be the most informative data obtained from the beehive. It is therefore planning further research to recognise more events from the audio data, such as the absence of a queen. The recognition of activity phases is also to be further improved.

Website of the we4bee project (in German): https://we4bee.org/