Richard Inwards wrote this in 1893 in his book ‘Weather Lore’, and it turns out that throughout history, bees have been used as an indicator of weather.
Bees typically fly at around 15mph, so will struggle in high winds and often refuse to fly if the windspeed is over 12mph. They also struggle in the rain as large raindrops can damage their wings, and flying with a wet body requires much more energy. Temperature is also a factor, bees typically prefer warmer weather.
So, it’s fair to say that bees are fair-weather flyers, and studies have shown that temperature, solar radiation, humidity, wind speed and air pressure are all factors that bees seem to take into account when deciding whether to leave the hive and how far to fly when foraging for food.
We tried to mimic this decision-making process using a Machine Learning algorithm. The World Bee Project is currently working on several worldwide projects using BeeHero sensors. These intelligent in-hive sensors provide detailed data about bee activity, including the number of bees entering and leaving the hive every 15 minutes. We combined this data with local weather data, including rainfall, air pressure, temperature, wind speed, etc., to create a 3-month data set for 25 beehives across three locations.
We then created a predictive model called a Random Forest Regression Model, a model which uses the historical data to create lots of decision “trees” — hence the “Forest” in the name. Below is the result of a 4-day prediction for one of the hives. The blue line shows the model’s predicted number of bee trips compared to the actual number of trips recorded by the sensors, which is shown in orange.
The results look pretty good and seem to support the theory that bees are using weather data to make decisions. However, the weather isn’t the only factor, and bees will also consider things like existing food stocks and hive health, so this type of modelling is probably too simplistic for an accurate prediction.
