Data collection has started, as below:

I've carefully chosen data parameters that seem to be important from past experience. The most obvious environmental parameter to collect is sun intensity as it's extremely unlikely that the bees will try and swarm in the dark. The importance of the various parameters can be checked when analysing the trained model but to get started we have to rely on experience and educated guessing. The next factor of importance is quite possibly the month of the year as the bees never swarm before April. Temperature is a another major factor and if the average 48 hour temperature is low, then it is probably because there is a cold snap. Rainfall is probably next important and this system is using a website that estimates rainfall from satellite information and is generally quite accurate. Wind speed is probably next important as the bees generally wont fly in strong winds and would find it hard to difficult to maintain the aerial structure of the swarm. Bees are unlikely to want to swarm in cold easterly winds and seem to much prefer to swarm in high humidity for some reason. Lastly, and probably least important, is atmospheric pressure although analysing the model could well prove me wrong.

Checking for swarms will be done every afternoon at about 6pm and probably more often if the CO2 levels in the catch hive dramatically increase. The catch hive itself will be checked if nighttime CO2 levels are significantly higher than normal.

A fellow bee keeper recently reminded me that swarming very often occurs after a period of wet or stormy weather, so a factor could be introduced into the data similar to 'average temperature in the last 48 hours' but it would relate to rain and 'storminess'. Also, bees tend to do a lot of scouting for their new home the day before they actually swarm so a value for 'max CO2 level on the previous day' was added to the data structure.