In 2011 I spent the Bumblebee nesting season painstakingly collecting activity data from the Bufftailed Bumblebee nest (Bombus Terrestris Audax) in our garden. The nest was equipped with a number of sensors (such as temperature) and several CCTV cameras.

At the time collecting activity data was a manual affair - scanning hours of CCTV footage for notable behaviours. One of the things I tracked and recorded was the time the first bee left the nest and the time the last one returned at night.

In the first instance this was a curiosity which allowed me to compute the length of the “working day” for the nest. In general we were noticing that the bumblebees would start at dawn and finish at dusk. And for a whole host of reasons this makes perfect sense: maximising their working day without putting themselves at undue risk, such as getting caught out in low visibility or cold.

The following chart visualises this and shows the period of time the bumblebees worked (green) compared to overall daylight (sunrise - sunset, shown in yellow). You can see they are fairly adept at making use of available hours.

The problem with the chart above is you can’t really quantify how much time has been worked/lost day by day because the y-axis is not fixed. So, the next chart fixes this problem by computing worked hours and day-length, rather than times-of-day.

The blue bars record the total amount of time between official sunrise and sunset, while the orange bars record the elapsed time between the first bee leaving and the last one returning. Additionally, the darkness of the blue represents the level of rainfall and although bumblebees can and do function in some rain, they won’t go out in a torrential downpour. (And we regularly found them trapped out on our lavender on such days if they hadn’t made it back to the nest).

You can see on the days where lots of rainfall was recorded the orange activity level is much lower, as a result of the bees leaving the nest later in the day or returning sooner.

(Note, that on the chart it looks like there is a slight lag between rainfall and lower bumblebee activity, because there is! The rainfall measure is a 24 hour rolling total, taken every hour by our weather monitor, rather than a historical daily total; so at the time a downpour starts, the amount of rain fallen so far is small, but the bees may stop working if it’s heavy. Then by the end of the day you have a sense of how much rain actually fell).

It’s possibly a little bit easier to see on this next chart. On the wettest (darkest blue) days, the bumblebees are out of the nest a lot less.

It was interesting to see that on the very best days the colony actually worked slightly longer than daylight hours. Indeed, we can identify 2 days (orange) where this was the case:

Also, at least during the first half of the season, the bumblebees tracked sunset so accurately, I became fairly adept at predicting when the last bumblebee would return; usually to within +/- 5 minutes with almost 100% accuracy.

But here was the odd thing. The return pattern at night seemed very predictable but the first exit in the morning was much less so. Actually, below is the pattern of first-exits and last-returns to the nest over the season. Particularly over the first half of the season, the bumblebees track sunset very closely (upper pane).

My initial expectation had been that they would leave the nest at sunrise but I could see from the data this was not the case, and when I plotted it, it was very obvious. The chart below expands the one above and shows sunrise (top edge of the blue area) and the circle marks show the time the first bumblebee left. It’s clear that this does not strongly track sunrise and some even leave, rather keenly, up to 25 minutes before sunrise. (Thankfully our cameras were equipped with infra-red).

Again, to normalise everything for the changing hour of sunrise, we can just calculate the differential between sunrise and first exit. So, recalculating the chart above, for each day, the chart below shows the time difference between the exit of the first bumblebee and the time of sunrise. Here the size and colour of the marks also indicates the rainfall.

It’s easy to see from the above that there’s almost no discernible pattern. So, why all the delay? The mean and median differential (“delay” in leaving the nest (y axis)) are both approximately around 60 minutes after sunrise but the range is vast, from 25 minutes before sunrise to 230 minutes after. And it’s not as if the extremes particularly correlate with the days of rainfall (of which there were only 3 that were particularly wet). So what’s going on here?

The obvious next thing is to perform the same comparison against wind. Here we are calling out the windiest days using the maximum wind-speed for the day: larger and redder being the windier days. Again, in terms of time taken to leave the nest, the windy days do not stand out as separate from any others.

The problem with the above analysis, though, is that while it’s easy to look at, we’re approximating to “windy days” and not really analysing the wind at the time the bumblebees left the nest. So, we can plot wind-speed in more granularity (hourly) and it looks like the below. However, it also starts to look a bit messy because to produce this chart required joining the nest data with measured wind-speed from our garden weather station (I did not have wind-speed data in our nest records). This does create a small snag, because wind was sampled on hourly intervals that of course don’t necessarily match either sunrise or the time our bumblebees choose to leave.

Notwithstanding, it gives us an impression of what’s going on. The line plots the wind-speed for each hour and the red circles show the first exit of the bumblebees.

At first glance this appears to tell us a story: doesn’t it look like the bumblebees wait for wind to drop before leaving?

On close inspection, it’s a bit of a yes and no. We can see evidence of a pattern showing that the bees prefer less wind (though we already know this) but the wind-speed is not really predicting the exact time of nest departure. For a start, there are numerous days where there is a choice of times of zero wind: why do the bees pick the particular time they do? What’s more, the graph shows they are willing to leave the nest during speeds of up to 2 mph, so clearly absolute wind-speed is not the deciding factor.

This does help us piece a story together, though. Typically it looks as though they wait for enough light (makes sense) - i.e. in-and-around sunrise; and by preference will tend to synchronise with lower wind speeds where possible. It’s difficult to go further with the conclusions, not just because of the problems with the data highlighted previously, but also because the wind itself has it’s own complex pattern which can’t be revealed in this hyper-local data. It might just be, for example, that wind-speed tends to drop at sunrise and isn’t necessarily the specific trigger.

Ok, so what about temperature?

Here’s a general readout over the season so we can see what we’re dealing with. Unlike the wind, the temperature has a much-easier-to-follow overall trend as well as repeating patterns day by day - the natural daily rise and fall we’re all familiar with. Although on some occasions the temperature does rise through part or all of the night.

Having explored the other variables (wind, light), my instinct by now was maybe the bumblebees are driven more strongly by temperature. We know, for example, that in spring once temperatures start hitting 13C or 14C we tend to see queens coming out of hibernation and setting up nest sites, so the hypothesis is not a huge leap of imagination.

So, maybe there is a minimum temperature that triggers the bees to leave the nest? This is easily assessed in the chart opposite which shows the range of temperatures over which the bees first leave. (The * is the lowest recorded temperature in the data set, which just happens to be 0C).

So, it tells us two things:

1) they never left below 7C. This could, of course, be a quirk of the darkness hours - meaning that this was nothing to do with a behavioural choice on the part of the bumblebees but it just so happened that sunrise was always 7C or higher.

2) There’s a large range of temperatures - there is nothing to suggest a cluster of behaviour around any particular trigger point. Frankly, it makes complete sense: if this is the range of temperatures that Mother Nature is going to throw at the bumblebees, then they have to be able to cope with it. And they do.

By this time I was thinking “is there anything to be uncovered here at all?” All my hunches has been disproved and it seemed there was no obvious pattern to be found. Maybe, in fact, the bumblebees were just more random than I thought/expected; or maybe I just hadn’t been measuring the right variables.

Anyway, there was one more obvious chart to plot - analogous to the wind-speed correlation chart - the time line of temperature with the exit times of the bumblebees also marked on it.

At the time I’d recorded all the nest data in 2011, this chart combining two data sources had evaded me, because - for my sins - I was using Google Sheets to manually capture the nest data. I think it’s fair to say that Google Sheets is not really a visualisation or analytics tool and I never figured out how to correlate two data sets with mismatching time series, without somehow hand-editing every data point and in the process eliminating all credibility of the data sets.

Eventually I had a dabble with Tableau and that’s where I started to make progress. Indeed, even though the two time series don’t perfectly align, Tableau is happy to fit them together as best it can without any real effort on my part. Here’s the required chart. The red circles show the exit time of the bees, while the line and colour plots temperature. The thickness of the line indicates last 24 hours cumulative rainfall.

And boom! Suddenly it leaps off the page - there is a very striking correlation. That almost without fail, the bumblebees leave the nest at the point the temperature stops falling and starts rising. This pattern even holds true for the wet days, so in fact we can see that modest precipitation doesn’t in fact influence the bumblebees’ decision to start foraging. The absolute temperature is also largely irrelevant.

First, let’s explain the red circles that do not appear to touch the blue line - are they are problem? Not really.

The reason the red circles are not touching the blue line is that the temperature data for the lines and circles comes from two data sets and there is a discrepancy between them: typically (sometimes) the sensor on the nest is reading lower than the weather station. There could be several reasons for this. One might be that the sensor on the nest was nothing more than a £1.50 fridge digital LCD fridge thermometer from Amazon, housed in a bit of Sugru, So, aside from possible variations due to quality, it’s possible that (say) on some days it was getting damp and readings were being affected. It’s also possible I misread the temperature from that sensor: all these readings were taken manually from CCTV images in low-light conditions, so we cannot rule out human/visual error.

In contrast, the horizontal alignment of points along the x-axis (i.e. time of day) is robust and automated, so I am satisfied the data is in lock-step along that axis.

Secondly, there are a few points where the exit point does not appear to correlate with a turn in temperature, as highlighted in the chart below (these points also have the problem above, so we need to draw a vertical line through them to establish where they meet the temperature line).

It turns out these points are actually quite easy to explain:

1) The left-handpoint (on the chart above) occurs on a day where the turn in temperature was actually at 5PM the previous day. I.e. this is one of our cases of the temperature rising through the night. My notes record that this day was very wet and windy and clearly the temperature was suppressed as a result. As the wind and rain eased off, the temperature rose through the night. Indeed, the notes also record that a bumblebee returned to the nest at 5:26 am that morning (with pollen) having been out all night (they do sometimes get caught in a storm), followed by the first exit of the day 10 minutes later at 5:36 am. (I have no data to determine whether it was the same bumblebee).

2) the right-hand point appears in the first instance to occur as temperature is falling away, also on a wet day, so seems again to fit a pattern where more-adverse weather is driving an unusual temperature pattern and different behaviour. However, closer inspection of that point (see opposite) reveals there is actually a micro reversal in temperature (several in fact) which appears to trigger/coincide with the bee behaviour. The notes also record similar behaviour to (1) - i.e. the first thing observed that morning is a bumblebee returning to the nest from a night out, and then the first exit being observed.

Conclusion

Indeed, if anything, this latter data-point adds more credibility to the whole argument that the bumblebees are sensing a reversal in temperature as a behaviour trigger. Moreover, it tends to suggest their ability to do so is highly tuned and is effective within a sub-1-degree range.

For me it is an absolutely fascinating discovery and only adds to my amazement at, and respect for, these remarkable creatures.