by Huw Evans
I have been a beekeeper for almost 15 years, I have always loved keeping bees as for me it’s the perfect mix between a science and a craft. However, I remember returning from some particularly unpleasant swarm inspections in the Spring of 2008. The bees were busy and did not want to be disturbed, meanwhile I was shaking the bees from each brood frame in search of queen cells.
Despite the stings my sympathy lay with the bees. I recall considering that in this day and age, if a surgeon can replace a valve in your heart via a vein in your leg, surely there is a way of finding out what is happening inside a bee hive without taking it to bits. I am an electronic engineer and my wife is a biologist, so we together started researching into physical parameters that could help non-intrusive and remote bee hive monitoring.
Most beekeepers can relate to the fact that there is a lot you can tell about the condition of a colony from the noise that greets you the moment you remove the inner cover so its hardly surprising that hive acoustics have been used as a diagnostic tool for thousands of years.
The first written document referring to a change of sound in a pre-swarming colony was in Georgics IV, by Virgil, ancient Rome’s most celebrated poet and a beekeeper back in 50BC. This was backed up by Columella a hundred years later in his De re rustica which says “. . . He will be able to find out beforehand their decision to escape by putting his ear to each of the hives in the evening . . .”. In 1609 Charles Buttler speaks of the sounds of a pre-swarming colony in the “Feminine Monarchy”. Charles went on to write a madrigal (song) using the sounds of pre swarming bees (see YouTube). Similarly, in 1759 Thomas Wildman writes about a peculiar humming noise in the hive three to four nights before the “the swarm sallies forth”, but also points out that the interpretation of this sound varies from author to author owing to the strength of imagination in each (quite pertinent even today). Probably the most famous pioneer of sound analysis came in the 1950s. Edward Farrington Woods, or “that nice Mr Woods” as the queen of England used to refer to him, was a sound engineer at the BBC. There is a phenomenon called the cocktail party effect in which human beings have the ability to eavesdrop on a conversation across a crowd even if they are not the loudest voices in the room; its like a filtering thing. Woods was a BBC sound engineer and had a very musical ear, you needed one to be a sound engineer back in 1954. Today you only need to be good with a computer, you could be stone deaf and a perfectly good sound engineer in this day and age! Eddie identified a sound in pre-swarming colonies he called a ‘warble,’ he attributed the source of this sound to be redundant house bees. He produced a portable electronic gadget called the Apidictor, this was in effect a band pass filter which allowed the beekeeper to better hear sounds of interest. A hole the size of a regular sink plug was cut in the rear of the brood box. The plug was simply removed and the microphone inserted in the evening, once the flight commotion has passed. Eddie conducted experiments for over a decade and published his work in both the New Scientist and Nature. In 1995 a modern day Apidictor called the Bee Tone Analyser was published, however the same operational drawbacks remained i.e. an evening trip to the hives remained necessary and, possibly more importantly, the interpretation of the sounds was left with the beekeeper.
The reliability of the Apidictor remains a moot point. Following an experiment at Rothemstead (a major bee research institute in the UK) brood boxes with sink plug holes in the rear could possibly still be found lying around today. Using Apidictors, hundreds of hives were tested by several beekeepers, one of them being Eddie Woods. Overall the results were fairly inconclusive, however Eddie himself did very well!
I think that it is now generally accepted that accurate detection of a queenless or pre-swarming colony cannot be simply recognised by the majority of beekeepers using an Apidictor. However, there is rarely smoke without fire and we felt that with the application of today’s technology (Digital Signal Processors and complex signal processing algorithms) it was worth investigating further. We set up a large experiment recording the sounds of bees while making regular physical inspections so we could correlate the evolution in sound with changes in colony behaviour/dynamics. We studied black bees (Apis mellifera mellifera) in the UK and yellow bees (Apis melifera ligustica) in Italy. This allowed us to consider differences in both bee breed and geographical location. To date we have over 30 Tb of bee sounds; throughout the experiment we also recorded other physical parameters such as temperatures, humidity and weight.
Our original focus was one of swarm prediction. Like Woods, we created a frequency spectrum of the sound in the hives. We took spectra from consecutive days and complied a 3D graph or ‘soundscape’. On the next page you can see the first two soundscapes we ever constructed.
These neighbouring colonies were of comparable strength with queens from the same breeder. A warble can be clearly seen in the right hand trace, as a discontinuity on the right hand face of the hump. This colony did indeed swarm! Indeed the third soundscape continues for a week beyond swarming (the warble disappears following the swarm).
The following year the swarming colony began to warble again, this was not warmly received as we expected role reversal this year. However, sure enough, the swarming colony swarmed for a second consecutive year! So those hives also gave us our first genuine swarm prediction. By ‘genuine prediction’ I mean looking at the data and predicting who was going to swarm before the event, rather than looking at the data retrospectively and convincing yourself that it could have been predicted.
Unfortunately our initial excitement was short lived, although accurate swarm prediction continues to thrive in the minds of researchers it is hampered by many obstacles when applied to the diverse world of ‘real life.’ Different hive types or even just adding supers can change the nature of the sound scape quite dramatically. Some bees preparing to swarm don’t warble much while some colonies appear to warble when superseding. It really boils down to colonies of bees having different accents or even speaking different languages. Furthermore, some bees begin preparations to swarm then give up! However, a seasoned swarm predictor with acoustic experience of their bees does stand a good chance. Arnia continues to work with several acoustic experts and universities around the world in search of a generic algorithm robust to bee breed and hive type and plans to be beta testing swarm prediction algorithms as soon as next season. If anyone reading this thinks they can bring anything to that party, please contact us for some sample data to play with.
However, our experiments were by no means a failure. We concluded that acoustics fell into two distinct categories. The first we call ‘smoke and mirrors’, in which you try to recognise behavior or health issues from a specific ‘acoustic signature.’ During our experiments, we also noticed anomalies in the soundscapes of bees with Varroa and Nosema. However, like generic swarm prediction, these remain ‘works in progress.’ The second category is things like flight noise, the amount the bees are fanning, even the total amount of noise is a good indication of colony strength; it’s only an indication but even that can be useful. There is no smoke or mirrors, these parameters are more straightforward to measure with considerable less influence from bee breed, hive type or microphone position.
We went on to develop hardware. Each hive is monitored for acoustics, temperature and humidity; the monitor fitted above the entrance or fitted to a dummy board inside the hive.
Weight is measured using a hive scale positioned below the hive. The low profile rectangular ‘doughnut’ design is particularly suited to open mesh floor hives as debris can fall and ventilation is not restricted. At a height of only 1¼ inches there is little need to adjust hive stand height. The standard scales are fully wireless, and come in a variety of colours as seen in Figure 2.
An intuitive user interface is possibly the most important component when it comes to making a monitoring system useful to beekeepers as seen in Figure 4.
At a glance the user can see the condition of each hive. Each hive icon represents a monitored hive in the apiary. Current readings from each sensor are displayed in sensor icons; the two hives on the right also have hive scales. The cloud of bees above each hive show that hive’s activity, so at a glance you can see which hives are strong and which are weak, very like when you enter your own apiary. The weather bar along the top shows the current weather conditions in the apiary and what they have been over the past week. The weather bar also displays cellular signal strength and battery condition of the Gateway, that of each monitor can be found under each hive. As seen in Figure 5.
If the user clicks on a sensor icon they are taken to the Graph View which shows historical data from that sensor. It is possible to compare other sensor data from the same hive, from other hives or even data from other beeyards, all on the same graph.
In Figure 6 we can see a very stable brood temperature (Green line), this is what we expect to see in a queen right colony as the bees thermo-regulate the brood area with surprising accuracy. The brood temperature from a neighbouring hive is simply added by clicking that sensor icon. Here we can see that the brood temperature becomes unstable (Blue line), at that time we received an automatic alert and following an inspection we confirmed the queen had stopped laying. Following the introduction of a new queen the brood temperature begins to stabilise once again. For comparison, we can also add the air temperature at the apiary, the day night fluctuations are clear as is a particularly warm week in June.
Weight is a very useful measurement. During nectar flow, an increase in weight is seen as bees return with nectar, what is possibly more interesting (but arguably less useful) is seeing the weight drop during the night as the bees process the nectar. This is demonstrated in Figure 7, where we can also see when honey supers are added. There is a system of comments, which allows the user to log such hive manipulations.We can also see a sudden 7lb drop in weight, it was a SWARM! Although we are not currently offering an automated swarm prediction algorithm weeks or even days before a swarm, as Tom Seeley notes in his book ‘Honeybee Democracy’, “One could almost predict with fair reliability when you would find your first swarm by noting when the hive of bees ends its six-month-long-free-fall in weight and begins to build up again on fresh nectar and pollen” and that is certainly true in this case.
Adding the neighbouring hive’s weight, we can see that initially the smaller colony is less productive but following the swarm it soon catches up. See Figure 8.
Meteorological information can also be added to any graph using the three weather icons to the left of the hive icons. Weather data is often useful when used to put other sensor readings into perspective. In Figure 9 a drop in productivity is simply explained by rainfall.
What sets arnia’s system aside from ‘hive scale’ based monitoring products is the diversity of measurements, the addition of acoustics in particular, and how easily they can be compared on the same graph. In Figure 10 we can see a sharp drop in weight, is this robbing? Or has a rock holding the roof down rolled off the hive? By simply including flight noise we can clearly see that this colony is being robbed. You can also see the bees briefly return on day three but soon give up as there is no honey left to steal. Users can zoom in on any area of interest using the lower scale, either by using the mouse or ‘pinch’ zooming on a tablet or smart phone.
Not only does this give us the opportunity to alert the beekeeper of robbing (15 lbs of honey went on day two!) but this is also valuable as ‘black box’ data, better explaining when and why the colony failed.
Most ‘backyard’ Beekeepers tend to treat each colony as an individual, almost like a pet. One day they will requeen one hive while adding a super to another. Therefore it is useful to monitor every colony in their Apiary. Commercial beekeepers tend to manage all colonies in the same way at the same location. Therefore they only need to monitor a handful of ‘sentinel’ colonies to alert them when the supers are full or the nectar flow is over or there is a dearth in forage etc.
Today we can reliably and remotely monitor Spring build up, brood state, queen status, track forager activity, accurately map nectar flow, nectar processing and monitor weather conditions at the apiary. We can compare trends over time and with neighbouring hives.
We can receive mobile alerts for robbing, if the bees swarm, if the colony becomes broodless, if you need to feed or ventilate during the winter months, if its time to add or remove a super, or simply when the bees have collected a lot of nectar that day. The latter has no managerial consequence but is possibly the most enjoyable alert to receive!
The system has a security feature, which informs the beekeeper when a hive is moved. Furthermore, there is a tilt alert if the hive falls over. A colony most often survives the Fall; the colony does not fail on impact. It’s the resulting exposure before discovery, often for several days or more that can be fatal. We get a lot of positive feedback on how much better some of our customers sleep when comforted by the knowledge that a text message will wake them if their hive blows over on windy nights.
Electronic/Remote hive monitoring is in its infancy and its ultimate application to beekeeping will continue to evolve with use over time.
Huw Evans is the co-founder of arnia, a research and development company that designs and builds hive monitoring equipment. arnia hive monitors are currently for sale in the U.S., for more information contact email@example.com.