Hypothesis: 

We hypothesized that:

• Past season movements will substantially increase the chances of winter mortality. 
• Movement will have an acute effect on hive health during the season.
• Reducing the amount of moves within a single season will be more economically viable in the long term.

Method(s)

We used an anonymized dataset of 184,199 hives over three seasons (2021-2024) collected with Nectar’s platform. Then, we used machine learning to develop a hive simulation for an operation with 5000 hives. Using economic information about hive yields anonymously extracted from Nectar customers, we then computed a range of outcomes for different movement scenarios. 

The results(s)

We found a strong negative effect of moves on colony survival. We also found a positive effect of past season moves on the survival of past season survivors, suggesting an artificial selection of hives may result in hives more acclimated to frequent moves.

Our simulation demonstrates the potential economic gain of distributing pollination contracts evenly across hives in an operation, and that moving bees less than 6 times and avoiding to travel long distances with the same group of hives can have a significant impact on bee mortality.

We also learned that, if moving hives multiple times per season cannot be avoided, whenever possible beekeepers should:

1. Alternate which of their hives are sent to fulfill pollination contracts to minimize the cumulative mileage and number of movement for each colony.
2. Plan movement schedules that don't assign hives to a sequence of stressful locations, such as locations with poor forage availability, potential spray risks or high colony density. 
3. Increase prices for contracts that require a quick pace of movement. 
4. Increase prices for mid to late summer pollination to compensate for the opportunity cost of sending hives for pollination when they could be moved to areas of high forage availability. 
5. Wherever possible, beekeepers should avoid dispatching their hives to holding yards before sending them to individual pollination yards. Nectar’s data suggests that this logistical approach places unnecessary stress on the bees that can be avoided with a more effective yard mapping system. 

Co-written by Nico Coallier, CTO at Nectar, Ryan Kuesel, Brandon K. Hopkins, Maxime Fraser Franco,  in collaboration with Sam Venis, Communication Specialist at Nectar.

Context

As far back as 2021, one of Nectar’s key objectives was to understand the hidden impacts of beekeeping practices that had become standardized and commonplace, but that beekeepers themselves believed were putting stress on the bees—like poor nutrition, or the effect of pesticides. Beekeepers would often talk about how these practices were causing hive related problems, whether in the form of reduced honey or hive death. But no one had the data to understand the extent. 

One of the most impactful of these “hidden killers” is hive movements: the number of times and frequency that a hive is moved from place to place, whether inside a particular region (from a holding yard to a honey yard, for example), or outside of it (ie. moving hives out of state/province to a pollination contract or honey yard). Some beekeepers suspected that each time a hive was moved it had a corresponding reduction in hive strength—leading to lower pricing for pollination contracts, and a subsequent reduction in a hive's likelihood of surviving the winter. Some beekeepers even build business models around this fact—sending hives to as many pollination events as possible within a given season, on the basis that, if that hive is going to die at the end of the season anyway, they might as well get as much money for it as they can.

At Nectar, this theory of pollination directly challenged our initial assumption that better, healthier bees lead to stronger, more sustainable businesses. Healthy bees have lower mortality rates, stronger overwinter rates, less disease. So to optimize for short-term gains, at the expense of bee health—to juice the hive for all it’s worth—seemed counter-intuitive. 

With this in mind, Nectar began a process of uncovering the real impact of movement on honeybee hives. By this time, over 280,000 hives across North America were using Nectar to manage their hives, creating the largest database of hive-level information ever generated. This database is colored with granular details on when hives are moved and where, plus when they were treated, fed, made queenright, sent for pollination and more. So we made a plan to learn as much as we could from the data—specifically, as it relates to movement.

Step One: Building the “Real World” dataset

Our first step was to create a working model.

Starting with the data of 29 operations who regularly use the Nectar app to track “deadouts”, we generated an anonymized dataset of 184,199 hives and looked at their data over a three season period (2021-2024). Typically, beekeepers evaluate the health of their hives and record dead outs many times per year, but especially after the winter, when they observe how many hives have survived.

One of the first things we noticed is that hive mortality increases among hives between each movement in a non-linear way, as visualized in the chart below. We also learned that colonies that underwent more movements had a larger variability in mortality. There was also a correlation between distance travelled and increased mortality.

The average survival of hives in the dataset is 79% with an average hive age (i.e., time since tracking of the hive began) of 271 days. We also showed that older hives generally had a higher chances of mortality, so survival probability decreases among older hives. This was expected because old hives, like older creatures of any species, are more likely to die.

Step Two: Building the model and the simulation 

With the real world data of 184,199 hives in hand, we began a process of ‘model testing’ to learn the best way to simulate beekeeping outcomes. 

To understand how simulation works, imagine that we have a record of all the NFL game outcomes stretching back to 2020, with player-specific game statistics. With this data in hand, imagine we built a model to simulate what might happen in a future season based on all this information. Because we know what really happened, we can check the accuracy of the model by comparing its predictions to the real, historical outcomes.

In short, that’s what we did with the bees: the studies we conducted in phase one (described above) were like the NFL game statistics—what really happened. So when we started building our model, we could validate its accuracy by comparing it to this history. Once we could prove that our model was accurate based on past data (that it accurately predicted what actually happened), then we could use it to predict scenarios that haven’t actually happened (ie. simulations). 

This was the focus of the next phase of our study. We created seven different kinds of models and compared their ability to accurately simulate the past. Eventually, we developed a model that predicted hive mortality based on past data with over 93% accuracy, as you can see below.

With this model in hand, we were ready to answer this question:

Is there any scenario in which it makes sense to increase the number of pollination contracts (and, consequently, movements) that a hive makes per season, rather than reduce it, in order to maximize its economic yield? 

Or, in other words: Is there an optimal number of moves in the life of a hive that maximizes revenue over the lifespan of a hive, rather than optimizing its revenue within a single season?

In order to answer this question, we needed to make some assumptions about the expenses and income sources a beekeeper might accrue while running a commercial apiary. For example, the income a beekeeper can expect from a given pollination visit, the number of days in between moves, the distance traveled between moves, whether or not it’s a pollination visit, etc. 

A breakdown of the inputs we used for the simulations can be seen below. While some of these estimates are highly subject to fluctuation, it should be noted that we wanted to build a model on the basis of conservative estimates. That way we could avoid the risk of overstating our case. 

Our inputs included following data:

• Distance traveled: From 0 km to 50 km (31.06 miles) per visit, with a maximum of 2700 km (1677.7 miles) over the lifecycle of the beehive.
• Cost of transportation: $0.001 per hive per km traveled, because we estimate that the beekeeper can move between 40 and 450 hives on a semi-truck (or other type of truck) with a gas price of 1.25$/liter (0.33$/ gallon).
• Number of days at location: A minimum of 7 days and a maximum of 60 days, with a mean of 35.
• Income from Pollination: Between $60 and $220 per pollination event to 40% of the moves. For example, a hive could do 6 moves in its life and fulfill two contracts of different values summing to $145 total while another hive could do 9 moves and 4 contracts for a total of $600. The number of movements are limited to 40% of the total per season which is similar to what we observe from our pollinators users in the database.
• Survival rate: After each move in the simulation, we use our predictive model to predict whether or not the hive would survive to the next move. If the survival rate dropped below 50%, it would be considered a deadout and would not be used for another move.

The simulation ran 5,000 imaginary beehive operations through thousands of different scenarios, and analyzed what happened when the bees were managed in different ways. Some had more movements while some had less; some yielded more money per pollination visit; some hives were moved at greater distances.

Note that we didn’t allow the AI beekeepers to create more hives within the seasons, in order to estimate their loss of income without the additional costs of scaling back up their operation after high seasonal mortality. In other words, for an operation of 5000 hives, if 2000 hives died in the first season, the next season starts with 3000 hives and so on. This way, we optimize the short term income of a pollination contract in function of its long term effect on survival, trying to find balance between revenue and negative impact of moves. 

In the figure above, we can see the answer to the question: do hives that move more per season die younger? As you can see, at 6 moves/season, a big drop in hive longevity occurs. This is true in both the model and the raw data. 

Next we wanted to learn: does this also correlate to revenue ? 

The figure above shows the average income for each hive in function of the number of moves done each year after 3 seasons. We can see that hives that move over 10 times per season present lower income. This is in line with the figure above where we show that longevity drops heavily at 6 moves per season. 

On the right, we see the relation between hive mortality and distance travelled during move, where a similar pattern is shown. However, we can see that when a hive reaches over 4,000 km travelled per season, the variability of impact becomes high. In other words, according to these studies, moving bees less than 6 times and avoiding to travel long distances with the same group of hives can have a significant impact on bee mortality.

Importantly, we observed this difference even though we didn’t include revenue generated from honey and nucleus production, which generally account for about half of beekeepers  revenue. In other words, when you include the additional revenue per hive that comes from these sources, it becomes clear that optimizing for survival over a longer period of time is a much better strategy. We didn’t include these income sources into the model so we could reduce the number of variables we were studying at one time. 

However, we also predict that with stronger bees that survive for more seasons, honey and nuc revenue will not only remain the same, but they will increase. If we even assume a small percentage increase in survival, the corresponding increase in revenue will be significant. Future Nectar studies will attempt to build these scenarios with real world data.

How can beekeepers use this information? 

In practical terms, we think this study has a few important implications for commercial beekeepers. 

1. The main takeaway is that, if moving hives multiple times per season cannot be avoided, when possible beekeepers should consider:
   1. Alternating which of their hives are sent to fulfill pollination contracts to minimize the cumulative mileage and number of movement for each colony.
   2. Planning movement schedules that don't assign hives to a sequence of stressful locations, such as locations with poor forage availability, potential spray risks or high colony density. 
   3. Increase prices for contracts that require a quick pace of movement. 
   4. Increase prices for mid to late summer pollination to compensate for the opportunity cost of sending hives for pollination when they could be moved to areas of high forage availability. 
2. Wherever possible, beekeepers should avoid dispatching their hives to holding yards before sending them to individual pollination yards. Nectar’s data suggests that this logistical approach places unnecessary stress on the bees that can be avoided with a more effective yard mapping system. Holding yards present high density of beehives, low floral resources and unknown disease level in the population. In other terms, they can be a great contamination source and have a significant impact on colony survival.

• As an example of how this kind of system can be designed, consider Miller Honey Farm, who practices a strategy called ‘low touch beekeeping.’ You can read about their approach on Nectar’s website here.

As of December 2024, beekeepers can build their own insights dashboard in Nectar’s Manager portal. On this dashboard, we added the moves count per hive so users can keep an eye out on the impact of moves on the health of their bees. Being a beekeeper who uses Nectar on a day to day basis, I personally track these impacts on survival and varroa level using this dashboard. If you need help setting up a similar dashboard, contact customer success!

While these changes could be hard to implement without a good tracking system, this information is easily obtainable with a system like Nectar. And this is only the beginning.