Computer simulation helps test land management techniques to help bumblebees
exeter.ac.uk
exeter.ac.uk
I guess I immediately have a few questions - How do we know that the model corresponds to reality at all? I mean, I know we have experiments to confirm what impacts the bees, but it seems like there could be a massive amount of variety depending on different factors. I'd love to see some research showing that BEE-Steward can work in real scenarios. Right now, they just have a simulation showing "hey, if we change x and y we can get 5x more bees", but does that actually correspond to reality?
Secondly, how do we make farmers actually want to use this? Do they get paid for taking care of bees on their land? I know this is just a policy question, but it's still important to answer. Do farmers need any training to use this? Can we collect data in an automated fashion to reduce the work the farmers have to do? This almost sounds like a deep learning problem - use a camera to map + classify areas of farmland, how many pollinators they have, the presence of bees, etc.
Finally, what is the fundamental problem of bee modeling? Is it fundamentally a computational problem (where we know the rules, but computing the model can be hard) or is it a data problem (where we know how to compute how the bees will behave, we can do it efficiently, but coming up with rules that bees actually follow is hard)? I've noted that the NetLogo environment by default will only take up 1GB of RAM, which is already not very much considering the JVM itself will take a pretty big chunk to run just about anything.
It would be interesting to start building some sort of "National bee model", rather than just working farmland by farmland. Is there any project like this in the works?
Either way, I'm sure all of my questions and more are already being asked/answered by people much more involved with the conservationist field. Just my 2 cents.
In the U.K. at least, they do. The government pays you to leave some part of your field unused and untouched. It’s called ‘wilding’ and it’s well paid enough that the subsidy sometimes is worth more than any reasonable crop that would grow on that patch of land.
Jeremy Clarkson’s Farm (of Top Gear fame) had a segment on this. My take is that farming is a very subsidy-heavy business.
Nobody in the UK (or in England at least, i'm not as familiar with devolved policy) gets paid for full rewilding (in the sense of taking land completely out of production and leaving it, a philosophy most famously espoused in England by the Knepp estate). English agri-environment policy expects some kind of active management of the land taken out of production. There are a range of options farmers can choose which they have to discuss and agree based on the land type and context with a specialist adviser. For bees this might be cultivating a plot of plant species chosen for pollen and nectar, for example.
The payment is calculated on a model based on "income foregone plus costs" so in theory should be worth exactly the same as an equivalent crop on that patch of land (although in practice this obviously varies a bit depending on how marginal the land is and fluctuations in the market). Generally the amount of admin and time required, and the limited funds available, means that nobody is making very much money off this.
The much more broken UK subsidy is the Basic Payment System - a leftover from the EU Common Agricultural Policy. Here farmers are paid a flat rate per hectare for any land used for agriculture - just cash handed over with no incentive to either improve productivity or manage for the environment. For obvious reasons this disproportionately benefits massive landowners. This is thankfully currently being phased out and the funds shifted towards the agri-environment model, which has its problems but at least delivers something.
I say this as someone who has spent many years writing ecosystem models for schools (some of them with NetLogo). Toy models often require more than that to be useful.
You'll have a set of assumptions. You'll have a number of formulae that you can individually say there is experimental evidence they roughly models reality, if everything else is held constant. But you almost certainly do not have the experimental evidence to say that when these six formulae are interacting together, they will behave in known, deterministic ways.
The paper does not appear to show any applications where researchers were able to predict something novel using the simulation, test it in the real world, and see how well it matched the results. Their "results" (if I'm interpreting correctly) appear to be mostly summarized in figure 3, which seems to be created by trying different settings until the model very, very roughly matched some historical dataset. Someone correct me if I'm wrong.
If there is one single person to read on system dynamics modeling, it's John Sterman. Also, the Systems Dynamics Review (SDR), https://systemdynamics.org/system-dynamics-review/ , is what I think of as the journal of record. I am also aware of some work by the Concord Consortium for simulation in education. What seems to get written up is either high effort/high cred (SDR) or edu/toy.
I would describe all my work as "toy" models, in the sense that they are designed to demonstrate some aspect of an ecological system in a pedagogical way. The purpose of any one model is generally (1) to show 2 or 3 emergent phenomena, (2) to simply showcase the complexity of interconnected systems, where (say) eliminating a predator species can have surprising effects on the ecosystem at large, and (3) to introduce students to the idea of modeling as a practice -- i.e. making it explicit that they are working with a model, that models are not reality but make certain assumptions and decisions, and that the students themselves may be involved in making those decisions, and needing to defend them (from some work with Uri Wilensky, I also know that this is explicitly why NetLogo models always allow the end-user to edit the code, and this cannot be turned off).
I would describe a credible ecosystem model as one from which justifiable predictions can be drawn. These would need to be demonstrated experimentally.
Personally I don't know any credible ecosystem models, because that's not my field, but I'm sure that some exist, to varying degrees of credibility. My skepticism of the Bee model was based more on my knowledge of how high that cliff would be to scale, and how this model seemed not significantly more complex than plenty of models I have written.
For instance, if bees are getting afflicted by some fungus, then there is some research that says that applying fungicide X will allow the colony to prevent y% decline in the ideal case, or whatever.
At that point, even though the model will then incorporate this "sub-model," you don't need the system dynamics model to say what will happen when you apply the fungicide, you only need that well-researched sub-model.
I feel (and I'm getting a little out of my depth because of my limited research on real-world uses of these models) that the system dynamics model is only useful for attempting to determine what happens when you have a conjunction of various interventions, or interventions with various environmental parameters that were outside the original research of the sub-model. But that's precisely were the limitations of the system dynamics model will come into play, because the non-linear or chaotic effects may mean that the model's prediction that the colony will grow or fail may have little to do with reality, because of the complexity of accurately accounting for so many parameters, and guessing how they interact without specific research on how they interact.
A friend of mine is a beekeeper in one of the northern US states. Bee keepers there keep bees on farmland in exchange for most of the honey. The farmers generally get more than they can personally eat. Sometimes keepers will take a farm from another keeper by offering more honey/a sweeter deal (ha!). Farmers also get crops pollinated for free, so it’s a win/win for them.
Original model implemented in NetLogo (as described on the page).
Some papers I found looking for the model:
https://besjournals.onlinelibrary.wiley.com/doi/epdf/10.1111...
https://besjournals.onlinelibrary.wiley.com/doi/full/10.1111...
Wish I had some time to explore making a game out of something like this while preserving as much of the accurate simulation as possible.
They use a different matrix.
There's a fine line between improving survival and finding the absolute bare minimum to prevent loss. Situations change, and red-lining nature is always, always going to end in unpleasant surprised.