247 – Scaling PBDM (Physiologically Based Demographic Modelling): Localising Climate and Pest Models in West Africa

[00:00:07] George: Hello and welcome to the IDEMS podcast. I am George Simmons and I’m joined today by David Stern.

Hello, David. How are you?

[00:00:13] David: I’m doing well, George. Looking forward to discussing. It’s been a while.

[00:00:17] George: It has been a long time and today we’re kind of going back to, I guess, both our roots in a way, but definitely back to the project that I originally joined on, which is working on this physiologically based demographic modelling paradigm, PBDM. And I guess today we’re gonna discuss a few kinds of new avenues that we’ve come into. We’ve had our grant with McKnight Foundation renewed and that’s opened up new avenues for us to actually start scaling the work that we’re doing.

[00:00:52] David: Well, I suppose it’s what I was always aiming for, that we actually want to use this with projects. And there is a specific project in the West African region, where they’ve been working for 20 years now on things including biological control of pests, and I was originally interested in the CASAS work, partly because I saw its relevance to the work that was happening in West Africa.

And as part of this new phase, we are aiming, and there’s challenges, but it should actually have postdocs from the region working on implementing models related to the Sahelian context and these specific predator-prey interactions and they also, of course, eat millet. So you have a tritrophic system, which is the basis of PBDM, and it’s really exciting that we’re gonna actually potentially do this.

[00:01:55] George: Yeah, I guess just to quickly remind people, PBDM is about trying to construct a more holistic model of a certain ecosystem, and specifically in this case, agro-ecosystem. So it is really accounting for the crops, in this case millet, it’s accounting for the immediate crop pests that eat the vegetative side, or the kind of productive fruit side of millet, and then also the insects and other enemies that eat and predate on those things.

And the idea is to try and capture the entire system in balance. And once you have that, then you can start bringing in what Sahel IPM are doing in terms of their pest control systems and actually start to model those. That’s the really exciting bit.

[00:02:43] David: Absolutely. And I’m just excited thinking about this. We are finally getting here and we are gonna be able to do this. Maybe just to sort of dig in, this is non-trivial to get started, still. I mean, there’s been a lot of work you’ve been doing over the last couple of years with Luigi and Andrew, who we’ve mentioned in the past, actually re-implementing what they’ve done in Pascal in the past in Python.

We now have these Python packages through which such a system can be implemented. We don’t yet have the full user interfaces that we would eventually want, but we could work now with these postdocs to help them to calibrate the models to actually build the models out.

Then there’s still going to be some work done on optimisation, presumably to actually get the models to run fast enough to be useful. And then to build out, as you say, the concrete models for this use case as a first place to be useful. The really interesting challenge in my mind being that at the moment these systems are really continental level, or at least national sort of level implementations, where temperature is the driving force.

[00:04:11] George: Exactly. Yeah.

[00:04:13] David: And I don’t know if that’s going to be true here. This is really interesting. I mean, millet in Niger is all rainfall, the actual base crop is very rain driven. And rainfall is a different beast in different ways. So will it be good enough? These are really difficult questions. We can certainly build the initial models, but then we’re gonna have to ask some of these really difficult questions, and I’m looking forward to that, to digging in, to sort of going through the first iterations, getting something that works, then digging in and actually asking some of these hard questions.

[00:04:50] George: Yeah. And it’s the kind of challenge that we both love because it involves bringing together so many different people, different experts, to actually make this not just representative, but useful and usable. I think this is the really cool challenge about this. But yeah, as you said, from our perspective, a lot of the initial work has just been really on understanding the structure of these models and how to capture them.

And it’s a whole different challenge of how to get other people to be using them, to be creating models, how to make these things, as you said, actually simulate properly, how to allow people to kind of fulfil our vision.  And we’ve talked about this before, this idea of being able to chop and change your models, be actually very modular, very intuitive.

So to actually get to that level where people can intuitively make these changes, saying “what happens if we try this intervention?” or “what happens if the rainfall does this?” Actually building the system that can kind of capture those inputs is really the journey that we’re embarking on.

[00:06:01] David: Absolutely. And we’re really grateful for this grant, which is part of our work for the Global Collaboration for Resilient Food Systems, which is enabling us to continue to work on this to some extent and to support this particular application.

I guess one of the things to dig into a little bit is: how are we actually gonna do this? And this is a discussion we haven’t had yet, but I was discussing with the team in Niger including the counterparts in Burkina and Mali, and they’ve identified some potential postdocs. And then there are lots of interesting questions.

What can we do remotely? Can we get them started building things? When might we want them to come and have a visit? And, when they have a visit, are they coming to visit you, or are they going to see Luigi, or are they going to see Andrew in the US? Who are they actually visiting and why? What is it that they really need? Why do they need to be outside of their environment and when do they need to be outside? What is it we’re really wanting them to become expert at?

I definitely want them to be expert enough to build these models, but they still need to be entomologists in their own right, and working in the regions. Does the data that they already have provide enough information to calibrate the models or are they going to need to do field experiments or lab experiments to complement that data? I don’t have those answers, I’m sure you don’t quite have them either. But this is exactly what we’re needing to now figure out over the next few months. I’m excited by this.

[00:07:34] George: I think the exciting thing about that is that what we do know is that these postdocs need to have a quite interdisciplinary set of skills. It’s kind of building a set of people who can function or interact in the way that we’ve built up this project.

So the example, I guess, is me coming in as a mathematician, knowing nothing about modelling or entomology or ecosystems or agriculture, and learning enough to be able to interact with the people in those environments, and be able to ask at least something that looks like the right question to be able to have that conversation, that interaction.

And I think for the postdocs that are gonna come onto this, the really exciting part of this, is that they will need to build up skills that probably traverse everything that has gone into this. They will need to understand some of the maths and category theory that have gone into this to be able to interact with the modelling framework in the current state because there’s no user front end to this.

They will need to be contributing to that translation from this more abstract system into something more functional. They will need to keep their entomological forefront. And that’s the really exciting thing about this is, then we’ll have been building a group of experts, which have home expertise in lots of these different areas, but who can communicate across.

So I suppose in answer to your question, they will probably need to visit most, if not all of us in some capacity, and/or we visit them. But there are practical issues doing those kinds of visits.

[00:09:21] David: Yeah. 

[00:09:22] George: Which is another dimension to this project. I mean, for me, I’ve never travelled to that part of the world, it’s something completely unknown to me. I know there’s all sorts of advice that says you should or shouldn’t go to this part of the world. And that is an additional challenge that you don’t get in traditional academic projects, I suppose, that’s certainly something to move through. The other thing for me is the language, and that’s going to be very interesting to figure out, definitely big considerations that we need to work through. 

[00:09:52] David: Absolutely. And the first step of course, and we should do this over next week – I’ve just come back, literally travelled back just yesterday when we’re recording this – first thing of course to do is now to set up some remote meetings to get things started remotely, because even if we wanted to start with face-to-face, visas are a challenge, actually organising travel either way on is not something which can happen overnight.

So as the first step, we are going to try to get people at least started remotely, and thinking about how we do that and how we put that together is going to be, I suppose, our first challenge. I’m looking forward to maybe digging into it, maybe we use a Moodle course, and we actually have a remote session. We could actually open that up, not just the core postdocs, but anybody from their team who wants to go through that process to be exposed to and introduced to this material. This could be a possible starting process.

So it’s right at the beginning, this is a new phase, just the starting of a bigger project, and within it, this really exciting opportunity.

[00:11:06] George: There is a question on my mind which might be a good thing to go into. The new title of our McKnight project is Responsible AI for RMS. And certainly part of the next phase, even for this PBDM work, is to understand how we can integrate AI into all of this. These are conversations we’ve been having in little pieces, and sometimes not around agroecology, you know, I went to a workshop a few weeks ago and I presented about AI in relation to education, for example.

So yeah, I guess I’m interested in your perspective on the ways that AI or responsible AI can come into.

[00:11:51] David: Well, I mean, what I’m really excited about on this, and this is where this particular project is so interesting, is that when we talk about responsible AI, in almost every example we talk about the deterministic cores. And what I see in this context is that what with these entomologists we can be doing is building a deterministic core where the responsible AI can then be enabling people to use these deterministic cores to ask good questions, and actually interacting with these modelling frameworks, which mean that there’s real contextualised substance at the heart of a system which farmers could then integrate.

So there have been some recent episodes with Digital Green about some of the work that they’re doing on responsible AI. I think they’re doing some exceptional work there. And I do see these things as coming together, but maybe not very soon because the work with Digital Green on the responsible AI is all about farmer advisory, and at the moment the CASAS work, this modelling work, is really political advisory, it’s at the level of a country, of a county, of a region. And getting that down to the level of a farmer, that’s a whole other question of whether we can get this modelling down to that so these two could meet.

So I guess if we take up the Digital Green work to now say, “okay, well how do we have AI agents to help political leaders with their decision making?” Well, they could then be using these deterministic systems. That’s something which is a few years away, maybe. We need to have the deterministic systems built to the core and then would need the AI agents to be able to interact with those in sensible ways to help policy makers with their decision making. Now, that is a realistic five to 10 year vision.

But the vision that I’m really excited about is: what about this small holder agriculture decision making? When could these models actually be the beating heart of an advisory system for small holder farmers? And that I don’t see happening as fast, there’s so many pieces that need to be worked on. 

[00:14:24] George: There’s so many pieces and we talked at the start about these regional level models, the driving thing is climate.

[00:14:32] David: Wait, wait, I want to pull you up on that. We said explicitly temperature, and that is good because actually the ERA5 data, the satellite data for temperature, these are pretty reliable. As soon as you get to rainfall, it’s not clear they capture dry spells or extremes well enough.

So even that transition from temperature to rainfall, there are questions about scalability and whether this can be made relevant. That’s a whole different ball game. And that’s not even what we are getting to if we are getting down to a farmer level, where it’s not even climate. So, you know, we’re so many steps away.

[00:15:07] George: Exactly, yeah. This is soil, this is the elevation change in your field, anything you can think of may become the driving factor of whether your particular field is successful.

[00:15:20] David: Exactly, so there’s so many things, the previous crops, the history of the field, there’s so many different components. And what’s of course so fascinating is that there’s work happening which relates to all of this in other contexts. So there are pieces which might be able to draw all of this together.

There’s a group who are working on the soils testing toolkit where you could be getting soil information. What if that could then feed it in in a way that was useful? But can you get the useful information? What is the useful information to do this at a small scale? How do you even make that affordable? How do you make it reliable? Oh, so many good questions. 

[00:16:01] George: For a longer term vision. But this is what we’re working towards and this has always been. You’ve always said that the thing that you’ve always been aiming towards is actually to get tools into the hands of farmers. 

[00:16:16] David: In this context, that is my driving motivation. In other contexts there are other things, but in this context, the driving motivation is that. And it comes back to when – and it’s probably a good place to finish this episode – we were in a meeting, a relatively high level meeting, and one of the people in charge joked (this was just a couple of years ago) “everybody’s talking about AI now, but at least small holder farmers in these low resource environments, at least we don’t need to think about AI there”. And I said, “no, wait a second, this is not right, they are the ones who could benefit the most from responsible AI.”

Now, it’s true that AI, as it’s being developed at the moment, is not necessarily what they need, but they need identification of plants and diseases or pests in their environment, they need things which you can do through AI. They need the potential for AI to extend what an extension agent can do, how many people they can reach. So the potential for responsible AI as part of a wider ecosystem in these low resource environments far outweighs its potential in many other environments, this could transform livelihoods in ways which is so important.

And so we can’t separate out these recent technological advances from the needs of smallholder farmers in Niger as my motivating example.

[00:17:51] George: Absolutely. Yeah, I couldn’t agree more. And, I’m sure you often find that some of the people who are using AI the most are exactly these sorts of people, and actually helping them to interact with it in a way that benefits them is absolutely what we want to do.

Yeah. Thank you David for this reintroduction to the next step of our work. We’ll be, I’m sure, doing a lot of chats updating on what we’ve done, how we’ve managed to kind of bridge those things that we talked about. 

[00:18:22] David: Over the next few years, it won’t happen overnight.

[00:18:25] George: Won’t happen overnight, but yeah, we’ll be talking a lot more. So yeah, thank you David for your time and we’ll speak next time. 

[00:18:33] David: Thank you very much.