090 – Introducing George Simmons: Mathematical Modelling in Agroecology

[00:00:00] David: Hi, and welcome to the IDEMS podcast. I’m David Stern, the founding director of IDEMS, and I’ve got the pleasure of being here today with George Simmons, an Impact Activation Postdoctoral Fellow at IDEMS. Hi, George.

[00:00:20] George: Hi, David.

[00:00:22] David: This is our first episode together, I hope not our last. It’s great to have you on our IDEMS podcast and I’m keen to dig in to what you’ve been doing this last year. You’ve been with us almost a year now.

[00:00:34] George: Over a year. I joined last September.

[00:00:37] David: It was last September, there you go, just over a year.

[00:00:39] George: Yeah, and time has flown.

[00:00:41] David: It has. And you came in specifically on a piece of work, I suppose, I have a very soft spot for this work. It’s the work that we do on agroecology. I’ve been working in West Africa for 10 years now in this area. And you came in related to this, but on a very specific piece of work, which has been great fun to me because it’s actually doing maths.

[00:01:04] George: Yeah, and I can see whenever we talk about this, me and David, his eyes light up and he’s always keen for me to actually come and visit him in Reading in person so we can actually work through stuff together. Yeah, I definitely echo that sentiment.

[00:01:16] David: Yeah, you understand that this is something where, for me, it’s really great to have something really mathematical to get my teeth stuck into. And basically, we’re doing mathematical modelling related to, if you want, invasive species or generally systems dynamics involving insects and plants, and so on.

[00:01:34] George: Yeah. This project started, you would go into a group called CASAS Global. So these are non for profit group of research scientists led by Andrew Gutierrez, who is a professor emeritus at UC Berkeley, California. And essentially his life’s work has been developing this style of modelling, which is known as Physiologically Based Demographic Modelling.

And the idea behind this style of modelling is to really account for the ecosystem level dynamics that happen in a farm field. So the growth of the crop, the herbivore eating the plant, the parasite or predator impacting the herbivore, and being able to then frame this as not just farm level models, but look at this at a regional scale, looking at ecosystem dynamics, looking at invasive species, the potential for invasive species, and particularly how you can actually manage these things. Yeah, very interesting piece of work.

[00:02:44] David: And we came in originally with the idea of just supporting them to actually modernise a little bit the implementation. It was originally Pascal code and we were redoing a Python implementation of this. But we’ve got sucked in, haven’t we? And we’ve been discussing this, we’ve integrated more modern mathematical modelling techniques and it’s just so much fun.

[00:03:07] George: Yeah. It is great. And, I guess where this has gone to, as you said we started with the brief of make this modelling type more accessible. And, yeah, we’ve now grown into actually thinking more generally about how can you make complex systems modelling more accessible, more impactful, easier for people with perhaps not the specialisms you would expect to actually go and build models or even contribute to them. So some of the structures behind that are really fascinating.

[00:03:40] David: Absolutely. And I suppose before we dig into you a little bit more, when I’ve been involved in this sort of modelling before, particularly around crop modelling in agriculture, the fact that the pests and diseases were not integrated into a lot of those models meant that they could be used in certain ways, but they wouldn’t necessarily be close enough to really help farmers with decision making in ways that could transform what we can do.

And I now believe very firmly that the modelling approach that they’ve developed, although it’s used mostly at these sorts of large scale models, there’s potential that this could be brought down to farm level decision making as well.

And so this is something which is not going to happen in the next few years. But I’m really excited that this is a journey we’re going to be on now for the next five, ten years, where this approach that they’ve got with the sort of more modern methods together, and the sort of collaborative model approach of our collaborators, Topos and so on, it’s just, it seems to be coming together at the right sort of time.

The computational power is there, the mathematical knowledge of what to do is there and their approach to modelling on this sort of systems model shows that this can be really impactful and reusable.

[00:04:56] George: And yeah, just to pick up on where you started there, there are very sophisticated crop models out there, which are used, I think, Apsinth used in Australia by farmers.

[00:05:10] David: And all over.

[00:05:11] George: But as soon as an invasive species comes along and starts eating your plants they suddenly become unhelpful. And a traditional view among farmers is you see an invasive species, you apply a pesticide. But what you don’t know without that systems modelling view is what that pesticide is actually doing. Is it just controlling the insect you’re spraying for? Is it controlling one of its natural enemies? And actually by spraying, you’re actually creating a niche for what you wanted to control or another herbivore coming in to actually come in and take over and actually that will start eating your plants instead. Without this kind of systems view sitting on top of your crop models, then, you as a farmer have no way of anticipating what issues you might face or even how to control them.

[00:06:00] David: And just to put this into perspective, if you don’t take this approach that has been developed by Andrew and his team, then quite often when people look at invasive species models, they’re looking really at sort of correlation models, which basically is just saying what’s happening in other places that look like this? That’s a fair description, isn’t it?

[00:06:22] George: Yeah, exactly. And the correlative style of modelling is very popular because these things have always been fairly easy to put together. You’re running, multilinear regression analysis, but even more increasingly so with the rise of machine learning technologies, developing one of these models, which essentially correlates between observational data taken from certain geographies or climates to try and predict how likely it is that invasive species will appear, much more easy to put together.

But that style of modelling does come with a lot of problems. It doesn’t allow you to extrapolate beyond where your original parameterization data set came from. So these models are not very good at predicting invasive species outbreaks in new geographies.

There’s a great example of that with tomato pinworm, which started coming to Europe in around 2009, 2010. And all of the models about tomato pinworm were based on its native environment in South America. And you might imagine the climate there is tropical, warm. And all of the models applied to Europe predicted that the species would settle in the Mediterranean band where climatic conditions most matched its native environment.

But what they found is actually this species started colonising much beyond the Mediterranean band. It spread north into Europe, up through France, even up to the UK, around the Black Sea up into near the Baltic region. And this just demonstrates, while, these correlated models are so easy to put together they, very rarely give you the full picture.

And the second part of that is beyond geography is actually what a lot of people are interested in is how our food systems are going to evolve with climate change. And, again, trying to use just these correlative techniques to understand how invasive species are going to migrate as climate changes due to climate pattern predictions, is going to have the same problems as trying to apply them in new geographies.

[00:08:49] David: And just to be clear, that actually, when you take the approach developed by Andrew, this PBDM, have I got that right?

[00:08:57] George: Yes, PBDM, Physiologically Based Demographic Modelling.

[00:09:00] David: Exactly. What you’re actually doing is because you’re modelling the physiology and demographics of the populations, then actually you’re looking at those interactions, you’re looking at how the systems interact with each other. And that’s what enables you to then, actually, go beyond what exists. You’re taking the knowledge of the ecosystems and you’re actually putting that in to create a simulated ecosystem.

[00:09:30] George: Exactly. And the correlation models there, the data of the insect physiology and its environment coupled together, and what PBDM gives you is this uncoupling and allows you to just parameterize the insect based on its physiology, its response to, in general, things like temperature, like humidity, like in the case of plants, like rainfall, and it uncouples that from those environmental factors. And by doing that, you actually then gain this predictive capacity outside of the envelope where he started.

[00:10:05] David: Exactly. And this is where, if you think about this more generally, and this is what I love with the way that we’ve now dug into this, you get towards the sort of collaborative modelling approach that Topos has put forward. So do you want to just tell us about that as well?

[00:10:20] George: Yeah. So the Topos Institute coincidentally are also based in Berkeley, California, at least started there. And they started by a group of mathematicians who are really interested in trying to understand real world modelling problems using mathematical principles.

And one of the, what I would describe as their goal, is to actually develop the mechanisms that allow a complex model to be not just built, but contributed to by lots of different people at once. I see the analogy as being, building a software on GitHub. GitHub is able to manage all these different contributions by lots of different people, it’s able to stitch together things in coherent and useful ways, and it also keeps track of all the history that came with your software package, who committed stuff, why things were committed, reasons for any changes, all this kind of stuff.

And I see that what Topos Institute is trying to build is exactly the same thing of being able to stitch together models in a general way from contributions from lots of different subject experts alongside all of the reasons, the assumptions that were made in the model, the track record of which publication of the data came from, the reasons why one data set was chosen above another.

[00:11:44] David: And there’s elements of implementation, which we won’t dig into, but broadly speaking, this resonated to us. And we were in the middle of building the system for the CASAS team. And we then came across the Topos work and it changed and reframed us. We both have a background in category theory because we’re pure mathematicians where this was part of where we’ve come from. And it really resonated and what we found as we brought this in is it’s also resonated with, and I think Luigi in particular from the CASAS team, as part of what we’re building.

And it’s just so exciting how these pieces are coming together. And it’s, as we say, it’s allowed us to build something which is almost certainly going to have applications well beyond this single use case.

[00:12:33] George: Yeah. I always joke, just going back to category theory, that, as a topologist, studied topology in my PhD, there’s this inevitable march towards the abstraction towards category theory of those topological concepts. I thought, no, I’m going to leave that academic route, go and search out something else. And here I am at IDEMS, studying category theory. Yeah, it’s great.

[00:12:55] David: I promised you something more applied, but there you go, back to category theory.

[00:13:00] George: You can’t escape it, but I think that is a very good lesson in the general things that Topos are trying to pursue, that actually, the ideas of category theory really do capture structure in everyday systems, be it models that we’re talking about, be it databases, even statistical theories, physical modelling theories, they’re all built up out of components, and that’s essentially what Topos is trying to study.

[00:13:32] David: Absolutely. Let’s get a bit back to your own journey. How did you end up at IDEMS?

[00:13:37] George: It’s a very interesting question. I was introduced to IDEMS by Lily Clements, who you’ve known for many years now. Was it 2016, 17, something like that?

[00:13:49] David: Absolutely, and regular listeners will recognise, Lily and I have had many episodes together.

[00:13:55] George: So Lily was a statistics PhD student at Southampton. We started at the same time, became very good friends. She finished a year ahead of me and came to work full time at IDEMS. And I took another year to finish off my PhD, and as part of my job search, I reached out to David to see if there were any opportunities going. And I guess it was a good timing, good coincidence that, at the time, IDEMS were preparing for this grant with McKnight CRFS scheme.

[00:14:33] David: It’s of course interesting that you actually didn’t apply for this. You applied for…

[00:14:38] George: Oh, that’s true.

[00:14:39] David: You applied for a position in collaboration with Edinburgh University, Chris Sangwin’s also done an episode with us on STACK, and we were really keen to take you for that and there was one position and there were two people we were keen to take, and Sal got that job and you didn’t.

And then this other opportunity came up and the timing on that meant that we didn’t want to go through another recruitment process and so I approached you as to whether this would fit. And as chance would have it, this was a better fit than what you applied for.

[00:15:12] George: Yeah and, I almost couldn’t believe my luck in a way. I guess I should say what kind of things actually attracted me to IDEMS in the first place was looking to pursue, the areas I was looking for were some kind of mathematical or scientific consultancy, some kind of data science roles. And when I came across IDEMS, I saw that while these two things were not the primary focus of the company, they are things that people get involved with as they work.

But my kind of other motivation is I’ve always had this thing in the back of my mind that I never thought I’d get into, which my uncle on my mum’s side grew up in Guernsey, got small island fever, and at the age of, 18 decided, no, I’m off. And he moved first to Portugal, but then down to Mozambique. And he’s been there ever since. He has a family there and he works in an organisation that essentially makes sure or tries to make sure that aid, international aid actually gets to the places where it’s supposed to go.

And I was always fascinated by this, I had really no common ground with him to talk about this, and really no way of, going through pure mathematics undergrad and into PhD, no way of actually seeing how I could reconcile those two viewpoints together. So suddenly IDEMS pops out as this kind of, oh, here’s a way of combining your mathematical expertise with ways that you can actually be impactful. Yep, sign me up. So….

[00:17:03] David: And you’ve been a great addition to the team in so many different ways. And one of the things that I think you’ve done really well in the way that you’ve settled in, is it hasn’t just been about your core focus, you’ve got involved in other things, building courses, and so on. And so I guess the final thing to finish this first discussion is, if you think about your different experiences that you’ve had over the last year, what is it that you can share that you expected, that you didn’t expect, about the experience? How has it challenged you? How has it put you out of your comfort zone? And how has it met with your expectations?

[00:17:45] George: I suppose, when we were actually discussing this role over a year ago, one of the key questions was, how do I feel about being this interface between IDEMS and CASAS? Actually, talking about, talking with and integrating with researchers whose primary focus is entomology, ecology, biology, completely different language mindset. And I think I said, I have no idea, but I’m happy to try. And, that has been a massive challenge, changing my vocabulary, training myself to take their concepts and articulate back to them through this kind of mathematical funnel, in a way. But it’s, on the flip side, been a very rewarding process.

And I think I mirror that kind of mindset in the other things I’ve done at IDEMS. I think one of my first projects, not on my main work was, you said, developing courses for the EU satellite agency EUMETSAT and we were developing courses to teach skills on time series analysis. But part of that was actually going in, downloading their real data sets and making courses based on their real data sets. And that was such a steep learning curve to understand how to work with data, how to fetch it, how to then do data science on it, actually getting my coding skills up to scratch so that I could get to the point where I could teach someone how to make plots about things.

And I’ve really enjoyed that kind of steep learning curve that’s come with everything I’ve done in IDEMS. It’s very challenging to get your head around, and one of the things I guess I’ve most struggled with is actually the concept of being approached to do something like, oh, go and make courses for time series analysis. There’s so many words in there that I’d never come across before. Make a course, time series analysis, statistics, data. But actually the process of going through that and learning all of those techniques and then actually putting it on paper for someone else to read is just so rewarding.

[00:20:06] David: Absolutely. And I think this is one of the things where the nature of your role is exactly to have these opportunities to be challenged, and to have people around you who can help you sort of unpack those words, as you put it. But at the same time, that this is something where you actually do have a skill set, which means that none of those words, when it comes to it, are actually that challenging. They’re just new words.

You actually have the skills to be able to deal with all of them, this is the nature of your fellowship, it’s about, activating the fact that you actually have the skills to do this. You just need to get exposed to, be supported, to be enabled to take on those challenges. It’s been great, and the way you’ve risen to these challenges has been great.

I want to come back to the way you put it, and I think I’ve seen this with both Andrew and Luigi, who I’ve seen you spend time with in different ways, but it’s that personal relationship that you’ve managed to build, because this is what many mathematicians struggle with. And I think you’ve just done brilliantly, you’ve built that relationship to understand their language. And by doing so, you suddenly realised, wow, what they’ve done is amazing.

Often I find many mathematicians, they like to remain in their domain. And that means that they don’t actually appreciate the value that the other domains bring. But the way you’ve immersed yourself and seen, and we’ve had discussions about how much you admire the work that’s happened, as you mentioned, Andrew’s sort of lifetime of his work. And it’s just so impressive. And I think that appreciation for it, and the threat of taking you to the field is something which you’ve…

[00:21:51] George: Yes, I’m sure Andrew will eventually follow through on actually getting me to sit and study a plant. But I guess, yeah, my final analogy based on that is: I was sat with Andrew, I actually went to visit him in Berkeley, in March, to learn about the style of modelling he’s developed. And one of the ideas he had as we were sat together was, I wonder if we can start with observations and observational data and somehow run the model backwards, as he puts it, to understand what some of the parameters were in the first place. And eventually you realise he’s just describing a machine learning technique.

And actually, once you have that, as you say, that relationship, that ability to communicate with people who have approached this their whole life in a completely different way, you can actually draw that parallel and say, yeah, that’s something we can actually start to attack.

[00:22:54] David: I think it’s a, it’s been such a pleasure to see you grow into this role and, yeah, I look forward to further discussions. I’m sure this will not be our last conversation.

[00:23:04] George: No, absolutely not. Yeah, you finally roped me into it.

[00:23:08] David: I have. It’s taken a while, but you’re here now. It’s not gonna stop.

[00:23:12] George: It’s been very fun. Thank you for talking to me.

[00:23:15] David: Thanks.