Amjad Karim (00:03):
Hi, I am Amjad Karim. In this show I’m speaking to Neil Strong and Tom August. Tom is a computational
ecologist at the UK Centre for Ecology and Hydrology, and Neil is the biodiversity manager at Network
Rail. Both of them use computer vision and AI to monitor habitat and biodiversity either in network rail
or in the UK countryside. It’s a really interesting conversation. I hope you enjoy the show.

(00:31):
I thought we could start with what you do, what you have been doing, and just some of your backgrounds really.

Tom August (00:37):
Yeah. Hi. So I’m Tom August. I’m based at the UK Centre for Ecology and Hydrology and we do research on large scale long-term changes in the environment, whether that’s in the air and the freshwater
habitats and fresh habitats and my work there focuses on how wildlife is changing over time, what’s driving that change and what we can do about that. So in the UK we’ve got a long history of monitoring
biodiversity. This goes back hundreds of years to natural historians who have been out recording nature,
but in more recent times, that’s kind of picked up with things like the invention internet and mobile
phones. More and more people are getting involved in recording the wildlife they see around them. We
call this citizen science and we can use these data that people collect to understand how wildlife is
changing. This data underpins our current understanding that quite a lot of biodiversity is changing in
the UK with some species groups seeing quite significant declines. I’m interested in how we collect this
data. I’m interested in the tools that we can develop to help these certain scientists do what they do,
and I’m also interested in how we can use technology to interpret that data, do analysis on that data
and disseminate what we learn from it. And most recently I’ve been interested in how AI can help in
species recognition from images. So this is a kind of tool which we can use or we can create for citizen
scientists to help them to get towards the right identification. So if you’re not too good with your plant
ID, for example, and these AI tools can help suggest what that plant might be and help you get towards
the correct identification. So off the back of that, we’re also interested in how these AI tools, not just for
vision like images of plants, but also for acoustic, say for example and beyond identified birds from their
song, how these sorts of needs can be deployed in the field long term. So kind of like a weather station,
but for monitoring biodiversity.

Amjad Karim (02:52):
Thank you, Neil.

Neil Strong (02:53):
I’m Neil Strong. I’m a biodiversity strategy manager with Network Rail, which might seem a slightly odd
organisation to be involved with this conversation, but we have a larger state that goes across England,
Scotland and Wales. It’s about 50,000 hectares, which if you get the Isle of Wight and a half and you
squash all of our network together, it’s about that same size and the citizen and science element, it’s
something that we have to really think about. We’ve got 7 million neighbours who peer over the fence
and wonder what we’re doing when we’re managing our estate, whether it’s the teams who are working
on the actual trucks and the ballast or whether it’s those teams who are going in managing the
vegetation and tying in, being able to let those people know what we’re doing and why we’re doing it.
But also there’s an opportunity using the techniques that Tom has talked about to get them to help us collect the data so that we already know when, for example, the bird nesting season is, it’s sort of that
March to August, September time depending on the weather conditions.

(03:53):
So we know that we have to take special cab when we’re going into a certain area. But there’s an
opportunity, we’ve got 7 million people and as I say, looking over the fence wondering what we’re
doing. Is there a way to engage them with what we’re doing so that when we go into an area we have to
go in to do the work in a particular area because there might be particular issues with the trees leaf fall
during autumn, so we have to go in and do the work at some other time during the summer. Can we use
data that we get from those people who can then identify where those nests are, now actually we know
where we should and shouldn’t be going. Getting that information, that buildup of quality data then
means that we can start predicting when, so the following year we know we’ve had so many records
that there was always birds nesting in this particular location at this time of year. So we then can put
that onto a map and just know when we should be going. There’s that sort of information, that
intelligence gathering that the railway does already. We have trains that go around the network, 125
miles an hour, and they have a camera underneath the train that is looking at each individual
component, that whole railway together, all the clips and the rails and the sleepers, and that is then
monitoring when there are ones that are missing and it uses the computer to identify where there could
be an issue. And what this is doing is taking away people from being stood in and around the track and
it’s that safety element moving those inspectors and those surveyors away from the track. If we can do
that same sort of thing with the environmental information, then we can remove people from the
railway estate when they don’t need to be there. And we can start targeting now, whether it’s with
cameras on the fronts and the sides of trains to look at the vegetation or whether we’re using drones or
aerial imagery from helicopters and planes or even satellite imagery, we can then start targeting where
we need to go. Only then send somebody to that site so that we’re not having to go three or four times
and put them potentially in a place where there are dangerous situations. They’ll manage it safely, but
by not having to go there in the first place, then you can’t injure yourself when you’re there.

Amjad Karim (05:59):
What are the things that you’d like people to know or you wish they did know about the work that you
do, what you do and they think that it’s probably not perceived appropriately or not understood as well
as you would like? If you had one thing.

Neil Strong (06:10):
I think it’s true from our side of things, it is trying to demonstrate what we’re doing and why and that
there is a reason to it. And primarily we have to look at the railway safety. The reason we manage the
trees and the habitats around the railway is to make sure that the trains and the passengers on it, get to
where they need to get to quickly, efficiently, safely. And then the teams who are doing that work, they
can also be safe. If they’ve got to stand next to trains when they’re going past a 125 miles an hour. If
we’re able to demonstrate what we’re doing and why, if we’re able to put the plans in advance of the
work so that people can see, oh, right, I can understand why they’re coming in to cut those trees down
because I can see the evidence that those trees have caused damage to the trains.

(06:55):
But I can also see what the plan for that site is. Yes, the trees are being cut down, but they’re either
being allowed to regrow, so that the trees will come back and then there’ll be a cycle of management
and we will go back regularly every 15, whatever years it is. But also then we might be changing the habitat type and then working with people like Tom and the data that we’re getting from the species
level and the orders of the animals that we’re finding. If we can see what’s there before we do the work
and then we can see what comes back, once the habitat structure has changed, we might be able to
demonstrate to people, yes, okay, the types of biodiversity in the species that you were seeing
beforehand, they won’t come back because the species you saw were woodland species, and actually
what we have now is a load of scrub and grassland. But actually look at all the species that are now
interested in coming to the new habitat. And so we we’ve got that ability to demonstrate that the
decisions we’re making, they might change the look of something, but actually the biodiversity isn’t
faded, it’s just different biodiversity. But the overall importance of those habitats still improved. We’ve
still got those green corridors, that connection between one habitat and another. It’s just different
things that are being able to use it.

Amjad Karim (08:06):
It’s really interesting what you said earlier about the total area of Network Rail estate is the size of the
Isle of Wight. So I don’t know what that is in terms of how it compares to a city. It’s large, right? There’s
a lot.

Neil Strong (08:18):
Yeah, it is massive, but it’s very long and thin. The closest rail to the fence to that fence, the average
width across the whole of network is only 12 metres. And if you imagine the first five or six metres of
that is where all the stuff sits that runs the railway, that’s where all the cables are, that’s where all the
signalling systems are. So you’ve only got about six metres of green stuff that is there on average across
the whole of the network. And we’ve got to fit all that biodiversity into that area, but we also need to
manage it so that it doesn’t get in the way of the railway.

Amjad Karim (08:48):
Yes. One of the questions that would be interesting, and we’ll come back to it, is to understand, given
that it’s so narrow, how much biodiversity can you pack into and do you have to think about the type of
biodiversity that you actually can realistically do. You’re not going to have elephants roaming around in
that space, so you’re not going to have savanna, so what can you do?

Neil Strong (09:05):
But the problem with biodiversity, it doesn’t read the books. So you will get the biodiversity and it’ll live
wherever it thinks is a nice place to live. It doesn’t actually have to be that academically perfect habitat,
which is always the challenge. And that’s the interesting thing.

Tom August (09:17):
Yeah, and I think Neil, with it being that long thin network, the context is extra important, right?
Because I guess you are interested in things like how does these long thin corridors link up other
habitats? The difference for you in managing your land compared to a large continuous nature reserve,
is that you’ve really got to think about what’s around you and how the habitat sites interact, right?

Neil Strong (09:45):
Exactly. And that’s where some of the work that we’ve done with you guys has come into it. So we’ve
got that 50,000 hectares stripped through the middle of Britain. But the data that we’re using to understand that landscape, that we’ve worked with CEOs to get, goes a kilometre beyond the fence, and
that gives us a data set that is three and a bit million hectares, which is the same size as all the national
parks in Britain lumped together. I like lumping things together. So that shows us the landscape and
then we can start seeing where that connection is.

Amjad Karim (10:17):
Like connected corridors is probably quite important for what you do or do you consider it, I suppose,
but Tom, then I suppose, do you want to talk about some of the projects that you guys are currently
doing and if they have some overlap with the work that Neil is doing it, it would probably be nice to talk
about that.

Tom August (10:33):
Yeah, so I guess one of the main themes around use of AI methods is scale. So usually we’ve come to
decide we need use sort of methods because we have data coming in at massive scale. So we have lots
of data coming in from sitting scientists who are using mobile phone apps to take pictures of animals
and plants that they see, but we have a limited pool of experts who can review those and say, yep, that’s
right, or no, that’s wrong. So we see these kind AI methods as a way to help those experts. So being able
to perhaps take care of the things that are really easy for an AI to take care of, leaving those really
challenging ones for the experts, the experts using their expertise where it’s most needed. I think the
same kind of goes for linking over to Neil is the satellite data that we have. So we have these satellites
going around the earth and they’re collecting images of the earth’s surface, and what we really want is
to translate what they see. So sort of red, green, blue, classic Google Earth image, translate that into
habitats. So we want a map of habitats. That’s what Neil needs. And so again, we can use kind of
machine learning methods to do that at scale. So rather than having a human go around and drawing
lines around hedgerows and woodland patches, you can have an AI which will say, yeah, this bit over
here is a grassland, this bit here is a woodland. And then you can then do that, like Neil was saying that,
or is it one kilometre either side of the entire rail network. You can do that at scale. So this kind of
operating with these big data sets of scale and interpreting that raw data into something meaningful,
the species ID or the habitat that’s there, it’s a really common challenge I think in modern ecology.

Amjad Karim (12:19):
What is a habitat? Is it a set of plants growing in the region? Does it include the animals in that area?
What do we mean when you say habitat?

Tom August (12:27):
Yeah, so I guess broadly you think of habitats as things like woodland, grassland, heathland, salt, marsh.
These are types of habitat and they’re defined by the community that’s there. A community of plants
and associated insects, animals that are there and they can be defined at various scales. So I say
woodland is a habitat, but also you could subdivide woodland into broadleaf woodland, evergreen
woodland. You could define it by beach dominated woodland. So you can have these various levels of
specificity, but yeah, they’re important for understanding what is there and then for merchanting, how
that changes over time.

Amjad Karim (13:07):
So you’re using satellites which are image data that you’re processing to identify and classify habitats.
You’ve done some work taking camera data from trains, but also you are looking at things like moth
traps. What is that?

Tom August (13:23):
Yeah, so we did some work with Neil where we kind of built these weather stations for biodiversity and
we tried all sorts of different technologies and one of the best technologies we used was a moth trap,
which had been developed by researchers, of this university in Denmark. And this is essentially a white
screen, white kind of wooden board and a little box that sits in front of it with a camera in it. So this
camera is looking at this whiteboard and at nighttime the whiteboard is lit up with a bright light and that
attracts moths to it. So moths as you know, are attracted to light, that’s why I see them flying around
the street lights and things and bump into you kitchen window. And they’ll land on this board and the
camera will take pictures. So again, there were thousands of images every night and then we use an AI
to look at these images, find where the moths are and then identify them there. So we’re working with
collaborators at Media Institute in Canada who work on that AI element. And what’s exciting about this
is that it means that you can put these systems out somewhere, which is perhaps hard to access. So
maybe halfway off a mountain where you don’t want to be trekking out all the time to go and collect
data. So you can put in these remote places. Also thinking about in the tropics and places like that, you
can put them out and they’ll run autonomously. They have solar panels, data storage, so they can sit
there for months at a time. And they’re also good for getting this fine temporal resolution of data. It’d
be quite expensive to send a field worker to go out and record moths every night for a month, but you
could put a station out on day one and then pick it up on day 30 and get that continuous time series.
That’s good for looking at actions which you can have impacts these small timescales. So it might be like
weather events, agricultural spraying events and how these might affect moths.

(15:11):
So yeah, it’s an exciting project, but at the moment we’re trying to test these systems in lots of different
places around the world. There are people as well all around the world who are doing similar things and
interested in these sorts of technologies, not just moths, but also daytime pollinating insects, which are
a big interest to people all over the world. So we’re trying to bring this community together as well and
work together towards these effective solutions.

Amjad Karim (15:37):
Oftentimes people would say the job of Network Rail is just to make sure that the trains run in time and
the track is safe. Why do you spend time and effort focusing on this other thing, that is about the
environment?

Neil Strong (15:50):
Why is a railway counting moths on…

Amjad Karim (15:53):
Yeah. Why are you doing this? Should you be?

Neil Strong (15:55):
Yes, we should be because we can have the best train system, but if you’ve got things that are getting in
the way of those trains, they won’t run safely and efficiently. If you’ve got trees growing up through the overhead power lines, then they won’t work and you won’t be running trains safely. So we have to
manage the habitats that we’ve got next to the railway and these things have built up over the last 200
years. It used to be grassy fields and meadows, which are well managed and maintained. With the
changes in technology, we’ve moved into allowing the trees to grow back because they weren’t the
issues we had with steam trades, but now we’re getting to the point when they’ve come too far back
and we need to get this even so that there’s balance between where we can have the vegetation and
where we need to manage it. And so in doing that, we are affecting the habitats that are growing
currently on the railway. And so being able to monitor them and understand what’s in them, and what
we will be affecting when we come in and do that management. That means we need to know what’s
there, but then we also need to know what we’re creating and to be able to then demonstrate that
actually the way that we’re going that the management technique is the right one to do. So we know
we’ve got to get rid of the trees in certain places, but what we then need to work out is whether we
allow those trees to regrow, so to coppice them and let them grow back and then just get that into that
routine management or whether we need to move the habitat, move the trees a lot further away and
actually start creating and maintaining a different sort of habitat. And then understanding what species
are using that sort of habitat. And then so the techniques that Tom’s talked about, the moth trap and
other methods of detecting small mammals, bats, that sort of thing, that then helps us to demonstrate.
It’s all about building up the data sets so that we can show that we’re having an impact, but a good
impact on that biodiversity.

Amjad Karim (17:45):
So the data that you get, so satellite data, you get data from the moth traps, you get data at quite high
resolution. How do you practically use it to change or kind of influence what you do? Because quite
often it’s quite difficult to just avoid cutting a single tree or avoid touching a single plant or a single
location on the network. So you have all of this high resolution data that’s coming in from machine
vision, AI, from citizen scientists. What are the cultural barriers to taking this and adopting some of
these techniques and things like that?

Neil Strong (18:19):
The challenge is having the knowledge within the rail industry to be able to make those decisions, and
that’s why we need to look to experts to help us make the right decision about the habitat site. So
you’re right, we’re going to get a whole, long list of moth species. The majority of which don’t even have
a common name. It’s all Latin, literally, to these railway engineers and they’ve got to make a decision as
to what the work they’re doing. So that’s where we have to work with the experts and it’s the group. So
when we’re talking about the definition of the habitat site, there will be that list of moth species. They’ll
be able to create this sort of this tree that demonstrates that this clump here likes a particular species of
plants or it likes a particular structure of habitat. And if we can see that and then work out from those,
whether any of those species are sort of the rare species or ones that are in decline. If that ties in with a
habitat type that works next to the railway, then that’s a massive opportunity the railway has because
the land on our side of the fence is relatively undisturbed. You can’t go and walk your dog through it.
People can’t generally access it. It is only the railway industry people who are allowed there. And so it’s
relatively undisturbed. And so if we get a match between having declining species, a habitat type that
works for a particular location next to the railway, then the opportunity for us is actually to remove the
vegetation that always causes us problems and to create that structure that is a good thing for those
rare or declining species. And then what will you then end up with is a corridor that’s connecting
communities of those round declining species and gives them the opportunity to start moving around.
As climate changes, they maybe need to be moving further north and is actually using literally the transport network as they transport to enable them to start moving around. But you then are linking up
using that habitat and the landscape data that we’ve got as well. There might be communities just
outside the fence but then can use our estate as a stepping stone to get to somewhere else. You end up
with that sort of bigger, better, more joint up and that’s the opportunity that railway has.

Amjad Karim (20:34):
So listening to what both of you’re saying, one of the things I was thinking about is how far could we
take this? What’s the kind of vision for the future? One of the things that sounds really cool about what
could happen here is if you had a system of moth traps or sensors across the country or almost globally,
is it possible to build a, not the real time, but like a vision, a view of how habitats are evolving and
changing at a very fine timescale? How useful would that be? Is that possible? And then how useful is
that? I’m just thinking off the top of my head.

Tom August (21:04):
Important just to say straight away that they will continue to be important right, is the actual surveys.
There are lots of things that boots on the ground can do that automated methods. They can give that
holistic view of the biodiversity there, they can assess the context, they can spot outliers, anomalies and
things like that. So there’ll always be a place for that. I guess I see these technologies as complimenting
and enhancing what people already do. So for example, when you’re going to do a breeding bird survey,
you can get a pretty good idea of the abundance of individuals and where in the landscape they are. But
with an AI method you’ll just get a sort of presence, absence species list. That might be fine for
answering some questions. But in terms of this kind of national or international network, we already do
analysis at the national scale for the state of nature. So how is nature in the round or specific habitats,
how are they fairing, how are changing at the national or international scale? And those rely on a mix of
data sets. Some of them will be standardised methods where people go and do a transect, at a certain
place once a year or twice a year and we can see how that changes over time. Some of that is done from
just ad hoc observations. People happen to be out on the weekend and spot something and they
reported, I think these automated methods are a new stream of data.

(22:41):
There are some countries, so Netherlands for example, have had these camera systems out for
monitoring pollinators and night flying insects for a couple of years now. And it wouldn’t surprise me if
we see a number of networks started to establish other countries which will be running for decades
collecting these sorts of data. So the challenge there for us as research is how do we bring these data
together. We don’t want to start from this year and then measure how things change, you want to
integrate that data that we’re collecting from this year onwards from the automated methods, with all
that data that’s gone before so that we can create a measure of change over a very long time period. So
that’s kind of one of the challenges we have have to face.

Amjad Karim (23:27):
I had this, well not this idea, but just visionary. You could have a real time view of what’s going on. So
the reports that you talked about come maybe on an annual basis or however. They aggregate lots of
data, they make sure it’s all coherent and then published for report. So are you sceptical about using it
as a real-time measuring tool, or do you think there’s problems with it because things are quite volatile
and you can’t learn a lot? Is that something you’d like to…

Tom August (23:55):
Definitely. Yeah, definitely these sorts of systems, same applies to satellites, right? We’re getting much
faster repeat visits to satellites. You start to produce that much higher frequency and like you say, these
systems, whether it’s a moth trap or acoustic things for birds and bats can produce data in near real
time. I think what’s important is don’t just do something because you can. So why would you need that
data in real time? And for a lot of the stuff we do, we simply don’t need it in real time. We’re interested
in seasonal patterns or year to year patterns. So it’s not required. However, there are applications where
it is useful. So for example, in terms of engaging the public, so we have a field site here just in
Oxfordshire, which we run with the Earth Trust, which is a charity, and there they have one of these
monitoring systems that’s out on this nice wetland habitat they created.

(24:49):
And we’re looking at ways that we could get that data in real time onto their websites, so that people
could see what birds have been heard this morning and maybe you see a really cool bird on this and you
decide you’re going to go out and see if you can twitch it. So, there are times when real time is useful.
Another thing to bear in mind is when you move to real time, you’re I guess, in most applications in real
time you are then removing the human from the loop. So you’d be taking the AI identification of things.
AI classifications of habitats for satellite, the AI classifications of birds and that song and you putting that
straight into use in whatever that real-time system is. Whereas when we do stuff at an annual scale, we
typically have a human go through and review and do some quality checks and that sort of thing before
we refuse it.

Amjad Karim (25:35):
One thing. So when you watch springwatch and things like that, they have a camera in a bird box. And I
thought for network rail, if you had a system where you’ve got all these monitoring things, but you have
a ways for people to access that and see, okay, this is what’s growing at what’s happening right now at
New Street Station, which probably isn’t a lot given where it is. There are stations that aren’t in the
central, that are probably more pleasant. I think that would be a nice way of engaging. One of the things
that you guys have both talked about is engaging the communities in which you operate, in the
communities in which collect that data and also the communities that analyse that data and
communities that actually act upon the data that’s been collected. So there’s a number of communities
involved.

Tom August (26:19):
I think this is a really interesting discussion point around engaging people with nature. You and I think
maybe all of us here see an opportunity here. We’re collecting this data via digital technologies and we
can tell people about what it’s hearing. That is exciting and I think that’s true. That is exciting. However,
if you ask someone to recall a interaction they’ve had with nature, which made them feel good, it won’t
be digital. It’ll be a childhood memory from being in the woodlands with their family and hearing the
birds sing or it’ll be on holiday snorkelling and seeing something cool. So we need to kind of, I think, just
sort of check ourselves, in how we want to help people engage in nature. So here’s an opportunity,
we’re collecting some digital stuff, we can share that, that’s fine, but how else can we use these
techniques to really help build that better connection with nature? And this is actually a bit of pushback
I often get when talking about this subject, is that these digital approaches to recording nature act as a
wedge between people and nature. So rather than looking at the butterfly in your garden and
appreciating the beauty and wonder of it and thinking about the pollination it’s doing and how amazing
that is, you get out your phone, you put a screen in between yourself and that butterfly and that
butterfly becomes a data point or it puts you up to the league table in the app you’re using. And so it’s getting in the way of that connectedness with nature, which we know is a problem. And we know that
spending time in nature is good for mental wellbeing. So one of the things I think is helping that
connection with nature is that I think it’s important to be able to put names to things and to know about
species and that increases your connection to it. So, I find that if I learn a bird call then when I’m out and
about, instead of just being noise, it’s like it’s that bird. I can now identify it and I can appreciate that
more. So I think using these AI methods to help people to understand a bit more about nature, that’s
random, and through these apps you can add extra information about this is a rare species or this is a
foreign species or the species is a specialist on roses or whatever. You can enrich these interactions with
nature.

Amjad Karim (28:39):
Are talking about slightly different things though? So one is, being there and recording, so yeah, I get
this sense that I get it with lots of people when I see them not looking at the fireworks and they’re
recording. So I get that. But I was thinking more about if you’re not there, so if you’re not at the station
or if you’re not in the habitat, you get a way of seeing or engaging with it. So you’re right, I guess it can
go either way. One is, when you’re not actually present, how you can use digital technologies to make
people closer to it. And the other one perhaps is when you’re actually present maybe the digital
technologies that you’re speaking about to act as a barrier.

Tom August (29:16):
So they’re not exclusive. I think what I was trying to say is that we need to see where the opportunities
are to also use these technologies in a way to connect people better with nature. And when we’re
talking about we’ve got this census out and we’re collecting data, we want to share it and make it
engaging, I think there’s also work to be done in thinking about what is it that’s engaging. So we could
put up the species list of birds for example, that’s been recorded every morning an outside Houston
station or whatever…

Neil Strong (29:43):
That so, species what was here yesterday and the big long list that ranges from Ostriches and Golden
Eagles to the Dunnocks and the House Sparrows. And when you are there looking through the bit and
the hide, there’s nothing. Absolutely. It’s trying to show people what they could have, you should have
just been here yesterday.

Amjad Karim (30:03):
Here’s why you could have won.

Tom August (30:04):
Yeah, but I also think it’s adding to that because still that list of bird species is only really accessible to
birders right, who know what those things are and why they might be interesting. So I think adding
information like this is a rarity or this is a migrant that’s just arrived from Africa. These sorts of things
that start to build that story and enrich that experience, I think is also really important.

Neil Strong (30:27):
And from our side of things, there’s two bits of engagement we have to do. There’s the one with the
people on the other side of the fence, so that they can see what we’re doing is good land management and there’s a right reason for it. And this is where that narrative comes in that we are protecting this
particular species and this one’s being able to expand. But then there’s also giving the data in a format
to my colleagues in the railway, so that they can then make those management decisions. Cause this has
to be a long-term thing. That the habitats we’re creating are not, you’re not going to turn it from a
woodland to a grassland in one season. And it’s giving them the right data to enable them to make the
right decision about what to do with the habitat. So it’s slightly different bits of communication using the
same data in a slightly different format.

Amjad Karim (31:16):
So when you’re talking about communication here, you’re talking about communicating to people
within say Network Rail who are responsible for vegetation management. They have specific needs,
don’t they in terms of what they need and how that needs to be presented to them. I think when I was
talking about the general public, I agree perhaps knowing what was here yesterday, isn’t that well
maybe that exciting. I do occasionally see them, but I don’t really understand them. But also if you have
digital ways of sharing, you can’t iterate quickly though, right? That’s the other thing. So you don’t have
to figure out everything ahead of time. You can try new things and see can be what can be shared and
what engages people and what engages them to do perhaps some of the things that you’d like them to
do perhaps. I kind of suggested this kind of thing, which is like this remote monitoring system. What are
you guys thinking about, that you’re going to be doing over the next few years or months?

Neil Strong (32:07):
As I said, railway, we want to make sure that we’re not putting people where they don’t need to be. So
the remote monitoring is something so we have to be able to justify and demonstrate why we need to
do a piece of work. We might need licences from one of the environmental organisations to be able to
go and do something. So being able to collect data about the particular species without having to go
there, that’s where the remote stations come into the road. And I think I mentioned earlier the ability to
see what impact we are having. We know we need to do this work, but there’s not a lot of work being
done on, so how is it changing and what is changing as a result? So being able to have that remote
station that sits in the same place whilst a whole load of other stuff goes on around it, will give us a
really useful data set in demonstrating what is happening to the habitats and therefore the associated
species and everything else that will come in and live there. How that changes so that we can then start
that narrative and demonstrate the progression from where we are now to where we want to get to in
the future.

Amjad Karim (33:14):
So either of you have a sense of how many of these devices we would need or how would you place
them or what’s a strategy that you’d follow for best getting the data that you feel is important. So it
doesn’t necessarily mean you need to put it everywhere. You might put it in strategic places or areas
that are more representative or less representative. Have you thought about that at all in terms of how
that might work?

Neil Strong (33:34):
From our side of things, it would be literally based on where the work is required. So we know we’re
going to certain places, there’s a level of work we need to get to before we have the complete, the right
structure across the whole network. So we’ll go to where we need to go, but Tom might have some
better, more scientific reasons for putting it in certain locations.

Tom August (33:53):
It’s totally question dependent, isn’t it, Neil? I mean, if you are interested in the impact of management
you’re doing, or if you want it at the sites that you’re doing your management on. I think from a
research point of view, if you are interested in the general state of biodiversity in landscape and how it’s
changing, then you’d kind of want to, we say stratify that sampling. So you’ve got sampling points and all
the relevant habitats that you’re interested in and you have enough that you would be able to detect
the change if a change was present. Obviously when you’re thinking about deployments at scale, then
cost becomes an important factor and some of these things are quite expensive. One of the cheapest
elements at the moment is the acoustics. So you can get these little units which can do some pretty
good acoustics and you can put out a large scale, but things like the moth trap, again fairly, fairly big, not
so mobile. So they’re better for sitting in kind of one place or moving every week or so to a new
location.

(34:54):
I wanted to come back in terms of, Amjad your question, about the future directions, and Neil you
mentioned about sharing data. So the global scale, one of the challenges we’re dealing with, with these
autonomous systems is what happens to all of that data? So we’ve got these hardware now, acoustics
and the moths trap, and there’s other designs around the world which are collecting data. They’re
collecting very large volumes of data. We’ve got some AI tools developed, which can variously classify,
analyse these things and produce data, but what do we do with all that? Where does it go and how do
we make sure that it’s useful to people? And it’s in a standard that you can take that dataset from it’s
being collected in Argentina and that data’s been collected in Israel and be able to compare them. So
yeah, this is a big challenge at the global scale and I think that’s something that we’ll be working on as a
community over the next two or three years, will be how do you store it first of all, and then what are
the standards that you put on that data to allow it to be used by as many people as possible and be as
useful as possible.

Amjad Karim (36:04):
I get that, but I also wonder. When The Centre for Ecology and Hydrology started with data sets that go
back over a hundred years. So people started recording something. Sometimes you just record
something and you don’t know what the value is of it over time, but actually when it becomes valuable
is when you have it for a long period of time. And when you have that, then you learn ways that you
could use that data. I think storing the data, holding that data, all of those things are technical
challenges, but I wonder sometimes whether we don’t know right now how we might use that data. It
will evolve on time and you might see an opportunity maybe five or six years down the line. Sometimes I
hear it quite a lot when we talk about, oh, we need to structure the data in a certain way for future use
cases or whatever it might be. But oftentimes you don’t know what they are. So you want to try and
preserve as much of that data so that you can ask as much questions of that data in the future rather
than you’ve tried to simplify it and compress it and you’ve lost something that you perhaps might find
useful in the future.

Tom August (37:05):
No, that’s absolutely right. And one of the things that’s important is the context. So was your sensor in a
woodland or grassland or what model of microphone did you use? Maybe in five years time we’ll find
out that model of microphone was terrible for picking up Blue Tits or whatever. So all these, like you say,
you’ve got to retain this information, which maybe right now you don’t think it’s useful, but actually it
might become crucial in years to come.

Amjad Karim (37:33):
Okay, thanks guys. Is there anything that you’d want people to go to, and links or locations or places or
things that you’d like people to read that you’d like to share? We can share it afterwards as well. We can
put it on the notes to do with this particular podcast. But is there anything that you’d like people to go
to or tell people about that they should look into? From both sides?

Tom August (37:53):
I’d encourage people who are interested in nature and spend time in nature to consider recording what
they see. We’ve recently been working with some artists actually on biological recording, and one of the
themes that has been picked up by them, it’s a painter and a poet, is this idea that wildlife doesn’t have
agency unless it’s recorded by humans. So we can’t protect species that we don’t know need protection
because they don’t have a voice for themselves. So people who are out spending time in nature and
interested in nature can be that voice of nature by using smartphone apps and things like that to record
what they see. And the biological record centre has a whole range of mobile phone applications that can
be used. And yeah, we can share that with your materials and link to those pages.

Amjad Karim (38:43):
And Neil?

Neil Strong (38:44):
Just get out. Basically. I like Tom’s points about getting the screen in the way of the thing. I’ve done it
myself and I’m trying to do it less often. So there is things to look at without needing to take a photo of
it. So just get out of there and record it, write it down, get a pen, a piece of paper, the old fashioned
way.

Tom August (39:02):
Can I ask you a question Amjad?

Amjad Karim (39:03):
Yes, go ahead.

Tom August (39:04):
So you are obviously an AI expert and that’s what your company does. Where do you see the future of
AI going in relation to this kind of biodiversity monitoring? Do you see new technologies techniques on
the high line, on the horizon, which you think will impact biodiversity? And I guess from the AI side of
things, what do you see as the kind of key next steps?

Amjad Karim (39:31):
From an AI side, I think for me, what’s really, I find really interesting and I think is really valuable is large
scale monitoring that can take place across a large area with very, very cheap sensors or satellite data.
That’s really important. So how do we get as much of the environment monitored as possible, as
cheaply as possible, and in a way that’s least or zero damaging to the environment? Because I’ve looked
at other technologies where there’s almost like things that you can drop off a drone or off an aeroplane
that you can distribute sensors across places, but they have an environmental impact. So how can we
create technologies that we can deploy across the globe to measure and monitor what’s going on? Secondly, I think, well, we need to make those data sets open source and available to people to do the
analysis and the models that fall off of them, make them open source.

(40:24):
And the second thing is maybe not a technical question, but perhaps with using AI and using digital
techniques and having a more of a way of monitoring the environment or the world in a more real time
manner, we can try to understand the impact of actions people take that have no official commercial
impact, so there’s no cost associated. And that’s one of the things that we need to get around is how do
you incorporate the benefits of increasing biodiversity in terms of financially and how do you
incorporate the costs of doing bad action? So recently in the UK there’s been a lot of controversy about
how sewage is being released into the water because perhaps we don’t have enough capacity. I don’t
know exactly what the reasons are for that, but if we were monitoring that, if we had remote sensing
technologies that were available to the public first to see that that’s happening. A, we would probably
have, we would’ve picked it up sooner. And secondly, it gives us a mechanism of trying to understand
how you would measure the economic cost of that and then charge the actors for that. So I think the
biggest problem that we have today globally, is the tragedy of the commons. So the important things to
us as a society are not costly. I mean, as a result, people can damage it. So if we can use technologies to
build a better overview or a better kind of global view of what’s going on, maybe we can start doing
that. Maybe costing things isn’t the right way of doing it, but it seems a mechanism of doing it right. At
least we start to understand the value of nature and I think these technologies will help with that. That
would be my wish.