Last Week

Habitat mapping with Tessera

Last week I spent most the week down in Cambridge, primarily for a group meeting that David Coomes' group were having to review what Tessera related habitat mapping projects they have going. Given I am also starting to get my hands dirty on this topic, I was grateful that David had included me on the invite. As someone getting up to speed on this domain it's invaluable to see how other people are approaching this domain and what is of interest.

As prep for the meeting I managed to do a little end-to-end run of a naive pipeline that takes a set of species observations from GBIF for a given country or subregion, looks up habitat preferences for them, and then uses Tessera to build a crude habitat map. Right now the habitat maps I'm making are clearly very broken, but I'm not to worried about that right now, my main aim was to try and work out if I had any gaps in my understanding of how to pull together the bits required to make any sort of map. Now I have that, I can work in actually getting good data into the pipeline to improve the quality of map produced. My main learning from this exercise is that some of the trends in the data didn't match my expectations, even allowing for my low expectations. And so whilst I didn't assume this task would be easy, it's already more challenging that I thought :)

I consider realising this a win - it's very easy in the large data-science projects to assume the data going into it looks a certain way, and then after being combined with many other bits of data and with it all being so voluminous and then finally aggregated down, it's hard to spot that you have a problem of this form.

For example, due to how large Tessera is (it's a 10m per pixel global model), I've been working with just the UK and Madagascar initially, as smaller areas where we have some experience of what the data should look like. Even so, it's easy to assume the form of the data, and so I've been building the pipeline from the start to spit out summary stats for me in the form of excel spreadsheets, so I can quickly make graphs to see distributions/trends. Here's a fun example, which highlights some of the challenges with using citizen science data. I took a set of species occurrence data within the UK for a set of 182 animals I was interested in, and then mapped those locations to Open Street Map data and pulled the tag used to describe the land usage at that point:

The chart removes 34603 occurrences where OSM had no land usage tag, leaving us with those that actually had a label. From this you can start to spot the perhaps somewhat predictable biases in the sample data. Most observations are "residential" for instance, despite that not being the primary land usage type for the UK, and more species listing forest as their habitat preference over urban. I was also amused to see how high "allotment" ranks - it ranks even higher when I was looking at data for specialist species with a single habitat preference (none of which listed human made habitats as their preference). Ultimately I'm dealing with two biases here: there's a bias in where the occurrence data is collected, and where people contributing to OSM focus.

None of this is particularly surprising: if you hang out with ecology data-scientists, discussions about the inherent biases in these data sources is well known, it was just interesting to try and make that bias obvious in my pipeline so I can make it a conscious decision to work with this data, rather than something unacknowledged. In building a new pipeline for my habitat map work, the first totally new pipeline I've written for a couple of years, I want to try and make sure that biases and trends in the data are made evident as early as I can, so that we as a team can understand these issues before they get lost as we combine them with other data sources and then start to build our highly aggregated results.

Conversations galore

Outside the meeting with David's group, I spent the remaining time in Cambridge working out the common room in the David Attenborough Building, where I could alternative hacking on my laptop with conversations with colleagues. Lots of good conversations (in rough chronological order):

• I met Vinicius Tonetti for the first time in person, after many years talking about AOH edge effects via zoom. He's now based in Cambridge for the next year or so, and so we chatted about how we can perhaps push forward some edge effects work, and he is going to help find some candidate areas of Brazil for testing in the aforementioned habitat work
• Caught up with Shane Weisz on his PhD work on building tools to help with Red List assessments which gave me some ideas on how to automate some of the data sperlunking I've been doing into the red list of late.
• On the topic of the Red List I bumped into Simon Tarr of the IUCN and we had a brief chat about what is going on with some of the bits we collaborated on last year.
• Caught up with Alison Eyres and Aneesh Naik on the habitat mapping work we're trying to do together.
• Chatted with Sadiq Jaffer on Tessera related topics, and got a cool demo of his work to compress Tessera down to make it usable on smaller computers.
• Went through some of the updated LIFE pipeline with Emilio Luz-Ricca, one of Andrew Balmford's PhD students.
• Finally had a catchup with Anil Madhavapeddy on future directions for our work, and learned about some interesting work from the Cambridge Evidence Tap.

Basically between this and the workshop I filled my brain up with ideas and enthusiasm, and now I have a month to try action some small portion of it before I return to Cambridge next!

Lacking Porpoise

Whilst I've spent five years or so in the field of nature conservation, as a computer operator I've not had any chance to do any field work. So I was excited to get a chance to help cetacean (porpoises, dolphins and whales) monitoring with the local group of the Sea Watch Foundation up here in Merseyside, as part of the foundation's National Whale and Dolphin Watch. Alas, there were strong winds, meaning a very choppy sea, and so after an hour and a half we didn't see any of the typical harbour porpoises or bottlenose dolphins that they typically get in this area, but we did see a good number of Cormorants and got to watch the MSC Santa Maria container ship setting off from Liverpool to Antwerp. Next time hopefully the weather is going to be more favourable :)

Books

Again, despite working in conservation for a longish time, the majority of things I focussed on over that time have involved terrestrial vertebrates, which is to say land animals (birds, mammals, reptiles, and amphibians). I'm still far from a zoologist, but improving my layperson's understanding of the area has been one of the things I've enjoyed about my job. But with getting into habitat mapping I'm now having to interact more with plant scientists, so I thought I should try to begin to understand a bit more about that side of things. To that end, for the drive to and from Cambridge, I listened to The Hidden Life of Trees by Peter Wohlleben, managing to complete about 2/3rds of it. It's not a technical book, rather a pop-science book, but it has been interesting learning about the importance of old growth versus new growth forests, and why they're different; learning about how beech trees can usurp oak trees; and just getting a sense of the very different time scales on which trees operate.

The closest it touched on work I have done in the past was a discussion on forest fragmentation, which I looked into a little for edge effects on area of habitat maps I did with Ali and Vinicius. In that work we were focussed on the fact that land animals won't fully occupy their habitat, avoiding the edges where it sits next to habitats they don't like, and so if you have very small patches of forest, then there might be no habitable area for the animals. The book also talks about how small forest patches are also problematic for trees due to lack of breeding diversity.

Anyway, baby steps in trying to improve my plant knowledge, but across a lot of the bits of ecology I interact with they're pushing on moving to include plants within methods that have thus far been mostly focussed on animals, and so I look forward to learning more about this side of things in the coming years as I collaborate with experts on this topic.

This week

• I have to get some data ready for Ali for a talk on Wednesday, so that's next after these notes!
• Even though I think my current approach to the habitat map work with Tessera isn't going to pan out, I want to do the next obvious refinement as part of limbering up how I'm going to work with Tessera, as it'll help inform alternative approaches and again give me some more experience with the datasets I'm using.