Same struggles, different year


Very belated Happy New Year! I’ve been meaning to write another post for a while, but I’ve been having problems with work and various other things which have prevented me from posting for a while. But actually, that is the BEST time to write a post. Science is full of ups and downs and I promised myself when I started doing this that I would write about the downs too. So here we go:

The crows (a relentless tale of horror)
As part of my work on soundscapes, I’ve been finalising automatic species detection models for several species: most recently, the endangered and endemic Okinawa Rail, and the ubiquitous Jungle Crow and Brown-eared Bulbul. This should be a fairly straightforward procedure whereby I run a first pass of the data and tell the model which clusters of vocalisations I’m interested in. After that, I re-run the model on the same training data and it should isolate only the sounds of interest, but often with some additional unwanted noise. After I then manually identify those noises that aren’t of interest, the model should be able to extrapolate to larger data inputs with relatively high accuracy. Should, being the key word here…

I’ve had a few problems with two of these species; specifically, the rarity of the Okinawa rail makes it hard to train a model that produces any hits in the second round, and the Brown-eared bulbul has several distinctive calls, which increases the amount of noise the models produce since they’re trying to multiple sounds to one species. After some frustration and playing around with input data and species vs noise identification, I eventually produced models that seem to be of publication standard. The Jungle crow, however, was a different story.

We have five locations that we are recording these species. In four of those locations, I succeeded in producing a crow recogniser with up to 99.3% accuracy in one case! However, in one of our sites which has a mix of urban and rural sounds, the model struggled. A lot. The model was identifying roosters, kingfishers, bush warblers, owls, pigeons, cicadas, crickets, construction sounds, people laughing, dogs barking and genuine crows as crows, producing models with <30% accuracy. I probably spent around 120 hours working on the crow detector, and nothing was working. After talking it through with my collaborators, we decided it might be something to do with the soundscape signature of this one site – the rooster was causing most problems, but a similar site which also has roosters did not have this problem. So, after several weeks and weekends of working until 11pm, I decided enough was enough. I re-ran the best model we had at the time and manually went through identifying false positives from 11,000+ hits (mostly roosters). We’ve identified some possible reasons our model was struggling so much at this one site and will discuss them when we write the paper. On a Thursday night at 10.20pm, I finished listening to birds once and for all (well, until I start the next project that is), and I even got to spend some time in the field this week.

On the plus side, if President Trump is serious about this whole torture thing, I’ve got a suggestion that both contributes to science and guarantees that any offending party is spilling the beans in no time…

Jungle Crow (Corvus macrorhynchos), a.k.a the bane of my existence.

More uncertainty
The PhD funding application I submitted back in November was rejected yesterday. Obviously, I’m disappointed and a little sad, especially because the project was exactly what I want to do, but I’m trying not to call it quits just yet. I’m going to try and find some way to still do that project, whether it’s through alternative funding or some other means. It’s just a bit disheartening to be back where I was six months ago. I’m not a huge fan of uncertainty, especially about the bigger things like this. But, as I keep telling myself, it’s not back to square one; more like square 1.5.

Silver linings?
The crows are over. In fact, all the birds are over now. So, the fun part can begin. Answering the questions we set out to answer several months ago, and writing the paper (I’ll be co-first author, which I’m excited about!). My functional diversity manuscript is also about to resubmitted with minor edits after a second redraft, so fingers crossed still that the paper will be out soon. I’m hoping to give a seminar on that work here at OIST before I leave, and I’ll be visiting the National Centre for Biological Sciences in India soon to give the same seminar there too. I also had a really nice Christmas and New year when my Mum came to visit, and I’ve spent a good amount of time on the mainland and in my favourite city (Tokyo) recently; I even got a Japanese tattoo, which I’m really pleased with. So it’s not all bad, at least.

This slideshow requires JavaScript.

Science Wins:

1. Manuscript advances – Continued progress on two of my manuscripts
2. Soundscape progress – Ready for data analysis of our first soundscape ecology project

Sciences Losses:

1. Crows, crows, crows – Spending too much time in the lab, and listening to crows. Well, roosters masquerading as crows.
2. PhD uncertainty – Funding application rejected for my PhD.

What’s next?
Finishing work on several manuscripts, trying to sort out some PhD funding, and analysing data from the first soundscape ecology project. I’ll also start making progress with the second soundscape ecology progress, which I’m taking the lead on; I’ll be learning how to do phylogenetic analysis to determine evolutionary histories and relatedness which I’m excited about!

Until next time,



One thought on “Same struggles, different year

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s