Just short weeknotes this week as I've been travelling back to Belfast for various matters and haven't had much computer time.
One exciting development in the week was that Shane Weisz continued the conversation with the IUCN Red list team about his developing dashboard, which went extremely well. There's so much excitement on both sides about how all this is going!
1 TESSERA developments
I also spent a chunk of time wrestling with understanding Zarr so I can port TESSERA to use this instead of Numpy arrays. There's been a long and helpful thread on our Zulip about this with a lot of people chiming in. Michael Dales has also been a source of coordinate inspiration from his experiences, so I'll put proper thoughts together on this soon.
On the OxCaml front, Mark Elvers has knocked up some OCaml Zarr so I'll be porting those to OxCaml soon and taking the TESSERA support in OCaml for a spin.
1.1 TESSERA activity around the web
There was quite a lot of TESSERA things going on alongside this.
First a preprint Earth Embeddings as Products: Taxonomy, Ecosystem, and Standardized Access that's a nice summary of how to use geoembeddings like TESSERA:
Geospatial Foundation Models (GFMs) provide powerful representations, but high compute costs hinder their widespread use. Pre-computed embedding data products offer a practical "frozen" alternative, yet they currently exist in a fragmented ecosystem of incompatible formats and resolutions. This lack of standardization creates an engineering bottleneck that prevents meaningful model comparison and reproducibility. We formalize this landscape through a three-layer taxonomy: Data, Tools, and Value. We survey existing products to identify interoperability barriers. To bridge this gap, we extend TorchGeo with a unified API that standardizes the loading and querying of diverse embedding products. By treating embeddings as first-class geospatial datasets, we decouple downstream analysis from model-specific engineering, providing a roadmap for more transparent and accessible Earth observation workflows.
Then Sadiq Jaffer and Frank Feng did a great podcast on Satellite Image Deep learning where they go through the journey of how we trained the model. I can't believe it's barely been a year!
TESSERA support also got merged into Torchgeo for those looking to work on customising the model itself. Most users don't have to use this as they can just use our pregenerated embeddings.
2 A fond farewell to dra27 from the Lab
After almost a decade of working with David Allsopp in both the University and later Tarides, he finally "graduated" and went off to join Jane Street where...we will continue to work together on OxCaml and OCaml.
Good luck to David as he no doubt enjoys the ridiculously nice Jane Street office, where I would overdose on the fresh fruit juice machine and be on a perpetual sugar high!
Some fun links:
- It was nice to see others getting excited about my OCaml ATProto client support
- And also WebFinger seems more important so I implemented that too.