TESSERA: Temporal Embeddings of Surface Spectra for Earth Representation and Analysis · HackerPapers discussions

TESSERA: Temporal Embeddings of Surface Spectra for Earth Representation and Analysis — mentioned by sadiq

Mon, 02 Nov 2026 16:03:23 GMT

University of Cambridge, University of Manchester | PhD in AI / Earth Observation / Environmental Science | ONSITE | UK home fee status only | £30k tax-free annual stipend Our group at Cambridge works on Earth Observation foundation models (Tessera: v1 https://arxiv.org/abs/2506.20380, v2 https://arxiv.org/abs/2607.03949, https://geotessera.org/) and we've started a collaboration with Manchester to understand how they can be used to improve pollution models and then environmental decision making in general. This involves training new models and doing high-resolution inference across national…

Paper and discussion history. A mention is not an endorsement.

TESSERA: Temporal Embeddings of Surface Spectra for Earth Representation and Analysis — mentioned by sadiq

Fri, 26 Sep 2025 08:30:19 GMT

It's possible to use embeddings as input to a convolutional network and then train that using labels. We've done that for at least one of the downstream tasks in the TESSERA paper: https://arxiv.org/abs/2506.20380 to estimate canopy height. The downside of that approach is that you need to spend valuable labels on learning the spatial feature extraction during training. To fix that we're working on building some pre-trained spatial feature extractors that you should only need to minimally fine-tune.

Paper and discussion history. A mention is not an endorsement.

TESSERA: Temporal Embeddings of Surface Spectra for Earth Representation and Analysis — mentioned by sadiq

Thu, 25 Sep 2025 20:38:04 GMT

Hi! You can find a bit more about Gabriel's model through some of his posts over the last few weeks: https://gabrielmahler.org/posts/ When it comes to the satellite images, the model actually used TESSERA (https://arxiv.org/abs/2506.20380) which is a model we trained to produce embeddings for every point on earth that encodes the temporal-spectral properties over a year. Think of it like a compression of potentially fifty or a hundred observations of a particular point in earth down to a single 128 dimension vector. Happy to answer any other questions.

Paper and discussion history. A mention is not an endorsement.