Search papers, labs, and topics across Lattice.
This study conducts the largest controlled scaling analysis of pixel-wise Earth-observation foundation models, involving 395 training runs across 1,024 GH200 superchips. The findings reveal that pretraining loss is a poor predictor of downstream performance, suggesting that traditional model selection methods are inefficient and wasteful of computational resources. By establishing a guideline for scaling鈥攚here encoder and data size increase together while keeping the projector fixed鈥攖he authors successfully train and distill models that outperform larger competitors, demonstrating a new approach to efficient model deployment in Earth observation tasks.
Pretraining loss misleads model selection, wasting compute, while a simple scaling rule unlocks superior performance in pixel-wise Earth-observation models.
Pixel-wise Earth-observation (EO) foundation models are now achieving state-of-the-art performance via generated spatial embeddings. However, how these models scale and how best to spend a pretraining budget remain poorly understood. We present the largest controlled scaling study for EO to date: 395 training runs on 1,024 GH200 superchips within a fixed pixel-wise Barlow Twins family, each evaluated on 15 downstream tasks. We find that pretraining loss barely predicts downstream performance (|Pearson r|<0.2), so selecting models by loss wastes a large share of the compute. We also find that, as the training budget grows, the encoder and the data should grow together while the projector stays fixed, which gives a simple rule for allocating compute. Using this rule, we train a family of pixel-wise models (0.5B and 1B, with a 2B model in training) and distill them into compact students for embeddings-as-data deployment. The 21-million-parameter distilled TESSERA v2-1B-M in aggregate outperforms all open and proprietary models tested, some of which are orders of magnitude larger. These students produce Matryoshka representations that are inexpensive to serve: a 16-dimensional prefix keeps 92% of the full 128-dimensional performance at 1/8 of the storage. Upon completion of training we plan to release v2 global embeddings covering 2017-2025. Together, these results give a concrete, empirically grounded recipe for scaling pixel-wise EO foundation models: train large encoders, select by downstream performance, and distil into flexible student models. All code will be released at https://github.com/ucam-eo/tessera.
