Back to AI News
arxiv

DARWIN: a Geospatial Foundation Model for Marine Intelligence

DARWIN is a geospatial foundation model adapted for marine intelligence, addressing missingness, multi-modal physical and biogeochemical variables, sparse labels, and non-stationary ocean dynamics. It adapts a masked-autoencoder backbone with missingness-aware encoding, expanded spatial and temporal masking, and Fourier-style positional embeddings, pretrained on 26 years of global Copernicus reanalysis plus GEBCO bathymetry. On chlorophyll-a prediction the pretrained backbone achieves the lowest monthly relative error in every month of 2022, beating no-encoder baselines and an identically architected random-init encoder, and at 10% of supervised data it surpasses the random-init model trained on 100%. Embedding analysis and qualitative reconstructions suggest it captures coherent oceanographic structure, supporting downstream uses like anomaly detection, species distribution modelling, and environmental monitoring.

AI Brief

It shows foundation-model pretraining transferring to ocean science, where Earth-observation models have largely stayed terrestrial, and quantifies the label-efficiency gain.

arXiv:2611.22496v1 Announce Type: new Abstract: We present DARWIN, a geospatial foundation model for marine intelligence. Recent Earth-observation foundation models have concentrated on terrestrial domains. The marine realm poses distinct challenges: missingness, multi-modal physical and biogeochemical variables, sparse labels, and dynamics that are transient, multi-scale, and non stationary over time. DARWIN adapts a masked-autoencoder backbone to these properties through missingness-aware encoding, expanded spatial and temporal masking, and Fourier-style positional embeddings, and is pretrained on 26 years of global Copernicus reanalysis together with GEBCO bathymetry. On predicting chlorophyll-a, the pretrained backbone attains the lowest monthly relative error in every month of 2022, outperforming both no-encoder baselines and an identically architected random-init encoder. At 10% of the supervised training data, DARWIN already surpasses the random-init model trained on 100% of the data. Embedding analysis and qualitative reconstructions show that DARWIN captures coherent oceanographic structure rather than memorising input statistics, providing a scalable backbone for downstream marine applications such as anomaly detection, species distribution modelling, and environmental monitoring.
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!