Climate

NIVA: A Multimodal Foundation Model for Actionable Earth System Intelligence

Anisha Pal, Aodhan Sweeney, Kyle Heyblom, Kalai Ramea

Featured July 12, 2026

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

A new AI model called NIVA learns how the ocean and atmosphere interact by looking at them together, helping predict climate changes better than current models.

In depth

NIVA is a multimodal foundation model that learns unified representations of coupled Earth system dynamics, specifically focusing on ocean and atmosphere interactions. It uses contrastive learning to align representations from separate encoders for each modality in a shared latent space, enabling improved subseasonal-to-seasonal prediction by capturing long-term memory.

Key Takeaways

1
Introduces NIVA, a multimodal foundation model for learning coupled Earth system dynamics across heterogeneous components.
2
Leverages contrastive learning with separate encoders (Spherical Fourier Neural Operators) for ocean and atmosphere to align representations in a shared latent space.
3
Demonstrates improved prediction of major climate indices, showing the model captures physically meaningful cross-modal relationships for subseasonal-to-seasonal predictability.

Conceptual Flow

HIGH LEVEL

Methodology: Learning Coupled Earth Dynamics

The model takes ocean and atmosphere data, processes them separately, and then learns to connect them in a shared understanding space.

Ocean Data

Atmosphere Data

Learn Connections

Shared Understanding

Results: Predicting Climate Patterns

By understanding ocean and atmosphere connections, the model can accurately predict big climate patterns like El Niño.

Shared Understanding

Predict Climate Events

Climate Index Values

This breakdown was generated by SciGrove. Get the same analysis — intuition, storyboard, peer review, a runnable prototype and a glossary — on any paper you upload or paste a DOI for.