NIM for Earth-2 CorrDiff

Type: Microservice Tags: NVIDIA, NIM, Earth-2, CorrDiff, weather, climate AI, downscaling, super-resolution, diffusion model Related: Earth-2, NIM-for-Earth-2-FourCastNet, PhysicsNeMo, NVIDIA-Modulus, NVIDIA-NIM, NVIDIA-AI-Enterprise, NVIDIA-Omniverse, cuDNN, PyTorch Sources: https://docs.nvidia.com/nim/earth-2/corrdiff/latest/overview.html, https://docs.nvidia.com/nim/earth-2/corrdiff/latest/index.html Last Updated: 2026-04-29

Summary

NVIDIA NIM for Earth-2 CorrDiff packages NVIDIA’s Correction Diffusion weather downscaling model as a deployable NIM microservice. Current docs describe CorrDiff as a neural network model that improves weather-data accuracy and resolution by correcting a mean machine-learning model with a diffusion model.

Detail

Purpose

Weather and climate workflows often need high-resolution local fields from lower-resolution forecasts. CorrDiff NIM provides a consistent API and deployment container for producing higher-resolution, probabilistic weather outputs as part of the Earth-2 services and software stack.

Current scope

  • Downscales surface and atmospheric variables to improve weather-data resolution.
  • Uses a two-step correction approach with a diffusion model.
  • Produces deterministic and probabilistic predictions.
  • Helps recover spectra and distributions for weather extremes.
  • Provides a NIM deployment route for self-hosted weather and climate AI applications.
  • Can be composed with other NIMs in local and global weather/climate prediction pipelines.

NVIDIA context

CorrDiff NIM makes one of Earth-2’s named AI weather models directly queryable in the wiki graph. It connects Earth-2, PhysicsNeMo, and NVIDIA-Modulus model development to the NVIDIA-NIM deployment layer used by enterprise teams.

Connections

  • Earth-2 - parent NVIDIA weather and climate AI platform.
  • NIM-for-Earth-2-FourCastNet - companion Earth-2 NIM for global medium-range forecasting.
  • PhysicsNeMo - training framework and model-development context for Earth science models.
  • NVIDIA-Modulus - neural operator and physics-ML lineage for Earth-2 models.
  • NVIDIA-NIM - deployment and API layer for the microservice.
  • NVIDIA-AI-Enterprise - enterprise support and deployment context for production NIMs.
  • NVIDIA-Omniverse - visualization and digital-twin context for Earth-2 workflows.
  • cuDNN and PyTorch - GPU AI software foundations for model training and inference.

Source Excerpts

  • NVIDIA docs describe CorrDiff as a neural network model for downscaling surface and atmospheric variables.
  • The Earth-2 CorrDiff NIM docs position the model as part of NVIDIA Earth-2 services and software for weather and climate research.

Resources