Monthly Diffusion v0.9: A Latent Diffusion Model for the First AI-MIP

arXiv cs.LG / 4/16/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

Monthly Diffusion v0.9 (MD-1.5) is presented as a climate emulator operating on a 1.5-degree grid and focused on low-frequency internal atmospheric variability.
The model uses an SFNO-inspired Conditional Variational Auto-Encoder (CVAE) architecture combined with latent diffusion, aiming to emulate temporal evolution via monthly-mean forward steps.
Training is described for a data-sparse regime, with the design goal of maintaining modest computational requirements.
The paper outlines the motivation, training procedure, and provides initial results, positioning MDv0.9 as an early step toward an “AI-MIP” climate modeling workflow.

Abstract

Here, we describe Monthly Diffusion at 1.5-degree grid spacing (MD-1.5 version 0.9), a climate emulator that leverages a spherical Fourier neural operator (SFNO)-inspired Conditional Variational Auto-Encoder (CVAE) architecture to model the evolution of low-frequency internal atmospheric variability using latent diffusion. MDv0.9 was designed to forward-step at monthly mean timesteps in a data-sparse regime, using modest computational requirements. This work describes the motivation behind the architecture design, the MDv0.9 training procedure, and initial results.