Marchuk: Efficient Global Weather Forecasting from Mid-Range to Sub-Seasonal Scales via Flow Matching

2026-03-25

AI Summary

This work addresses the sharp performance degradation of conventional models in subseasonal (up to 30-day) global weather forecasting, a challenge rooted in atmospheric chaos. To overcome this limitation, the authors propose Marchuk, a generative latent-space autoregressive model based on flow matching. By replacing Rotary Position Embedding (RoPE) with trainable positional embeddings and extending the temporal context window, Marchuk substantially enhances its capacity to model long-range temporal dependencies. Remarkably, with only 276 million parameters—far fewer than the 1.6 billion in comparable models—it achieves forecasting accuracy on par with LaDCast over the full 30-day horizon while offering significantly faster inference. The code and model weights have been publicly released.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Planning, Routing, and Scheduling: Planning with Language Models
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Accurate subseasonal weather forecasting remains a major challenge due to the inherently chaotic nature of the atmosphere, which limits the predictive skill of conventional models beyond the mid-range horizon (approximately 15 days). In this work, we present \textit{Marchuk}, a generative latent flow-matching model for global weather forecasting spanning mid-range to subseasonal timescales, with prediction horizons of up to 30 days. Marchuk conditions on current-day weather maps and autoregressively predicts subsequent days' weather maps within the learned latent space. We replace rotary positional encodings (RoPE) with trainable positional embeddings and extend the temporal context window, which together enhance the model's ability to represent and propagate long-range temporal dependencies during latent forecasting. Marchuk offers two key advantages: high computational efficiency and strong predictive performance. Despite its compact architecture of only 276 million parameters, the model achieves performance comparable to LaDCast, a substantially larger model with 1.6 billion parameters, while operating at significantly higher inference speeds. We open-source our inference code and model at: https://v-gen-ai.github.io/Marchuk/

Problem

• subseasonal weather forecasting
• chaotic atmosphere
• predictive skill
• mid-range forecasting
• weather prediction horizon

Innovation

• flow matching
• latent space forecasting
• subseasonal prediction
• trainable positional embeddings
• efficient weather modeling