MFC-RFNet: A Multi-scale Guided Rectified Flow Network for Radar Sequence Prediction

2026-04-21 · Source: cs.AI updates on arXiv.org · Field: Science & Research — Environmental Science & Earth Systems, Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences · Depth: Expert, extended

Summary

MFC-RFNet is a novel generative framework designed for accurate and high-resolution precipitation nowcasting from radar echo sequences. It integrates multi-scale communication with guided feature fusion to address challenges like complex multi-scale evolution, inter-frame feature misalignment, and efficient long-range spatiotemporal context capture. The network incorporates a Wavelet-Guided Skip Connection (WGSC) for high-frequency component preservation, a Feature Communication Module (FCM) for bidirectional cross-scale interaction, and a Condition-Guided Spatial Transform Fusion (CGSTF) for aligning shallow features. Utilizing rectified flow training for stable, few-step sampling and lightweight Vision-RWKV (RWKV) blocks for long-range dependencies, MFC-RFNet demonstrates consistent improvements over strong baselines on four public datasets (SEVIR, MeteoNet, Shanghai, CIKM). It achieves clearer echo morphology at higher rain-rate thresholds and sustained skill at longer lead times, using approximately 27 million parameters and 28.6 GFLOPs.

Key takeaway

For AI Scientists and Machine Learning Engineers developing weather forecasting models, MFC-RFNet's approach to combining rectified flow with specialized modules for multi-scale communication, spatial alignment, and frequency-guided fusion offers a robust path to improved precipitation nowcasting. You should consider integrating similar architectural components, particularly for handling complex spatiotemporal dynamics and preserving high-intensity details, to achieve more accurate forecasts at longer lead times and higher rain-rate thresholds.

Key insights

MFC-RFNet enhances radar precipitation nowcasting through a generative framework that integrates multi-scale communication, spatial alignment, and frequency-aware fusion.

Principles

• Rectified Flow training enables efficient, stable generative modeling.
• Explicit spatial alignment mitigates motion-induced feature misalignment.
• Frequency-aware fusion preserves fine-grained details and structural consistency.

Method

MFC-RFNet uses a U-KAN backbone with a conditional encoder. It applies FCM for cross-scale communication, CGSTF for shallow feature alignment, WGSC for wavelet-guided skip fusion, and VRWKV blocks for long-range spatiotemporal context.

In practice

• Use rectified flow for efficient, high-fidelity generative tasks.
• Implement explicit feature alignment in early network stages.
• Incorporate wavelet transforms for frequency-aware feature modulation.

Topics

• Precipitation Nowcasting
• Rectified Flow Networks
• Multi-scale Feature Fusion
• Spatial Alignment
• Wavelet-Guided Fusion

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Toward AI-Driven Digital Twins for Metropolitan Floods: A Conditional Latent Dynamics Network Surrogate of the Shallow Water Equations — CLDNet significantly accelerates metropolitan flood forecasting by using a latent neural ODE and coordinate-based decoding.
• Weighted quantization using MMD: From mean field to mean shift via gradient flows — MMD minimization via WFR geometry and MSIP offers robust, efficient probability distribution quantization with variable particle weights.
• Meta Flow Maps enable scalable reward alignment — Meta Flow Maps enable efficient, one-step stochastic posterior sampling, resolving a core bottleneck in generative model control and reward alignment.
• NPMixer: Hierarchical Neighboring Patch Mixing for Time Series Forecasting — NPMixer uses hierarchical patching and adaptive wavelet transforms for superior multivariate time series forecasting.
• DSFNet: Learning Dual-Domain Spectral Operators for Multi-Modality Spatio-Temporal Forecasting in Urban Transportation Systems — DSFNet uses dual-domain spectral filtering to model complex spatio-temporal and cross-modality couplings in urban traffic.
• A Hybrid LSTM--Vision Transformer Architecture for Predicting HRRR Forecast Errors — Hybrid LSTM-ViT models improve weather forecast error prediction by integrating temporal surface data with vertical atmospheric profiles.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.