LangRetrieval: Language-Guided Self-Evolving Satellite-to-Radar Retrieval via CSI-Driven Reward

2026-06-08

AI Summary

This study addresses the ill-posed inverse problem in satellite-to-radar precipitation retrieval, where significantly different precipitation states can exhibit similar cloud-top radiance signatures, rendering conventional static semantic guidance ineffective due to its inability to adapt to dynamic convection and misalignment with the inversion objective. To overcome this, we propose a language-guided conditional flow matching framework that jointly optimizes learnable meteorological semantics and the retrieval task. Our approach leverages a vision-language model for initialization and injects dynamic semantics via cross-attention mechanisms. Furthermore, we introduce a Group Relative Policy Optimization (GRPO) mechanism driven by multi-threshold Critical Success Index (CSI) rewards, enabling closed-loop co-evolution between semantic generation and retrieval accuracy. The method substantially improves retrieval accuracy, robustness, and generalization—particularly in radar-sparse regions and complex scenarios such as intense convection.

Abstract

Satellite-to-radar (S2R) retrieval estimates ground radar precipitation from geostationary satellite observations, providing a critical solution for precipitation monitoring in radar-sparse regions. However, S2R retrieval is intrinsically ill-posed: similar cloud-top radiances can correspond to distinct precipitation regimes, storm organizations, and surface intensities, which are difficult to uniquely determine the underlying meteorological state from local spectral cues alone. Meteorological semantics offer complementary scene-level information that can help resolve this ambiguity. Yet existing static semantic conditioning is often insufficient, as externally predefined semantics cannot adapt to dynamic convective scenes or align with retrieval objectives. To this end, we propose LangRetrieval, a language-guided conditional flow matching (CFM) framework that establishes a closed-loop optimization mechanism between meteorological semantics and retrieval accuracy. Specifically, LangRetrieval consists of two core components: (i) Semantic Warm-up: structured meteorological attributes are injected into the CFM backbone through cross-attention conditioning, enabling continuous semantic guidance throughout the generation trajectory; and (ii) Self-Evolving Semantic Optimization: a lightweight attribute policy is first initialized from vision-language model annotations and subsequently refined via Group Relative Policy Optimization (GRPO) using multi-threshold Critical Success Index (CSI) rewards, enabling semantic generation to evolve directly toward improved retrieval accuracy.

Problem

• Satellite-to-radar retrieval
• ill-posed problem
• meteorological semantics
• precipitation estimation
• semantic ambiguity

Innovation

• language-guided retrieval
• conditional flow matching
• self-evolving semantics
• CSI-driven reward
• satellite-to-radar precipitation