Infrared‐Guided Super‐Resolution of Remotely Sensed Passive Microwave Sea Surface Temperature

Abstract

Abstract High‐resolution sea surface temperature (SST) is essential for weather forecasting and climate applications. Passive microwave (MW) SST offers largely cloud‐penetrating coverage but is coarse and spatially smoothed, whereas infrared (IR) SST resolves fine‐scale structures but is frequently cloud‐obscured. Using IR as auxiliary guidance for MW super‐resolution is challenging due to cross‐modality discrepancies in effective resolution and radiometric consistency; unconstrained IR injection can introduce spurious textures and local biases. We propose MambaIR‐MSP, a MW‐anchored, IR‐guided super‐resolution framework. A multi‐source prior (MSP) module combines spatial‐frequency alignment (SA) in the frequency domain with physics‐aware affine fusion (PAAF) in the temperature domain to regulate IR guidance, suppress modality‐induced bias, and preserve the MW radiometric baseline. In addition, a channel–frequency attention block (CFAB) selectively enhances gradient‐relevant components while avoiding unstable high‐frequency amplification. Experiments on 1‐year AMSR2–VIIRS collocations over the Kuroshio Extension and an independent Gulf Stream test set show that MambaIR‐MSP achieves competitive RMSE, PSNR, and SSIM relative to representative CNN, Transformer, and SSM baselines, while improving structural fidelity in high‐gradient regions, as reflected by gains in high‐frequency energy ratio, gradient variance ratio, and frontal overlap.