Accurate time-series monitoring of wetland vegetation and water conditions using hyperspectral remote sensing is essential for the precise assessment and comprehensive evaluation of karst wetland ecosystems. However, the low spatial resolution of existing satellite-based hyperspectral images (satellite-based HSI) limits their ability to capture the complex spatial details of wetland vegetation, while the ultra-high-resolution unmanned aerial vehicle hyperspectral images (UAV-HSI) is insufficient for large-scale, time-series monitoring. In addition, existing super-resolution reconstruction methods face challenges in supporting cross-platform reconstruction between satellite-based and UAV-based images, and these methods could not fully meet the demands of time-series monitoring across extensive wetland areas. To address these challenges, this study proposes a cross-platform multiscale image feature mapping module (CMIFM). This module first performs spatial alignment and transformation between UAV-HSI and satellite-based HSI. Then, based on ground-measured spectral data acquired with the analytical spectral devices (ASD), UAV-HSI and satellite-based HSI are mapped into a unified spectral feature space, facilitating the integration of spatial and spectral characteristics. Finally, super-resolution networks, including enhanced super-resolution generative adversarial network (ESRGAN) and swin-transformer for image restoration (SwinIR), are employed to reconstruct high-quality satellite-based HSI. For comparison, this study employs a deep learning-based fusion network (infrared and visible image fusion via dual attention transformer, DATFuse) alongside a traditional fusion method (Gram—Schmidt, GS). The evaluation focuses on the spectral and spatial quality of wetland vegetation and water in the reconstructed and fused results derived from Sentinel-2 and OHS-02 images. This study highlights that (1) The CMIFM-based super-resolution network enhances the spatial resolution of satellite-based HSI by learning the spatial—spectral features of UAV-HSI. This approach effectively reconstructs fine spatial textures of wetland vegetation and water outperforming the traditional GS fusion method in both visual perception and quantitative metrics. Specifically, the reconstructed results of Sentinel-2 and OHS-02 imagery achieve average peak signal-to-noise ratio (PSNR) of 11.06 and structural similarity index measure (SSIM) of 0.3102, when compared with reference data. (2) The spectral features of three typical wetland vegetation communities (Cynodon dactylon, Cladium chinense Nees, and Miscanthus) and water in the reconstructed images exhibit high reliability. Specifically, the OHS-02 reconstructed imagery achieves an average band-wise RMSE of 0.1154 and an R² of 0.7239 when compared with ASD ground-measured data. (3) The CMIFM+ESRGAN and CMIFM+SwinIR methods provide strong generalization capabilities in spatial—spectral reconstruction. They could perform image reconstruction in similar wetland-type scenes even in the absence of UAV-HSI data, achieving average PSNR of 12.74 and SSIM of 0.1897. (4) This study verifies the feasibility of CMIFM-based super-resolution technology in hyperspectral reconstruction images of complex wetlands.

karst wetland;CMIFM module;cross-platform super-resolution reconstruction;DATFuse fusion algorithm;hyperspectral images;quantitative evaluation of spatial-spectral reconstruction quality