Wetland mapping by fusing fine spatial and hyperspectral resolution images

Abstract

Despite efforts and progress have been made in wetland mapping using multi-source remotely sensed data, a fine spatial and spectral resolution dynamic modeling of wetland coverage is limited. This research proposed a fusion model to generate fine-spatial-spectral-resolution images by blending multispectral images with fine spatial resolution and hyperspectral images with coarse spatial resolution. Applying the China Environment 1A series satellite (HJ-1A) CCD/HSI data, we showed that the proposed model produced reliable dataset that was not only able to capture spectral fidelity, but also could preserve spatial details. By integrating both fine spatial details and hyperspectral signatures, we further conducted a guided filtering based spectral-spatial mapping on the Poyang Lake wetland. Compared with the classification result of the CCD image, a significant higher classification accuracy of the synthetic fused image was achieved. Results also showed that the final guided-filtering based mapping result could remove potential misclassification biases and achieve higher accuracy than previous pixelwise classification methods Our study indicated a straightforward approach to blend multi-source remotely sensed data to generate reliable, high-quality dynamic dataset for wetland mapping and ecological modelling. The synthetic combination of spatial and hyperspectral details could improve our understanding of the significance of wetland ecosystem.