Wetland classification using parcel-level ensemble algorithm based on Gaofen-6 multispectral imagery and Sentinel-1 dataset

Wetland ecosystems have experienced dramatic challenges in the past few decades due to global climate change and human activities. Wetland maps are essential tools for conservation and management of terrestrial ecosystems. The objective of this study was to obtain an accurate wetland map using a parcel-level ensemble method based on Sentinel-1 SAR time series and segmentations generated from GF-6 MPS images. The Dongting Lake wetlands in China, which has a heterogeneous landscape, was deliberately chosen as a challenging case study. First, both VV and VH polarization backscatters (σ0 VV, σ0 VH) were generated from time series Sentinel-1 SAR. Speckle noise inherent in SAR data can affect the performance on image segmentation by resulting in over-segmentation or under-segmentation, and that will degrade the accuracies of wetland classification. Subsequently, optical data (Gaofen-6), which is effectively capable of delineating meaningful parcels, were applied to get the segmentations in this study. Finally, the ensemble method was utilized to extract the wetland information at parcel-level. The overall accuracy and Kappa coefficient of the object-based ensemble method are 90.58% and 0.88, which are 4.25% and 0.04 higher than that of the pixel-based method, respectively. Moreover, the object-based ensemble method for classification in high heterogeneity areas is superior and can greatly improve the performance compared with single classifiers.