Mapping and classification of Liao River Delta coastal wetland based on time series and multi-source GaoFen images using stacking ensemble model

• Published in: Ecological informatics Vol. 80; p. 102488
• Main Authors: Qian, Huiya, Bao, Nisha, Meng, Dantong, Zhou, Bin, Lei, Haimei, Li, Hang
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01-05-2024
• Subjects: Chinese GaoFen satellite data; Spatial-temporal fusion; Stacking ensemble model; Wetland classification; Wetland classification; Chinese GaoFen satellite data; Spatial-temporal fusion; Stacking ensemble model
• ISSN: 1574-9541
• DOI: 10.1016/j.ecoinf.2024.102488

The precise mapping of coastal wetlands holds great significance for monitoring carbon sequestration and storage within coastal ecosystems, particularly in light of climate change and human-induced activities. Time series and multi-source remote sensing data offer distinct advantages in spatial and temporal land use mapping, particularly in wetland systems encompassing various vegetation types. The primary aim of this study was to delineate the spatial distribution of land use within the Liao River Delta (LRD) wetland. This was achieved by employing a stacking ensemble model that integrates time-series GaoFen-1 (GF-1) optical imagery, GaoFen-3 (GF-3) synthetic aperture radar (SAR) imagery, and Moderate Resolution Imaging Spectroradiometer (MODIS) Normalized Difference Vegetation Index (NDVI) imagery. The first step involved the application of an enhanced spatial and temporal adaptive reflectance fusion model (ESTARFM) to fuse the GF-1 NDVI and MODIS NDVI datasets, producing time-series NDVI data. Subsequently, the parameters pertaining to vegetation phenology were obtained by employing the threshold method on the time-series NDVI data. We compiled feature datasets encompassing GF-1 spectral bands, indices, phenological parameters, and GF-3 SAR features. A Recursive Feature Elimination and Cross-Validation (RFECV) model was employed to identify and select significant features to mitigate data redundancy. Finally, a stacking ensemble model was constructed by combining five base models [K-Nearest Neighbors (KNN), Random Forest (RF), Adaptive Boosting (AdaBoost), Extreme Gradient Boosting (XGBoost), and a Light Gradient Boosting Machine (LightGBM)] to perform wetland classification. The findings were as follows: (1) ESTARFM was able to successfully fuse GF-1 NDVI and MODIS NDVI data, resulting in spatiotemporal fusion with a coefficient of determination (R2) of 0.85 and a root mean square error (RMSE) of 0.07. (2) the RFECV algorithm was employed to select relevant features in the wetland classification process. Specifically, 75 spectral band features, 89 spectral index features, 13 SAR features, and seven phenological parameters were identified as significant for this task. (3) A stacking ensemble model was constructed using the aforementioned multi-source features. This model exhibited a robust and consistent performance in wetland classification, achieving the highest overall accuracy of 94.33%. Notably, this accuracy improvement ranged from approximately 0.09% to 10.02% compared to the individual base models. Thus, the present study has the potential to be utilized for fine-scale wetland monitoring, thereby offering valuable assistance in the field of wetland environmental research.
•Time series NDVI were generated based on fusion of GF-1 and MODIS imagery.•The classification feature dataset from multi-source remote sensing data were constructed.•A stacking ensemble model was proposed in wetland classification and mapping.•Stacking ensemble model outperforms five base models in mapping LRD wetland.