Wetland Classification Based on Sentinel-1/2 Time Series Feature Optimization
Year: 2026
Type: article
Abstract: Accurate classification of wetlands remains challenging due to complex geomorphological conditions and diverse vegetation communities. Although cloud-based remote sensing data have been widely used for wetland mapping, redundant features can increase computational costs and reduce classification efficiency. This study focuses on the Liaohe Estuary, a representative coastal saltmarsh wetland, and develops a feature optimization framework using multi-temporal Sentinel-1 and Sentinel-2 imagery on the Google Earth Engine platform and the random forest-based recursive feature elimination (RF-RFE) algorithm to achieve wetland classification, assess classification accuracy, and analyze land-cover changes. The results indicate that: (1) Integrating Sentinel-1 and Sentinel-2 data with RF-RFE improved classification performance, achieving an overall accuracy (OA) of 93.75%, a Kappa coefficient of 0.9217, and an F1-score of 81.62%. Compared with the baseline scheme without feature optimization, OA and F1-score increased by 7.75% and 3.92%, respectively, while the Kappa coefficient increased by 0.0096, demonstrating the effectiveness of RF-RFE in reducing feature redundancy and improving classification accuracy. (2) Natural wetlands dominated the study area, while artificial wetlands accounted for a relatively small proportion. Water bodies and Phragmites australis were the most extensive land-cover types, followed by farmland, bare tidal flats, and aquaculture areas. (3) From 2014 to 2024, wetland land-cover patterns changed considerably. Areas of Suaeda salsa marshes and tidal flats decreased after 2020, whereas aquaculture areas and farmland remained relatively stable. Built-up areas and bare soil increased, indicating intensifying anthropogenic influences on the wetland landscape. This study provides an efficient approach to accurate coastal wetland mapping and characterizes land-cover changes over time, offering valuable spatial information to support wetland conservation, ecological restoration, and sustainable management.
Source: Sensors
DOI: 10.3390/s26227256
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
SDGs: Life on land, Climate action
Open Access status: gold
Estimated APC paid: $3,137
OpenAlex ID: W7235309803
