Annual change in the distribution and landscape health of mangrove ecosystems in China from 2016 to 2023 with Sentinel imagery

This is the authors' abstract. We don't add key points for this paper.

As a type of unique coastal wetland, mangroves are subjected to the compounded effects of human activities and climate change. Acquiring reliable and up-to-date information is crucial to support their conservation and sustainable blue carbon development. The integration of remote sensing data and deep learning models enables precise identification of mangroves. This study evaluated the potential of various classical deep learning models for annual mangrove identification in China and analyzed their changes from 2016 to 2023, with a specific focus on the landscape health of mangrove ecosystems within 37 natural reserves and non-reserves. The research shows that compared to the combination model of PSPNet and DeepLabV3 with Resnet, the U-net+ResNet34 model gave the most optimal results in identifying mangroves. In addition, the F1-score of the model reached 0.843 when validated with the samples collected from the national mangrove dataset. Based on the mangrove distribution generated using this U–net+ResNet34, it is observed that the total mangrove area in China exhibited an upward trend during this period. Notably, the centroid of China’s mangroves shifted 26.23 km northeast from the western part of Guangdong. To evaluate the landscape health of mangrove reserves, various landscape pattern metrics were synthesized into the Landscape Health Composite Index through the application of the entropy weight method. Among the 37 natural reserves, 27 of them exhibited an upward trend in landscape health, with the most significant improvements observed in the Taishan Zhenhai Bay Mangrove Nature Reserve. This demonstrates that China has made considerable achievements in the mangrove conservation, encompassing both the quantity and quality of mangroves. However, future efforts should focus particularly on natural reserves where landscape health trends are declining, with an increase in human and material resources investment. This study may serve as a reference for efficiently assessing the health status of mangrove landscapes based on remote sensing data. It can enhance to understand the effectiveness of conservation efforts in mangrove reserves and further promote the scientific management of mangrove protected areas in China. • The U-net+ResNet34 model performs best in mapping large-scale mangrove. • Areas of mangroves in China showed an increasing trend from 2016 to 2023. • More than half of the natural reserves showed an upward trend in landscape health.