Toward a better understanding of coastal salt marsh mapping: A case from China using dual-temporal images

Coastal salt marsh vegetation is an assembly of herbs, grasses and low shrubs, playing vital ecological roles in promoting sedimentation, providing habitat, and carbon sequestration. Remote sensing-based mapping of coastal salt marsh vegetation has been enhanced through the application of phenological information and various black-box algorithms. However, existing research based on dual-temporal images faces the following challenges: 1) low efficiency in constructing dual-temporal images, 2) dependence on post-processing rather than reducing misclassifications at the source, and 3) failure to reveal implicit knowledge implied in a trained model. This study proposed a comprehensive approach addressing these issues. It efficiently constructed dual-temporal images by utilizing a latitude gradient-based common time period of the growing season in the north and that in the south (the same common time period for the senescence season), iteratively optimized samples to reduce misclassifications, and approximated the trained model through a decision rule to reveal implicit knowledge. Using independently collected samples, the produced 10-me resolution map in China for 2020 achieved an overall accuracy of 92.5%. The study elaborated on the potential mechanism for identifying coastal salt marsh vegetation based on dual-temporal images, indicating that water signals during the growing season are obscured by flourishing vegetation, while these signals are observable during the senescence season when vegetation withers. This research provides a novel approach for fine-scale mapping of coastal wetland land covers based on remote sensing, facilitating coastal conservation and blue carbon estimation studies.滨海盐沼植被（Coastal salt marsh vegetation）由草本植物、禾草与低矮灌木组成，在促进泥沙沉积、提供生物栖息地及碳固存（carbon sequestration）方面发挥着至关重要的生态功能。基于遥感的滨海盐沼植被制图，可通过引入物候信息（phenological information）与各类黑箱算法（black-box algorithms）实现优化。然而，现有基于双时相影像（dual-temporal images）的相关研究仍面临以下挑战：1）双时相影像构建效率偏低；2）依赖后处理手段而非从源头减少分类错分问题；3）无法揭示训练模型中蕴含的隐性知识。本研究提出了一套可有效解决上述问题的综合研究方法：基于纬度梯度，利用南北生长季的共同时段（衰老季亦采用同一共同时段）高效完成双时相影像的构建，通过迭代优化样本以降低分类错分率，并借助决策规则对训练完成的模型进行近似还原，以揭示其中的隐性知识。本研究利用独立采集的样本，生成了2020年中国全域10米分辨率的滨海盐沼植被分布图，总体精度达92.5%。本研究阐明了基于双时相影像识别滨海盐沼植被的潜在机制：生长季的水体信号会被繁茂的植被遮蔽，而在植被枯萎的衰老季，此类信号则可被清晰观测到。本研究为基于遥感技术的滨海湿地地表覆盖精细制图提供了全新路径，可为海岸带保护与蓝碳（blue carbon）估算相关研究提供有力支撑。