Mapping Mangroves Ecosystems Using Synthetic Aperture Radar (SAR)

Mangrove ecosystems provide essential ecological services, including shoreline protection, carbon
storage, and habitat support. However, they are increasingly under threat from anthropogenic
pressures such as agricultural encroachment, logging, and climate-induced sea-level rise.
Monitoring mangrove extent and condition is therefore critical, yet challenging due to their
location in remote intertidal zones and frequent cloud cover that limits field-based and optical
observations. Remote sensing offers an effective solution by enabling consistent, large-scale
monitoring of vegetation across time and space.
This study assessed the effectiveness of Sentinel-1 Synthetic Aperture Radar (SAR) data for
mangrove mapping in Ambanja Bay, Madagascar, and evaluated whether fusing it with Sentinel-2 optical imagery enhances classification performance. Using a Random Forest classifier, both
SAR-only and multi-sensor datasets were analyzed to classify land cover and assess changes
between 2019 and 2024.
The SAR-only classification achieved 60% accuracy, with notable confusion between mangroves
and water due to similar backscatter responses. In contrast, the fusion approach significantly
improved classification, achieving an overall accuracy of 94%. Land cover change analysis
revealed transitions from barren land to non-mangrove vegetation and localized expansion of
mangrove cover.
These findings demonstrate that integrating SAR and optical data substantially improves
classification accuracy, reinforcing the value of multi-sensor remote sensing for environmental
monitoring and conservation planning in dynamic coastal ecosystems.

红树林生态系统（Mangrove ecosystems）提供至关重要的生态服务，包括岸线防护、碳固存与栖息地支撑。然而，其正日益受到人为活动压力的威胁，例如农业侵占、伐木以及气候驱动的海平面上升。因此，监测红树林的分布范围与健康状况至关重要，但由于红树林地处偏远潮间带，且频繁的云层覆盖限制了实地观测与光学观测，该任务极具挑战性。遥感技术则提供了有效的解决方案，能够实现跨时空的、一致的大规模植被监测。
本研究评估了马达加斯加安班扎湾地区利用哨兵1号（Sentinel-1）合成孔径雷达（Synthetic Aperture Radar，SAR）数据进行红树林制图的有效性，并探究了将其与哨兵2号（Sentinel-2）光学影像融合是否能够提升分类性能。研究采用随机森林（Random Forest）分类器，分别基于仅使用SAR数据与多传感器融合数据集开展土地覆盖分类，并分析了2019年至2024年间的土地覆盖变化。
仅使用SAR数据的分类准确率仅为60%，由于后向散射响应相似，红树林与水体之间存在显著的分类混淆。相较之下，多传感器融合方法显著提升了分类效果，总体准确率达到94%。土地覆盖变化分析显示，存在裸地向非红树林植被转化的现象，以及局部区域的红树林扩张。
上述研究结果表明，整合SAR与光学影像数据可大幅提升分类精度，进一步证实了多传感器遥感技术在动态海岸生态系统的环境监测与保护规划中的重要价值。