A meta-analysis of the research on the extent and species identification of mangroves has been conducted using remote sensing since 2000. The study clarified the overall accuracy status of mangrove extent and species identification, as well as the influence of remote sensing data source, classification algorithm, feature type and species number on the overall accuracy. The results showed that the overall accuracy range of mangrove extent identification was 55.7% – 99.7% and about 66% of the researches were based on Landsat series satellite data, and had the highest overall accuracy (75% – 99.7%). Optical remote sensing and radar data fusion could effectively improve the overall accuracy of extent identification (>90%). The simpler the type of ground features (≤3 types) or the more complex (≥6 types), the overall accuracy of extent identification is higher and more stable. The overall accuracy of mangrove species identification ranged from 64% to 98.6%; the closer the spatial resolution is to the size of the plant canopy, the higher the overall accuracy of species identification. Among the high spatial resolution remote sensing data sources, the overall accuracy of the species identification of data sources with shortwave infrared bands was higher than that of non-shortwave infrared bands. Multi-source remote sensing data fusion and plants themselves feature information that helps to improve the overall accuracy of category identification. Support vector machines (SVM), maximum likelihood classification (MLC), and random forest (RF) algorithms in supervised machine learning algorithms were the most widely used and had better overall accuracy. As the number of species increases, the overall accuracy of species identification varies with remote sensing data sources and classification algorithms. In summary, the identification accuracy of mangrove extent and species may be improved to a certain extent, and remote sensing data sources, classification algorithms, feature types, and the number of species will all affect the identification accuracy.

对2000年以来基于遥感数据的红树林范围与种类识别的研究结果进行荟萃分析, 阐明红树林范围和种类识别精度的现状, 分析遥感数据源、分类算法、地物类型和物种数对总体精度的影响。结果表明, 红树林范围识别的总体精度范围为55.7%~99.7%; 约66%的研究基于Landsat遥感数据开展, 且总体精度最高(75%~99.7%); 光学遥感与雷达数据融合可有效地提高范围识别的总体精度(>90%); 地物类型越简单(≤3种)或越复杂(≥6种), 范围识别的总体精度越高, 越稳定。红树植物种类识别的总体精度为 64%~98.6%; 空间分辨率越接近红树植物冠幅尺寸, 种类识别的总体精度越高; 在高空间分辨率遥感数据源中, 有短波红外波段的数据源种类识别总体精度高于无短波红外波段; 多源遥感数据融合和植物特征信息有助于提高种类识别的总体精度; 种类识别算法以监督机器学习算法中的支持向量机(SVM)、最大似然分类(MLC)和随机森林(RF)算法应用为最广, 总体精度更高; 随物种数增加, 种类识别总体精度因遥感数据源和分类算法而异。红树林范围和种类识别精度还有提升空间, 遥感数据源、分类算法、地物类型和物种数均会影响识别精度。