附件
|
|
|
| |
|
|
| 本文已被：浏览 692次 下载 329次 |
|
|
|
| 联合Sentinel-1 SAR和Sentinel-2 MSI的湖泊浮叶和挺水植被自动分类算法 |
|
辛逸豪1, 罗菊花1, 徐颖1, 秦海涛1, 孟迪2, 何锋2, 鲁露2, 陈青3, 徐亚田3
|
|
1.:中国科学院南京地理与湖泊研究所;2.昆明市滇池高原湖泊研究院;3.中国科学院南京地理与湖泊研究所湖泊
|
|
| 摘要: |
| 浮叶植被和挺水植被是湖泊的重要初级生产者，在湖泊生态系统中发挥着不同的生态功能。利用卫星遥感技术监测浮叶植被和挺水植被空间分布和变化，对湖泊生态评估和碳源汇核算具有重要意义。但是由于两类植被均具有典型的植被光谱特征，仅使用光学遥感难以进行区分，并且在富营养化湖泊中其分类还会受到具有相似光谱特征的藻华干扰。针对这些问题，本研究提出了一种联合Sentinel-1 SAR和Sentinel-2 MSI的浮叶植被和挺水植被自动分类算法。算法首先通过归一化植被指数（NDVI）和Otsu算法获取湖泊中具有植被光谱特征的地物区域，然后使用Sentinel-1 SAR影像的第一主成分（PCA1）和K-means聚类算法从该区域中提取浮叶植被和挺水植被。其中，PCA1是算法的核心分类指标，可以去除藻华影响并实现浮叶植被和挺水植被的准确分离。算法在四个典型湖泊中开展了精度验证，平均总体分类精度为83.76%，Kappa系数为0.71。基于该算法，我们获取了太湖年内和年际浮叶植被和挺水植被的变化。结果表明，两类植被的年内覆盖度峰值均出现在7-10月份；2016-2023年间，浮叶植被面积显著增加，从24.21 km2增至68.03 km2，而挺水植被面积则相对稳定，年均面积约为41.48km2。该算法不仅解决了浮叶和挺水植被识别难的问题，还实现了自动化。在大尺度湖泊浮叶和挺水植被时空变化监测中具有广泛应用前景，为未来的湖泊生态评估和碳源汇核算提供了技术支撑。 |
| 关键词: 湖泊 遥感 浮叶植被 挺水植被 Sentinel-1 SAR Sentinel-2 MSI |
| DOI： |
| 分类号: |
| 基金项目:国家自然科学基金面上项目“湖泊沉水植被区水体二氧化碳分压(pCO2)遥感估算研究”(42271377)；云南省省市一体化专项“基于AI的高原湖泊生态系统关键种群智能感知技术研发及示范”(202202AH210006) |
|
| Automatic classification algorithm for floating-leaved and emergent aquatic vegetation in lakes using the joint Sentinel-1 SAR and Sentinel-2 MSI data |
|
Yihao Xin,Juhua Luo,Ying Xu,Haitao Qin,Di Meng,Feng He,Lu Lu,Qing Chen,Yatian Xu
|
|
: Nanjing Institute of Geography and Limnology,Chinese Academy of Sciences
|
| Abstract: |
| Floating-leaved and emergent aquatic vegetation play crucial roles as primary producers in lake ecosystems, each fulfilling distinct ecological functions. Monitoring the spatial distribution and changes of floating-leaved and emergent aquatic vegetation using satellite remote sensing is essential for lake ecological assessment and carbon source-sink accounting. However, distinguishing between the two types of aquatic vegetation using only optical remote sensing data is challenging due to their typical spectral characteristics. This challenge is further compounded by algal blooms in eutrophic lakes, which also exhibit similar spectral characteristics. To address this issue, we proposed an automatic classification algorithm for identifying two types of aquatic vegetation by combining Sentinel-1 SAR and Sentinel-2 MSI data. Firstly, we identify areas with vegetation spectral characteristics using the Normalized Difference Vegetation Index (NDVI) and Otsu"s method in lakes. Then, in these regions, the first principal component (PCA1) of Sentinel-1 SAR image and the K-means clustering algorithm are used to extract floating-leaved and emergent aquatic vegetation. It"s noted that PCA1 is a key classification indicator of the algorithm, which can remove the interference of algal blooms and achieve the separation of floating-leaved and emergent aquatic vegetation. The algorithm was conducted accuracy validation in four typical lakes, with an average overall classification accuracy of 83.76% and a Kappa coefficient of 0.71. Based on the algorithm, we mapped and analyzed the intra-annual and inter-annual variations of floating-leaved and emergent aquatic vegetation in Lake Taihu. The results showed that the area of both groups reached their coverage peaks from July to October. From 2016 to 2023, the area of floating-leaved aquatic vegetation significantly increased from 24.21 km2 to 68.03 km2, while the area of emergent aquatic vegetation remained relatively stable and the average annual area is 41.48 km2. This algorithm not only addresses the difficulties in identifying floating-leaved aquatic vegetation and emergent aquatic vegetation, but also achieves automation. It has great potential for monitoring large-scale spatial and temporal changes of floating-leaved aquatic vegetation and emergent aquatic vegetation in lakes. This provides the technical support for future lake ecological assessments and carbon source-sink accounting. |
| Key words: Lake Remote sensing Floating-leaved aquatic vegetation Emergent aquatic vegetation Sentinel-1 SAR Sentinel-2 MSI |
|
|
|
|
