| 基于多时相多光谱遥感影像的珊瑚礁面积估算方法研究 |
| |
| 引用本文： | 熊媛,黄荣永,余克服.基于多时相多光谱遥感影像的珊瑚礁面积估算方法研究[J].海洋学报,2022,44(8):151-168. |
| |
| 作者姓名： | 熊媛 黄荣永 余克服 |
| |
| 作者单位： | 1.广西大学 海洋学院，广西 南宁 530004 |
| |
| 基金项目： | 国家自然科学基金( 42090041, 42030502)；广西科技项目( AD17129063, AA17204074)。 |
| |
| 摘 要： | 准确计算珊瑚礁的面积是评估其资源、环境效应的基础，但我国迄今对南海珊瑚礁的面积估算仍缺乏共识，缺少可靠的估算方法是导致这一现象的重要原因。针对这一问题，本文以西沙群岛羚羊礁为例，提出了一种利用多时相多光谱遥感影像低成本半自动化估算珊瑚礁面积的方法。首先快速目视确定地貌带分界线的粗略位置，然后利用基于梯度向量场的主动轮廓线模型（Gradient Vector Flow-Snake, GVF-Snake）实现这些分界线位置的自动精化，最后将不同时相的瞬时分界线转换为面要素进行多时相的融合，从而得到珊瑚礁的面积。基于53景Sentinel-2 多光谱成像仪（MSI）影像的实验表明，羚羊礁的总面积为17.22 km2（Landsat 8 陆地成像仪（OLI）用于方法稳定性的验证，得到的羚羊礁面积为17.29 km2），其中礁前斜坡、礁坪?潟湖坡、潟湖的面积分别为1.76 km2、10.29 km2、5.17 km2。该数值与实测数据具有较好的一致性。具体地，该方法获得的地貌带分界点与实测水深所指示分界点的位置偏差能控制在0.2～4.9 m的范围内（不超过0.5个像素），珊瑚礁最外轮廓线与30 m等深线的位置偏差亦在1个像素大小内（5.7～9.5 m），而估算面积与高分辨率WorldView-2影像解译得到的面积差异为0.02%。同时，该方法获得的珊瑚礁边界线的完整度、正确度、提取质量精度能够由单时相平均的60%、64%和54%分别提高至84%、83%和72%。此外，该方法能够减小基于不同遥感数据源的珊瑚礁面积估算结果的差异，即6景以上的多时相Sentinel-2 MSI和Landsat 8 OLI影像提取的珊瑚礁面积标准差分别不超过0.01 km2和0.05 km2，仅相当于珊瑚礁总面积的0.2%和0.5%。总而言之，该方法能够用低成本的10 m分辨率Sentinel-2 MSI和30 m分辨率Landsat 8 OLI影像获得接近1.8 m分辨率WorldView-2影像的面积估算精度，且具有良好的稳定性和可靠性。
|
| 关 键 词： | 珊瑚礁 遥感影像 面积估算 Sentinel-2 主动轮廓线模型 多时相影像 |
| 收稿时间： | 2021-12-21 |
Estimation of coral reef area from multi-temporal and multi-spectral satellite images: A case study on Lingyang Reef,Xisha Islands |
| |
| Affiliation: | 1.School of Marine Sciences, Guangxi University, Nanning 530004, China2.Guangxi Laboratory on the Study of Coral Reefs in South China Sea, Nanning 530004, China3.Coral Reef Research Centre of China, Guangxi University, Nanning 530004, China4.Southern Marine Science and Engineering Guangdong Laboratory (Zhuhai), Zhuhai 519080, China |
| |
| Abstract: | Estimating coral reef area accurately is fundamental for assessing the resource and environmental effects on coral reefs. However, there is little clear agreement on the areas of coral reefs until now. The main reason is that there is lack of a reliable method to estimate the areas of coral reefs. To address this problem, a low-cost semi-automatic method of coral reef area estimation by using multi-temporal and multi-spectral satellite images is proposed in this paper. The method contains extraction of instantaneous boundaries and fusion of the boundaries. Firstly, the boundaries of the coral reef geomorphologic zones are automatically delineated by using gradient vector flow-snake model (GVF-Snake) after roughly locating the positions of the geomorphic zone boundaries. Thereafter, the extracted multi-temporal geomorphic zone boundaries are converted to geomorphic zone areas and then fused to establish a reliable and accurate geomorphic zone. According to our experiments on the 53 images of Sentinel-2 MSI (the Landsat 8 OLI images are used to verify method stability, the area of Ling Yang Reef is 17.29 km2), the area of Ling Yang Reef is 17.22 km2, among which the areas of the front reef slope, the reef flat-lagoon slope, and the lagoon are 1.76 km2, 10.29 km2, and 5.17 km2, respectively. The results are consistent with in-field survey data. Specifically, the differences between the positions of the geomorphic zone boundaries extracted by using the proposed method and those determined by bathymetric data are in the range of 0.2?4.9 m (less than 0.5 pixel of Sentinel-2 MSI images). The differences between the outline of the coral reef and the 30 m isobath line is also within 1 pixel (5.7?9.5 m). The difference between the area extracted from multi-temporal images by using the proposed method and the area determined by using a high-resolution WorldView-2 image is 0.02%, i.e. Coral reef area calculated from multi-temporal Sentinel-2 MSI images by using our method is able to compete to high-resolution WorldView-2 image in accuracy. Furthermore, the complete, the correction, and the quality of the boundaries are improved from 60%, 64%, and 54% for single-image method to 84%, 83%, and 72% for our multi-temporal method, respectively. Besides, the proposed method can also reduce variations of the estimated coral reef area caused by using satellite images with different sensors. In other word, if more than 6 scenes of satellite images was utilized, the standard deviations of the estimated coral reef area are shown to be less than 0.01 km2 and 0.05 km2 respectively for Sentinel-2 MSI and Landsat 8 OLI images. They are only equivalent to 0.2% and 0.5% of the total coral reef area. In summary, the proposed method is accurate, reliable, and stable for coral reef area estimation. |
| |
| Keywords: | |
|
| 点击此处可从《海洋学报》浏览原始摘要信息 |
|
点击此处可从《海洋学报》下载全文 |
|
