地球信息科学学报 ›› 2023, Vol. 25 ›› Issue (3): 606-624.doi: 10.12082/dqxxkx.2023.220672

• 遥感科学与应用技术 • 上一篇    下一篇

GEE平台下结合滤波算法和植被物候特征的互花米草遥感提取最优时间窗口确定

巫磊(), 吴文挺()   

  1. 福州大学 卫星空间信息技术综合应用国家地方联合工程研究中心 空间数据挖掘与信息共享教育部重点实验室 数字中国研究院(福建),福州 350108
  • 收稿日期:2022-09-08 修回日期:2022-12-06 出版日期:2023-03-25 发布日期:2023-04-19
  • 通讯作者: * 吴文挺(1990— ),男,福建福州人,博士,副研究员,研究方向为海岸带遥感。E-mail: wuwt@fzu.edu.cn
  • 作者简介:巫磊(1997— ),男,安徽合肥人,硕士生,研究方向为海岸带遥感。E-mail: wulei972022@163.com
  • 基金资助:
    福建省自然科学基金青年创新项目(2022J05024);福建省中青年教师教育科研项目(科技类)(JAT210027)

The Optimum Time Window for Spartina Alterniflora Classification based on the Filtering Algorithm and Vegetation Phonology Using GEE

WU Lei(), WU Wenting()   

  1. National & Local Joint Engineering Research Center of Satellite Geospatial Information Technology, Key Laboratory of Spatial Data Mining and Information Sharing of Ministry of Education, The Academy of Digital China (Fujian), Fuzhou University, Fuzhou 350108, China
  • Received:2022-09-08 Revised:2022-12-06 Online:2023-03-25 Published:2023-04-19
  • Contact: WU Wenting
  • Supported by:
    Natural Science Foundation of Fujian Province(2022J05024);Education Department of Fujian Province(JAT210027)

摘要:

互花米草的快速入侵严重影响湿地生态系统平衡。因此,精确监测互花米草扩张的时空动态变化过程具有重要意义。尽管现有基于植被物候特征的互花米草遥感提取方法,避免了光谱特征相似性引起的分类误差,但受到云和潮汐的严重影响,难以获取大尺度湿地植被提取特征信息。本文提出一种结合最大值合成法和Savitzky-Golay(S-G)滤波算法提取互花米草关键物候特征,减弱大尺度云和潮汐对时序遥感信号特征的影响,精准重构符合植被生长趋势的NDVI时间序列数据。通过获取关键物候特征,确定互花米草提取的最优时间窗口,基于Google Earth Engine(GEE)平台精准获取互花米草空间分布状况并分析典型地区的互花米草空间分布特征。研究结果显示,生长季初期(6—7月)为互花米草提取的最优时间窗口,该时期总体分类精度为89.81%,Kappa系数为0.88,相比其他时期的总体分类精度提高10.09%,Kappa系数提高0.11。互花米草提取结果表明,2020年福建省互花米草入侵面积总计100.78 km2,主要分布在宁德、福州、泉州以及漳州等地。其中,宁德市互花米草分布面积最广,共计38.08 km2,占全省互花米草分布总面积的37.79%。福建沿岸的互花米草在空间分布上呈现多种地理特征,在半封闭型海湾和河口地区的沿岸附近主要以连续的条带或片状斑块分布,而在低潮位区域则多是零星斑块。本文研究成果能为互花米草扩散的长时期、大范围空间监测提供可行性方案,为湿地植被精准提取提供技术支撑,为实现海岸带资源的高质量可持续利用提供数据基础。

关键词: 互花米草, 最优时间窗口, 最大值合成, S-G滤波算法, 物候特征, GEE, NDVI时间序列, 空间分布特征

Abstract:

Ecological catastrophes occurred in China's coastal region as a result of rapid invasion of Spartina alterniflora. It's critical to monitor the spatiotemporal dynamics of Spartina alterniflora. Remote sensing has a superior ability to monitor the dynamics of intertidal wetland vegetation on a broad scale as compared to traditional surveying techniques. However, it is of great challenges to precisely determine the amount of Spartina alterniflora using a single-phase image due to the spectral similarities of vegetation with various species. The extensive remote sensing data still need to be further explored in order to better understand the temporal and geographical characteristics of vegetation. The phonological cycle of the plant can be tracked using the time series remote sensing data, which can then provide additional temporal features for vegetation classification. However, the impacts of cloudy weather and tides in intertidal zones make it challenging to extract the phenological parameters, even though previous phenology-based algorithms can reduce misclassification caused by spectral similarity. In this study, we suggested a method to reduce the impact of cloudy weather and tides on time series remote sensing data by integrating the Maximum-value composite algorithm with Savitzky-Golay (S-G) filter. Then, based on the NDVI time series derived from archived Sentinel-2 data, the two-term Fourier function is used to determine the optimal time window and phenological parameters for the classification of Spartina alterniflora. The landscape of Fujian's coastal zone and the distribution of Spartina alterniflora were finally mapped using a random forest classifier on the Google Earth Engine (GEE) platform. The results revealed that, with an overall accuracy of 89.81% and a Kappa coefficient of 0.88, respectively, the beginning of the growing season (June to July) is the optimum time window for classifying intertidal vegetation. It demonstrates how effectively this technology may be used to monitor Spartina alterniflora on a broad scale in coastal areas. According to the results, Spartina alterniflora had a total area of roughly 100.78 km2 in Fujian's coastal zone in 2020, with the most of it being concentrated in Ningde, Fuzhou, Quanzhou, and Zhangzhou. The largest patches of Spartina alterniflora, with a distribution of 37.79%, were found in Ningde. Due to the coastal geomorphology, Spartina alterniflora displayed a large diversity in its spatial distribution along the Fujian coast. The demonstration in Fujian Province shows that the suggested method can offer significant potential for long-term and extensive spatial scale monitoring of Spartina alterniflora dynamics, which could assist coastal high-quality and sustainable development.

Key words: Spartina alterniflora, Optimum time window, Maximum-value Composite, S-G filtering algorithm, Phenology, GEE, NDVI time series, Characteristics of spatial distribution