Bulletin of Surveying and Mapping ›› 2023, Vol. 0 ›› Issue (6): 36-43.doi: 10.13474/j.cnki.11-2246.2023.0165

Previous Articles     Next Articles

Extraction of invasive plant Spartina alterniflora by combining vegetation phenological characteristics and machine learning supported by GEE

LIU Mingyue1,2,3,4, ZHENG Hao1,2, CHEN Xingtong1, YANG Xiaowu1,2, SONG Jingru1,2, ZHANG Yongbin1, MAN Weidong1,2,3,4   

  1. 1. College of Mining Engineering, North China University of Science and Technology, Tangshan 063210, China;
    2. Tangshan Key Laboratory of Resources and Environmental Remote Sensing, Tangshan 063210, China;
    3. Hebei Industrial Technology Institute of Mine Ecological Remediation, Tangshan 063210, China;
    4. Collaborative Innovation Center of Green Development and Ecological Restoration of Mineral Resources, Tangshan 063210, China
  • Received:2022-08-23 Published:2023-07-05

Abstract: The invasion of alien species threatens biodiversity and destroys ecosystem structure and function. As the only coastal salt marsh plant in the list of the first 16 invasive alien species in China, the rapid and accurate identification of Spartina alterniflora (S. alterniflora) is of great significance for the sustainable development and management of coastal wetlands. Based on time-series Sentinel-2 images on the GEE, HANTS algorithm is used to fit the NDVI time-series curves, then J-M distance is used to preferentially select the key phenological periods of the S. alterniflora. Sentinel-1, Sentinel-2, and DEM datasets during the key phenological periods are integrated to construct spectral, radar, topographic, and texture features. On this basis, four machine learning methods: SVM, CART, RF and GTB are applied to extract S. alterniflora in Yancheng coastal wetlands. The results show that: ①the key phenology extracted by HANTS algorithm and J-M distance optimization for S. alterniflora identification are maturity and early senescence (October—November), and the discrimination of S. alterniflora from native plants in the key phenology is significantly improved; ② 4 classifiers are trained based on multi-source features of the key phenology, the F1-score of S. alterniflora classification resulted from SVM, CART, RF and GTB classifiers are 0.95, 0.93, 0.97 and 0.95, respectively, and RF had the best classification result; ③The existing area of S. alterniflora in the core area of Yancheng Wetland Rare Bird National Nature Reserve (YNNR) is 3 741.86 hm2, accounting for 16.56%. The patches of S. alterniflora show point-source and community boundary-source diffusion, which occupy the ecological niche of native plant communities and threaten the ecological balance of YNNR. Based on the GEE cloud platform, combined with vegetation phenology characteristics and machine learning algorithms, this study utilized multi-source remote sensing data to extract S. alterniflora information in coastal wetlands accurately, rapidly, and efficiently.

Key words: Google Earth Engine, Spartina alterniflora, vegetation phenological characteristics, J-M distance, machine learning

CLC Number: