| Abstract: |
The exotic Sonneratia apetala is widely planted in mangrove afforestation in China due to its high adaptability and fast growth rates. This species has triggered intense debate on its ecological invasion risk during the past decades because of its natural reproduction, dispersal, and spread. However, national plans for the management and control of this exotic species are unclear, partly due to the lack of an accurate distribution map of the species for broad latitudinal areas. Mangrove species with subtle spectral differences and varied growth phases require plenty of samples to describe their spectrum; however, the scarcity of samples resulting from the low accessibility of their habitats hinders the mapping of the species across the national coastal zone. To overcome this problem, we derived S. apetala samples from existing discrete localized studies and then iteratively optimized the trained binary model by incorporating new negative samples until a threshold converged. Negative samples were more easily acquired in areas where the absence of S. apetala had been confirmed. This approach avoids the prereq-uisite that S. apetala can be distinguished by visual inspections, which is commonly used in routine classification procedures or active learning classifiers. The approach was applied to derive classification results with the help of a Random Forest classifier using both Sentinel-1 and-2 imagery hosted on Google Earth Engine, considering that S. apetala differs from native mangrove species in terms of the large crown, drooping branches, and biochemical properties. The generated S. apetala map was evaluated using three prepared datasets and achieved overall accuracies of 98.1 % and 96.4 % using the test dataset and independent evaluation dataset, respectively, as well as an accuracy of 91.7 % using 145 field samples provided by mangrove specialists. The total area of exotic S. apetala in China reached 2,968 ha in 2020, accounting for 11.0 % of the total mangrove area in China. This study is the first attempt to delineate the detailed national-scale distribution of S. apetala in coastal China. The information provided in this study can support the management and control of S. apetala. The developed approach can be generalized to other vegetation species in broad latitudinal areas, and can be further improved by probing the internal details of the trained classifier.
|
| Participation of the author: |
Zhao, C. P. Qin, C. Z. Wang, Z. M. Mao, D. H. Wang, Y. Q. Jia, M. M. |