Development of spectral-phenological features for deep learning to understand Spartina alterniflora invasion
Jinyan Tian (Capital Normal University)
Le Wang (State University of New York at Buffalo)
Dameng Yin (State University of New York at Buffalo)
Xiaojuan Li (Capital Normal University)
Chunyuan Diao (University of Illinois at Urbana Champaign)
Huili Gong (Capital Normal University)
Chen Shi (Capital Normal University)
Massimo Menenti (Chinese Academy of Sciences, TU Delft - Optical and Laser Remote Sensing)
Yong Ge (Chinese Academy of Sciences)
undefined More Authors (External organisation)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Invasive Spartina alterniflora (S. alterniflora), a native riparian species in the U.S. Gulf of Mexico, has led to serious degradation to the ecosystem and biodiversity as well as economic losses since it was introduced to China in 1979. Although multi-temporal remote sensing offers unique capability to monitor S. alterniflora over large areas and long time periods, three major hurdle exist: (1) in the coastal zone where S. alterniflora occupies, frequent cloud coverage reduces the number of available images that can be used; (2) prominent spectral variations exist within the S. alterniflora due to phonological variations; (3) poor spectral separability between S. alterniflora and its co-dominant native species is often presented in the territories where S. alterniflora intruded in. To articulate these questions, we proposed a new pixel-based phenological feature composite method (Ppf-CM) based on Google Earth Engine. The Ppf-CM method was brainstormed to battle the aforementioned three hurdles as the basic unit for extracting phonological feature is individual pixel in lieu of an entire image scene. With the Ppf-CM-derived phenological feature as inputs, we took a step further to investigate the performance of the latest deep learning method as opposed to that of the conventional support vector machine (SVM); Lastly, we strive to understand how S. alterniflora has changed its spatial distribution in the Beibu Gulf of China from 1995 to 2017. As a result, we found (1) the developed Ppf-CM method can mitigate the phonological variation and augment the spectral separability between S. alterniflora and the background species regardless of the significant cloud coverage in the study area; (2) deep learning, compared to SVM, presented better potentials for incorporating the new phenological features generated from the Ppf-CM method; and (3) for the first time, we discovered a S. alterniflora invasion outbreak occurred during 1996–2001.
help_outline