Scholarlink

<p>Climate change and anthropogenic activities are driving substantial shifts in mangrove community structure and accelerating species turnover. However, persistent cloud cover, frequent rainfall, and strong tidal dynamics in coastal regions hinder access to high-quality satellite imagery, thereby limiting species-level identification in mangrove ecosystems. To overcome these constraints, this study integrates niche theory with remote sensing by incorporating environmental variables that constrain species growth into species-level spatial distribution analysis. In addition, interpretable machine learning is employed to quantify the relative contributions of environmental and remote sensing data. Specifically, the Jeffries-Matusita distance was employed to quantify species separability across monthly composite datasets, and the month exhibiting optimal separability was selected as the remote sensing baseline for subsequent classification. Compared with the remote sensing baseline, the proposed framework improved overall accuracy by 16.7% points and increased AUC ROC from 0.803 to 0.900 for 18 classes derived from 26 taxa. Mapping results showed dominance by <i>Avicennia marina</i> (31.6%), <i>Aegiceras corniculatum</i> (17.0%), and <i>Kandelia obovata</i> (16.8%). Introduced species <i>Sonneratia apetala</i> and <i>Laguncularia racemosa</i> together accounted for 14%. Gini importance analysis indicated that coarse-resolution environmental variables define the overall habitat framework and niche constraints. Shapley Additive Explanations demonstrated that high-resolution spectral features, including B5, NDVI based on Band 8A, and EVI<sub><i>asm</i></sub>, are critical for resolving fine-scale species-level variability and spatial boundaries. This study highlights the value of combining niche-based environmental constraints with multi-scale remote sensing information to improve the accuracy and interpretability of species-level mangrove mapping under complex coastal conditions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Mapping Mangrove Distributions in China Using Sentinel-2 Data and Interpretable Machine Learning

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Zhiyi Kan,
  • Huijuan Hu,
  • Bin Chen,
  • Qinhua Fang,
  • Youshao Wang,
  • Shunyang Chen,
  • Hao Cheng,
  • Weiwei Yu,
  • Jiao Fei,
  • Lingyang Feng,
  • Guangcheng Chen,
  • Dian Zhang

摘要

Climate change and anthropogenic activities are driving substantial shifts in mangrove community structure and accelerating species turnover. However, persistent cloud cover, frequent rainfall, and strong tidal dynamics in coastal regions hinder access to high-quality satellite imagery, thereby limiting species-level identification in mangrove ecosystems. To overcome these constraints, this study integrates niche theory with remote sensing by incorporating environmental variables that constrain species growth into species-level spatial distribution analysis. In addition, interpretable machine learning is employed to quantify the relative contributions of environmental and remote sensing data. Specifically, the Jeffries-Matusita distance was employed to quantify species separability across monthly composite datasets, and the month exhibiting optimal separability was selected as the remote sensing baseline for subsequent classification. Compared with the remote sensing baseline, the proposed framework improved overall accuracy by 16.7% points and increased AUC ROC from 0.803 to 0.900 for 18 classes derived from 26 taxa. Mapping results showed dominance by Avicennia marina (31.6%), Aegiceras corniculatum (17.0%), and Kandelia obovata (16.8%). Introduced species Sonneratia apetala and Laguncularia racemosa together accounted for 14%. Gini importance analysis indicated that coarse-resolution environmental variables define the overall habitat framework and niche constraints. Shapley Additive Explanations demonstrated that high-resolution spectral features, including B5, NDVI based on Band 8A, and EVIasm, are critical for resolving fine-scale species-level variability and spatial boundaries. This study highlights the value of combining niche-based environmental constraints with multi-scale remote sensing information to improve the accuracy and interpretability of species-level mangrove mapping under complex coastal conditions.