欢迎访问《草业学报》官方网站,今天是

草业学报 ›› 2025, Vol. 34 ›› Issue (12): 73-84.DOI: 10.11686/cyxb2025024

• 研究论文 • 上一篇    下一篇

基于多特征融合的新疆托乎拉苏草原白喉乌头分布区识别及草地退化监测

尹龙1(), 韩其飞1(), 赵阳2,3, 刘文新2,3   

  1. 1.南京信息工程大学地理科学学院,江苏 南京 210044
    2.伊犁师范大学资源与生态研究所,新疆 伊宁 835000
    3.伊犁师范大学生物与地理科学学院,新疆 伊宁 835000
  • 收稿日期:2025-01-20 修回日期:2025-03-10 出版日期:2025-12-20 发布日期:2025-10-20
  • 通讯作者: 韩其飞
  • 作者简介:E-mail: hanqifei@nuist.edu.cn
    尹龙(1999-),男,江苏南通人,在读硕士。E-mail: 2740309362@qq.com
  • 基金资助:
    国家重点研发计划(2023YFC3206803);国家自然科学基金(42271493);国家自然科学基金(42177436)

Identification of areas of Aconitum leucostomum incursion and monitoring of grassland degradation in the Tuohulasu grassland of Xinjiang based on multi feature fusion

Long YIN1(), Qi-fei HAN1(), Yang ZHAO2,3, Wen-xin LIU2,3   

  1. 1.School of Geographical Sciences,Nanjing University of Information Science & Technology,Nanjing 210044,China
    2.Institute of Resources and Ecology,Yili Normal University,Yining 835000,China
    3.College of Biological and Geographic Sciences,Yili Normal University,Yining 835000,China
  • Received:2025-01-20 Revised:2025-03-10 Online:2025-12-20 Published:2025-10-20
  • Contact: Qi-fei HAN

摘要:

毒杂草型草地退化没有表现出地表裸露、植物生物量减少等特征,对其大尺度的遥感识别较为困难,纹理特征、时相作为遥感影像的重要衍生,能够在影像上提供更多的地物细节,降低了“同物异谱”“同谱异物”的现象,能够很好地提高分类的准确性和可靠性。本研究以伊犁河谷托乎拉苏草原为研究区,利用Sentinel-2卫星数据,对该区主要的毒杂草白喉乌头进行特征提取;基于像元尺度识别白喉乌头分布范围并计算其在混合像元中的占比;最后,通过计算剔除白喉乌头后的植被覆盖度,分析2018-2024年托乎拉苏草原草地退化趋势。结果表明:1) 特征优选有效减少信息冗余,光谱与纹理特征结合有效提高分类精度(总体精度91.67%,Kappa系数0.83)。2) 白喉乌头主要分布于阳坡中海拔平坦区及河谷地带,占研究区40%以上的面积,以稀疏覆盖(0%~0.25%)为主,2018-2024年间各密度等级分布变化为0.67%~1.17%。3) 经校正后草地退化指数2018与2024年均由轻度转为中度,但未退化面积占比提升1.17%,中度与重度退化面积分别下降1.15%与0.70%。本研究为基于多光谱数据进行大区域有毒杂草识别和草原退化监测提供重要方法支撑。

关键词: 托乎拉苏草原, 特征优选, 白喉乌头, 草地退化

Abstract:

Land degradation through colonization by poisonous weeds does not exhibit typical land degradation characteristics such as bare ground or reduced plant biomass, making large-scale remote sensing identification of colonized areas challenging. Texture features and temporal characteristics, as important derivatives of remote sensing images, provide more detailed information on land cover, reducing the ambiguities sometimes referred to as “same object, different spectra” and “same spectra, different objects”. Information from these texture features can significantly improve classification accuracy and reliability. This study focuses on the Tuohulasu grassland in the Ili River Valley, using Sentinel-2 satellite data to extract features indicating presence of the toxic weed Aconitum leucostomum. Based on pixel-scale identification, the distribution range of A. leucostomum was determined, and its proportion in mixed pixels was calculated. Finally, the vegetation cover after excluding A. leucostomum was calculated to analyze the grassland degradation trends in the Tuohulasu grassland from 2018 to 2024. The results show that: 1) Feature selection effectively reduces information redundancy, and the combination of spectral and texture features effectively improves classification accuracy (overall accuracy 91.67%, Kappa coefficient 0.83). 2) A. leucostomum is mainly distributed in the flat areas of sunny slopes and river valleys. A. leucostomum was found in 40% of the study area, with sparse cover (0-0.25%) as the most common scenario. From 2018 to 2024, the distribution of various density levels has changed by 0.67%-1.17%. 3) After correction, the grassland degradation index changed from mild to moderate between 2018 and 2024, but the proportion of non-degraded areas increased by 1.17%, while the areas with moderate and severe degradation decreased by 1.15% and 0.70%, respectively. This study provides important methodological support for large-scale identification of toxic weeds and monitoring of grassland degradation based on multispectral data.

Key words: Tuohulasu grassland, feature selection, Aconitum leucostomum, grassland degradation