基于星-机-地多源遥感的湿地植物群落冠层氮素含量反演

Plant is the important indicator of wetland degradation and restoration and is highly sensitive to global change. Identification of wetland plant communities and retrieval of ecological parameters play important roles in the sustainable wetland management. This study select the Momoge reserve which is one of the typical marsh reserves in Northeast China as the study area. Integrating unmanned aerial vehicle imaging hyperspectral data, lidar satellite data, and the ground spectral measurement data, the aims of this study are to analyze the remote sensing mechanism of identifying wetland plant communities and retrieving nitrogen content, to map the distribution of wetland plant communities using object-based image analysis (OBIA) and decision tree (DT) methods, to develope the model for estimating canopy nitrogen content using hyperspectral data and based on the merits of coupling a genetic algorithm to select spectral variables and partial least squares (GA-PLS) for regression, and further to examine the spatial pattern of canopy nitrogen content of wetland plant communities. This project characterized by discipline and regional features, has novelty during the identifying wetland plant communities and retrieving nitrogen content through the unmanned aerial vehicle and hyperspectral imaging technology. The implementation of this study will help to understand the mechanism and method of classifying wetland plant communities and retrieving nitrogen content, develop assessment of wetland habitat health, and provide scientific theory and technical support for making sustainable ecosystem management strategies.

湿地植物是湿地退化与恢复的重要指示因子，群落尺度上对全球变化极为敏感，植物群落制图和生态参量反演是实现湿地可持续管理的重要的课题。本项目选择我国东北典型草本湿地—莫莫格保护区为研究区，集成无人机高光谱成像数据，雷达卫星数据，地面光谱测量数据等，深入挖掘星-机-地多源遥感信息，解析植物群落遥感识别和氮素含量高光谱反演的机理，应用面向对象和决策树相结合的方法开展湿地植物群落遥感分类与制图；利用基于遗传算法的偏最小二乘法模型（GA-PLS）开展湿地植物冠层氮素含量反演模型构建与评价；进而实现湿地植物冠层氮素含量空间格局及多尺度分析。项目具有明显的学科特色和地域特色，在应用无人机高光谱成像技术开展植物群落遥感识别和冠层氮素含量反演方面具有创新性。本项目的实施将有助于深化理解植物群落分类和氮素含量反演的遥感机理与方法，为合理开展湿地生境健康评估，制定科学的湿地生态系统管理策略提供技术理论支撑。

本项目以莫莫格国际重要湿地为研究区，集成地面观测、无人机高光谱成像、及Sentinel 1/2卫星等天空地多平台数据源，重点解析了不同遥感数据源开展湿地植物群落分类的应用潜力，对比分析了不同湿地植物群落的光谱异质性。对比了应用不同机器学习算法开展基于无人机高光谱成像数据的分类优势，构建了莫莫格湿地植物群落遥感分类模型和冠层氮素含量高光谱遥感反演模型，揭示了不同湿地植物群落类型冠层氮素含量及其空间分异特征。研究发现：面向对象的随机森林算法能够实现湿地植物群落的最优分类；莫莫格湿地核心区以芦苇和香蒲群落为主，分别占研究区总面积的57%和16%；藨草和碱蓬群落分布相对集中。针对芦苇和香蒲群落而言，冠层氮含量均表现为四周高，中间低的空间分布特征。碱蓬群落的冠层氮含量呈现为南北高、中间低的趋势。芦苇、香蒲、碱蓬和藨草群落的冠层氮含量平均值分别为12.96 g/kg、15.39 g/kg、17.39 g/kg、15.98 g/kg，以藨草为最大。氮素含量是植被健康的重要表征指标，本研究凸显了无人机高光谱成像遥感技术和人工智能技术在湿地中的应用潜力，可为湿地保护和管理策略的制定提供科学理论依据。