随着再生水越来越多的应用于城市湿地, 湿地植物生长状态的大面积监测对于利用再生水的湿地恢复与重建具有重要意义。 目前遥感技术已成为植物生长状态大面积监测的重要手段。 本研究以北京市典型再生水城市湿地奥林匹克公园南园湿地为研究区, 以反映植物生长状态的重要指标全氮(TN)为研究对象, 在测定研究区湿地植物芦苇(Phragmites australis)和香蒲(Typha angustifolia)的叶片光谱及TN含量的基础上, 对数据进行预处理并建立二者的关系模型, 包括单变量模型(比值光谱指数(SR)模型和归一化差值光谱指数(ND)模型), 与多变量模型(逐步多元线性回归(SMLR)模型和偏最小二乘回归(PLSR)模型), 并利用交叉验证决定系数(R2CV)和均方根误差(RMSECV)对模型精度进行检验。 结果表明, 不同湿地植物类型相比, 利用芦苇反射光谱建立的各种预测模型的精度都高于香蒲； 不同回归模型相比, 多变量回归模型的精度较高； 多变量回归模型中, PLSR模型的精度高于SMLR模型, 其R2CV可达0.80, RMSECV仅为0.24, 是建立湿地植物光谱与TN含量关系的最优模型。 研究成果不仅为湿地植物生长状态遥感探测提供参考借鉴, 而且可以为利用再生水的城市湿地监测与管理提供有力的科学依据。

More and more urban wetlands have been supplied with reclaimed water. And monitoring the growth condition of large-area wetland vegetation is playing a very important role in wetland restoration and reconstruction. Recently, remote sensing technology has become an important tool for vegetation growth monitoring. The South Wetland in the Olympic Park, a typical wetland using reused water, was selected as the research area. The leaf reflectance spectra and were acquired for the main wetland plants reed (Phragmites australis) and cattail (Typha angustifolia) with an ASD FieldSpec 3 spectrometer (350～2 500 nm). The total nitrogen (TN) content of leaf samples was determined by Kjeldahl method subsequently. The research established univariate models involving simple ratio spectral index (SR) model and normalized difference spectral index (ND) model, as well as multivariate models including stepwise multiple linear regression (SMLR) model and partial least squares regression (PLSR) model. Moreover, the accuracy of all the models was tested through cross-validated coefficient of determination (R2CV) and cross-validated root mean square error (RMSECV). The results showed that (1) comparing different types of wetland plants, the accuracy of all established prediction models using Phragmites australis reflectance spectra was higher than that using Typha angustifolia reflectance spectra. (2) compared with univariate techniques, multivariate regressions improved the estimation of TN concentration in leaves. (3) among the various investigated models, the accuracy of PLSR model was the highest (R2CV=0.80, RMSECV=0.24). PLSR provided the most useful explorative tool for unraveling the relationship between spectral reflectance and TN consistence of leaves. The result would not only provide a scientific basis for remote sensing retrieval of biochemical variables of wetland vegetation, but also provide a strong scientific basis for the monitoring and management of urban wetlands using recycled water.