及时监测干旱与半干旱区光合/非光合植被覆盖度时空变化，可以为指导荒漠化防治工程及植被衰退机制研究提供重要信息。本文以甘肃民勤典型植被白刺灌丛为研究对象，通过地面控制性光谱实验获取混合光谱、端元光谱与丰度信息，开展线性与非线性光谱混合模型（包括核函数非线性和双线性混合模型）估算光合和非光合植被覆盖度的对比研究，采用全限制最小二乘法进行模型解混，分别获取各样本数据中各类端元丰度及其精度信息，通过模型分解的均方根误差（RMSE）与地面验证精度确定用于光合和非光合植被覆盖度估算的最佳光谱混合模型，其中参考端元丰度采用神经网络（NNC）分类算法对数字影像进行分类获取。结果表明：（1）引入阴影端元的四端元模型相对于传统的三端元模型（光合/非光合植被与裸土）能有效提高光谱解混的精度，并提高光合和非光合植被覆盖度估算精度；（2）对白刺灌丛来说，光合植被、非光合植被、裸土及阴影间多重散射混合效应存在，但混合效应不够显著；考虑非线性参数的核函数非线性光谱混合模型表现略低于线性光谱混合模型，因此非线性光谱混合模型在估算白刺灌丛光合和非光合植被覆盖度时相对于线性光谱混合模型没有明显优势；（3）基于光合/非光合植被、裸土与阴影四端元的线性光谱混合模型可以实现白刺灌丛光合和非光合植被覆盖度的准确估算，光合植被覆盖度估算RMSE为0.1177，非光合植被覆盖度估算RMSE为0.0835。

Sand invasion is intensified by the serious degradation and disappearance of Nitraria bushes, which has a serious effect on the oasis ecological security of deserts. Quantitative analysis of different multiple scattering factors in mixed spectral contribution for the ecological environment on deserts is particularly important. Timely monitoring of spatial and temporal variations in photosynthetic/non-photosynthetic vegetation(PV/NPV) fraction cover provides essential information for guiding management practices on land desertification and research on vegetation recession mechanism.
In this paper, taking the typical vegetation of Nitraria bushes in Minqin County of Gansu Province as an example, mixed and endmember spectra, and fraction information were acquired by ground-controlling spectroscopy experiment. Then, the fractional cover of PV(f_pv) and that of NPV (f_npv) were estimated by linear and nonlinear spectral mixture models (NSMM) (including Kernel NSMM (KNSMM) and bilinear spectral mixture model (BSMM)), respectively. Fully constrained least square method was adopted to mix the models, and the fraction of every endmember and the accuracy information of all the samples were calculated. The performances of the models were compared based on root mean square error (RMSE) of the unmixing model and accuracy of field validation, and the endmember fraction of field validation is based on the abundance of digital image classification by the neural network classification algorithm.
Results show that (1) compared with the traditional three-endmember model (PV, NPV, and bare soil (BS)), the four-endmember model, which incorporates an additional shadow endmember, can effectively improve both the accuracy of spectral mixture model(RMSE decreased from 0.0429 to 0.0052 and improved 16% in accuracy) and the estimation precision of f_pv and f_npv(increased by 44% and 83%, respectively). (2)Moreover, the precision of the unmixing of model could be improved by BSMM considering the multiple scattering between NPV and BS endmembers. However, the improved precision was insignificant. Also, considering the nonlinear parameters, the performance of KNSMM was slightly lower than that of the LSMM model. (3) The validation RMSE of f_pv was 0.1177(R²=0.7049), and that of f_npvwas 0.0835 (R²=0.4896) with LSMM based on PV/NPV, BS, and shadow endmembers.
Process monitoring describes the multiple photon-scattering effect among PV/NPV, BS, and shadows in Nitrariabushes.The selection and application of the types of NSMMs should be confirmed according to specific research object and the required precision. Shadows cannot be ignored in estimating vegetation fractional cover, especially in improving f_npv accuracy. This finding illustrates that the types and number of endmembers chosen are significant in improving the accuracy of fraction estimation. The conclusion also shows that LSMM is suitable to estimate f_pv and f_npv of Nitraria bushes accurately based on PV/NPV, BS, and shadow endmembers.