高光谱反射率的滨海地区土壤全磷含量反演

反射光谱在近年来广泛应用于土壤属性的估算。作为一种有效估算土壤全磷含量的手段，反射光谱技术可以很大程度上减少传统化学测量方法所损耗的人力物力。以江苏滨海土壤为研究对象，在30个采样点采集了共147个土样，测量土壤样品光谱反射率及全磷含量。利用原始光谱反射率数据及6种不同的光谱变换结果，通过随机抽样（RS）、KS、SPXY三种样本集划分方法，基于偏最小二乘回归（PLSR）和支持向量机（SVM）方法分别建立土壤全磷含量的估算模型，对比分析了三种样本集划分方法对估算结果精度的影响。结果表明：（1）以原始光谱反射率为数据，PLSR模型，RS方法在多数情况下可以获得较为稳定的模型精度，明显优于KS和SPXY方法；在SVM模型中，采用SPXY方法获得的模型结果最优，KS次之，RS结果最差。（2）不同的样本集划分方法所合适的光谱变换方法不同，对于三种划分样本集方法，PLSR和SVM对应的最优光谱变换分别是对数的倒数和一阶导数（KS方法），原始光谱和一阶导数（RS方法），一阶导数和多元散射校正（SPXY方法）。其中采用KS方法划分样本集，PLSR和SVM均能获得最佳的预测结果。并非所有光谱变换方法都可以提高模型精度，部分光谱变换后PLSR模型预测精度显著降低；（3）在所有的样本集划分方法中，SVM的建模效果优于PLSR，采用RS方法划分样本集，PLSR的预测精度高于SVM，而采用KS和SPXY方法划分样本集，SVM的预测精度整体高于PLSR。综上所述，本研究区域估算土壤全磷含量的最佳模型是基于KS样本集划分方法和一阶导数光谱变换建立的SVM模型，此时拟合优度（R2_p）为0.82。结果表明反射光谱可以对滨海地区的土壤全磷含量进行有效预测，对土壤磷元素的高效快速反演具有一定的指导意义。

Estimation of Soil Total Phosphorus Content in Coastal Areas Based on Hyperspectral Reflectance

Spectral characteristics are the inherent attributes of ground objects. Analyzing spectrum is help to improve the accuracy of ground objects recognition and a basis of quantitative remote sensing. However, limited by scale effect, the spectrum acquired in near-earth space is often quite different from that of remote sensing pixels. Therefore, revealing the spectral characteristics of typical wetland landscapes on the scale of remote sensing pixels is useful to improve the accuracy of large-scale wetland remote sensing classification and inversion of vegetation parameters. Based on the satellite-borne EO-1 Hyperion data, the reflectance of lotus field, reed land, woodland, paddy, highland, construction land, river，lake and pond were extracted from Lake Nanyang, one of the grass lake wetlands in North China Plain.The spectral characteristics of the pixel-scale ground objects were quantitatively analyzed by using the first derivative of the spectrum and calculating a variety of hyperspectral vegetation indexes. The results showed that: (1) The reflectance spectrumof eight wetland landscapes were significantly different, andthere were also differences in the 5 vegetation types. The reflectance of the lotus field was significantly higher than that of other landscapes in the whole wave-band range. It sreflective peak in the green band and absorptive valley in the red band was the most obvious. The reflectance spectrum of the reed field and paddy were similar in visible light and red edge region. The reflectivity curves of paddy and upland farms were different, and the green paddy’s reflective peak was higher than that of upland. (2) The first derivative spectrum of eight landscapes were obviously different at the blue, yellow, and red edge regions, especially at the red edge.The red edge slope of the lotus field was the largest, and the red edge position was obviously blue shift (712 nm), indicates that it has high chlorophyll content and the best health condition. The red-edge slope of woodland was the second, but its red edge position was an obvious red-shift (722 nm). (3) Woodland hadthe highest vegetation index, the vegetation index of water bodies and construction mode rate landscapes land was low, and other. There was no significant difference in the mean values of indexes related to normalized difference vegetation index (NDVI) among reed land, paddy, upland and lotus fields, but only in the Enhanced Vegetation Index (EVI) and Chlorophyll Index RedEdge 710. It suggested that EVI and Chlorophyll Index RedEdge 710 index can more effectively indicate the difference of greenness and coverage between wetland vegetation types. The research has great significance for the high-precision classification wetland of and inversion of vegetation parameters.