亚热带红壤全氮的高光谱响应和反演特征研究

利用高光谱遥感技术反演土壤性质已经成为土壤学和遥感科学研究领域的新手段，特别对土壤化学元素含量的高光谱反演，已成为土壤元素快速监测方法的的研究热点。以往研究往往关注不同类型土壤的化学元素光谱响应特征模型，以试图找到普适性的元素-光谱反演模型。由于成土因素的复杂性，土壤类型及其化学元素分布具有明显的空间异质性特征，宏观尺度上的土壤-光谱统计反演模型客观上具有较大的不确定性。若范围缩小到同一个气候带，土壤生物地球化学反应过程较相似，土壤化学元素-光谱反演模型的不确定性相对较小。以福州市为研究区，采集福州市典型红壤样品135个，研究土壤全氮含量的高光谱响应特征，对土壤样品在350～2 500 nm的光谱反射率分别进行倒数对数、微分等五种变换，分析变换后的光谱信息与土壤总氮含量的相关性，筛选出强相关敏感波段，通过设计不同的建模和验证样品比例，用逐步多元线性回归获得福州土壤的氮元素高光谱反演优化模型。结果表明：亚热带红壤全氮的敏感光谱波段为：可见光634～688 nm和红外872，873，1 414和1 415 nm;亚热带沿海地区土壤全氮—高光谱反演的优化模型为: Y=5.384X664-1.039(决定系数R2为0.616，均方根误差为0.422 mg·g-1，检验R2为0.608，均方根误差为0.546 mg·g-1)，该模型可以用于福州地区土壤全氮的光谱快速监测。

Spectral Inversion Models for Prediction of Red Soil Total Nitrogen Content in Subtropical Region (Fuzhou)

The present paper studied the hyperspectral response characteristics of red soil, with 135 soil samples in Fuzhou city. After monitoring the hypersectral reflection of soil samples with ASD (analytical spectral device) and total nitrogen contents with Vario MAX(for nitrogen and carbon analysis), the paper gained the spectral reflection data between 350～2 500 nm (resolution is 1 nm) and soil total nitrogen contents. Then the paper treated the hyperspectral reflection data with 5 mathematic conversions such as first derivative and second derivative conversions of original reflection, reciprocal logarithmic conversion and its first derivative and second derivative conversion in advance. The next step was to calculate the correlation coefficient of soil nitrogen and the above spectral information, and select the sensitive spectral bands according to the highest correlation coefficient. Finally, by designing different proportions of modeling and validation sample data sets, the paper established the quantitative linear models between soil total nitrogen contents and hyperspectral reflection and its 5 converted information, the final optimal mathematic model between soil nitrogen and hyperspectral information was significantly determined. Results showed that 634～688, 872, 873, 1 414 and 1 415 nm were the main sensitive bands for soil total nitrogen, and Y=5.384X664-1.039 (Y represents soil nitrogen content, X664 is the soil spectral absorbance value at 664 nm) was the optimal soil total nitrogen predicting model (in the model, the determination coefficients R2 and the RMSE of total nitrogen were 0.616 and 0.422 mg·g-1, the inspection coefficient R2 and the RMSE were 0.608 and 0.546 mg·g-1 respectively). The model can be used to rapidly monitor soil total nitrogen with hyperspectral reflection in Fuzhou area.