论文部分内容阅读
基于高光谱(350~2 500 nm)数据,研究了我国中、东部地区5种主要类型土壤全氮含量与高光谱反射率之间的定量关系,构建了基于偏最小二乘法(PLS)、BP神经网络(BPNN)和特征光谱指数的土壤全氮含量估算模型。结果表明,以500~900 nm、1 350~1 490 nm区域波段反射率经Norris滤波平滑后的一阶导数光谱为基础,构建的基于PLS和BPNN的土壤全氮含量估算模型精度较高,建模决定系数分别为0.81和0.98;独立观测资料检验结果显示,模型预测决定系数分别为0.81和0.93,均方根误差RMSE为0.219 g.kg-1和0.149 g.kg-1,相对分析误差RPD为2.28和3.36,说明PLS和BPNN模型对土壤全氮含量具有较高的预测精度。在光谱指数的分析中,基于近红外872 nm和1 482 nm两个波段的差值光谱指数DI(NDR872,NDR1482)对土壤全氮含量最敏感,建模决定系数、预测决定系数、RMSE和RPD分别为0.66、0.53、0.31 g.kg-1和1.60。比较而言,三种方法估算土壤氮含量的精度顺序为BPNN模型>PLS>DI(NDR872,NDR1482),基于PLS和BPNN两种方法建立的土壤全氮含量高光谱估测模型具有较高的精度,可以用来精确估算土壤全氮含量;基于两波段构建的DI(NDR872,NDR1482)预测效果低于前两者,但也可以用来粗略估测土壤中的全氮含量。
Based on the hyperspectral data (350-2 500 nm), the quantitative relationships between total nitrogen and hyperspectral reflectance of the five main types of soils in the eastern and central regions of China were studied. Based on partial least squares (PLS), BP Estimation model of soil total nitrogen content using neural network (BPNN) and characteristic spectral index. The results show that the model based on PLS and BPNN for estimating soil total nitrogen content based on the first-order derivative spectra of the band reflectivity of 350 ~ 900 nm and 1 350 ~ 1 490 nm after the Norris filtering is smooth, The coefficient of determination of model was 0.81 and 0.98, respectively. The test results of independent observation data showed that the model predictive determination coefficients were 0.81 and 0.93, the root mean square error RMSE was 0.219 g.kg-1 and 0.149 g.kg-1 respectively, and the relative analytical error RPD 2.28 and 3.36, respectively, indicating that PLS and BPNN models have higher prediction accuracy for soil total nitrogen content. In the analysis of spectral index, the difference spectral index DI (NDR872, NDR1482) based on the two bands of near infrared (872 nm) and 1 (482 nm) are the most sensitive to the total nitrogen content of soil. The coefficients of modeling, prediction coefficient, RMSE and RPD 0.66, 0.53, 0.31 g.kg-1 and 1.60, respectively. In comparison, the accuracy of the three methods for estimating soil nitrogen content was BPNN model> PLS> DI (NDR872, NDR1482). The hyperspectral estimation model of soil total nitrogen content based on PLS and BPNN had higher accuracy , Which can be used to accurately estimate the total nitrogen content in soil. The prediction results of DI (NDR872, NDR1482) based on two bands are lower than those of the former two, but they can also be used to roughly estimate the total nitrogen content in soil.