土壤含水率对高光谱反演紫色土有机质含量的影响研究

Effect of Soil Water on Organic Matter Content of Purple Soil by Hyperspectral Inversion

  • 摘要:
    目的 探究土壤含水率对高光谱反演紫色土有机质含量的影响机制,构建紫色土有机质高光谱估算模型。
    方法 以坡耕地紫色土为研究对象,在室内配置土壤含水率的基础上,以光谱反射率和有机质含量作为数据源,对原始光谱反射率进行去包络线处理、一阶微分和二阶微分等3种数学变换,并采用相关性分析和双因素方差分析方法,筛选土壤有机质、土壤含水率的敏感波段,最终构建偏最小二乘回归和反向传播神经网络模型。
    结果 ① 当土壤含水率介于4% ~ 18%时,水分对光谱反射率的影响相对较大,当土壤含水率介于18% ~ 28%时,水分对土壤光谱的影响较小;随着有机质含量的增加,光谱反射率变化趋势不明显,土壤有机质含量为1.937%时,土壤光谱反射率最大。② 与土壤有机质相比,土壤水分与光谱反射率的相关性较高,其中与原始光谱反射率的相关性最高(|r| = 0.96)。③ 当土壤含水率为23%时,考虑交互作用建立的神经网络模型精度最高且稳定性最强,建模集和验证集的决定系数均为0.97,均方根误差分别为1.34 g kg–1和1.46 g kg–1
    结论 考虑土壤含水率与有机质交互作用时,采用反向传播神经网络模型反演土壤有机质含量可有效提高模拟精度。该研究可为紫色土有机质遥感监测提供理论参考。

     

    Abstract:
    Objective The objective of this study is to investigate the influence mechanism of soil moisture content on hyperspectral inversion of purple soil organic matter content, and establish the hyperspectral estimation model of purple soil organic matter.
    Method Purple soil in sloping farmland was taken as the research object. On the basis of indoor allocation of soil moisture content, spectral reflectance and organic matter contents were used as data sources. The original spectral reflectance was transformed by continuum removal, first-order differential and second-order differential. Correlation analysis and two-factor analysis of variance were used to screen the sensitive bands of soil organic matter and soil moisture content. Partial least squares regression and back propagation neural network models were constructed, followed by establishing partial least squares regression and backpropagation neural network models.
    Result ① Moisture had a significant influence on spectral reflectance when the soil water content ranged from 4% to 18%, but this influence decreased when water content was between 18% and 28%. The spectral reflectance of soil did not exhibit a significant change with the increase in organic matter content, and reached its peak when the soil organic matter content was 1.937%. ② Soil moisture exhibited higher correlation with spectral reflectance compared to organic matter content, with the original spectral reflectance showing the highest correlation (|r| = 0.96). ③ A neural network model considering interaction achieved optimal accuracy and stability at a soil water content of 23%. The coefficient of determination for both modeling set and validation set was 0.97, with root mean square errors of 1.34 g kg−1 and 1.46 g kg−1 respectively.
    Conclusion Considering interaction between soil moisture and organic matter, the inversion of soil organic matter content by neural network model can effectively improve the simulation accuracy. This study would provide theoretical reference for remote sensing monitoring of organic matter in purple soil.

     

/

返回文章
返回