Scholarlink

Purpose <p>The content of soil organic matter (SOM) is an important indicator for evaluating soil quality and degradation status. Hyperspectral data have certain application value in the estimation and prediction of soil organic matter (SOM) content, provides effective ideas and approaches for quickly obtaining desert soil quality assessment. Fourier transform infrared (FTIR) technology can provide an enhanced understanding of the chemical composition and dynamic changes of SOM.</p> Methods <p>The content of soil organic matter (SOM) is an important indicator for evaluating soil quality and degradation status. Fourier transform infrared (FTIR) technology can provide an enhanced understanding of the chemical composition and dynamic changes of SOM. Therefore, advancing beyond conventional mathematical methods, our study combined thermal infrared technology, hyperspectral technology, and machine learning algorithms to establish an SOM estimation model. The optimal model was determined and quantitatively evaluated.</p> Results <p>Third order differential (E″′) and third order differential of inverse-log (lg (1/E)″′ ) were the optimal mathematical transformation methods in this study, and could increase the correlation between SOM and emissivity. The accuracy and stability of the SOM estimation models are as follows: BP neural network &gt; quadratic function of E″′ (third-order derivative of thermal infrared hyperspectral data) &gt; random forest. The spatial distribution characteristics of SOM predicted by the three estimation models were highly consistent, and the SOM content of surface soil gradually increased from southeast to northwest. Through model verification and uncertainty analysis, the BP neural network was identified as an optimal estimation model. The average value of SOM estimated using the BP machine learning model ranged from 2.42 to 38.38&#xa0;g/kg at a 95% confidence interval.</p> Conclusion <p>The combination of hyperspectral technology and machine learning algorithms not only provides a theoretical basis for the rapid evaluation of soil quality in arid areas but also lays a technical foundation for the restoration of regional ecosystems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Construction and verification of thermal infrared hyperspectral mathematical model for soil organic matter in arid areas

  • 期刊论文

  • 发表日期：

    2026年1月28日

查看全文
  • Qian Sun,
  • Liang Guo,
  • Guizhen Gao,
  • Aliya Baidurela,
  • Liu Li,
  • Tingwei Song

摘要

Purpose

The content of soil organic matter (SOM) is an important indicator for evaluating soil quality and degradation status. Hyperspectral data have certain application value in the estimation and prediction of soil organic matter (SOM) content, provides effective ideas and approaches for quickly obtaining desert soil quality assessment. Fourier transform infrared (FTIR) technology can provide an enhanced understanding of the chemical composition and dynamic changes of SOM.

Methods

The content of soil organic matter (SOM) is an important indicator for evaluating soil quality and degradation status. Fourier transform infrared (FTIR) technology can provide an enhanced understanding of the chemical composition and dynamic changes of SOM. Therefore, advancing beyond conventional mathematical methods, our study combined thermal infrared technology, hyperspectral technology, and machine learning algorithms to establish an SOM estimation model. The optimal model was determined and quantitatively evaluated.

Results

Third order differential (E″′) and third order differential of inverse-log (lg (1/E)″′ ) were the optimal mathematical transformation methods in this study, and could increase the correlation between SOM and emissivity. The accuracy and stability of the SOM estimation models are as follows: BP neural network > quadratic function of E″′ (third-order derivative of thermal infrared hyperspectral data) > random forest. The spatial distribution characteristics of SOM predicted by the three estimation models were highly consistent, and the SOM content of surface soil gradually increased from southeast to northwest. Through model verification and uncertainty analysis, the BP neural network was identified as an optimal estimation model. The average value of SOM estimated using the BP machine learning model ranged from 2.42 to 38.38 g/kg at a 95% confidence interval.

Conclusion

The combination of hyperspectral technology and machine learning algorithms not only provides a theoretical basis for the rapid evaluation of soil quality in arid areas but also lays a technical foundation for the restoration of regional ecosystems.