JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Enhancing soil nitrogen measurement via visible-near infrared spectroscopy: Integrating soil particle size distribution with long short-term memory models

  • 0School of Resources and Environment, University of Electronic Science and Technology of China, Chengdu 611731, PR China.
Spectrochimica Acta. Part a, Molecular and Biomolecular Spectroscopy +

|

October 29, 2024

|

October 29, 2024

View abstract on PubMed

Summary

This summary is machine-generated.

This study introduces a deep learning model integrating soil particle size distribution (PSD) with Visible Near-Infrared (Vis-NIR) spectroscopy to accurately measure soil nitrogen. The InSGraL framework significantly improves accuracy by accounting for PSD, crucial for precision agriculture.

Area Of Science

  • Agricultural Science
  • Environmental Science
  • Spectroscopy

Background

  • Accurate soil nitrogen data is vital for agriculture and environmental monitoring.
  • Traditional chemical methods for soil nitrogen analysis are labor-intensive.
  • Visible Near-Infrared (Vis-NIR) spectroscopy offers a rapid and efficient alternative, but its accuracy is hindered by soil particle size distribution (PSD).

Purpose Of The Study

  • To develop an innovative deep learning methodology integrating PSD with Vis-NIR spectroscopy for accurate soil nitrogen measurement.
  • To explore different strategies for integrating PSD and spectral data in deep learning models.
  • To enhance the reliability and accuracy of soil nitrogen content determination.

Main Methods

  • A deep learning framework, InSGraL, was developed, incorporating mixed features of PSD and Vis-NIR spectra as Long Short-Term Memory (LSTM) inputs.
  • The LUCAS dataset was utilized to train and validate the models.
  • Shapley Additive exPlanations (SHAP) analysis was employed to understand the model's behavior and feature importance.

Main Results

  • The InSGraL framework achieved superior performance compared to models using only Vis-NIR data, with a 39.47% reduction in RMSE and a 42.55% decrease in MAE.
  • The model demonstrated robust performance across diverse land cover types, achieving an R² of 0.94 for grassland samples.
  • SHAP analysis indicated that PSD integration effectively mitigated spectral interference from particle size and highlighted previously obscured critical wavelengths.

Conclusions

  • Integrating PSD with Vis-NIR spectroscopy using deep learning offers an innovative and effective strategy for accurate soil nitrogen measurement.
  • The InSGraL framework provides a reliable method to overcome the limitations imposed by PSD on Vis-NIR spectral analysis.
  • This approach contributes to a deeper understanding of the interplay between soil properties and spectral data for agricultural and environmental applications.

Keywords: