Scholarlink

<p>Remote sensing and machine learning techniques together could be useful for assessing some of the chemical properties of soils, which are essential for sustainable agriculture and precision farming. This study investigated the application of Sentinel-2 imagery and the&#xa0;Random Forest Regression model (RFR) to predict key soil properties, including Humus (%), Nitrate Nitrogen (N-NO<sub>3</sub>), and Phosphorus (P), and Potassium (K). Among these, K was predicted with the highest accuracy, achieving R2 values up to 0.66 and an RMSE of 0.62. Feature importance analysis revealed that vegetation indices such as Normalized Difference Red Edge Index (NDRE) and Green Normalized Difference Vegetation Index (GNDVI), along with spectral bands like NIR, were the most influential predictors, highlighting the importance of red-edge and NIR-sensitive regions for potassium estimation. In contrast, the RFR model struggled to predict Humus (%), N–NO<sub>3</sub>, and P, indicating weaker correlations between these properties and the spectral data used. These findings underline the utility of the&#xa0;RFR model in leveraging remote sensing data for potassium monitoring while highlighting challenges in modeling other soil properties. Future research should explore the integration of additional data sources, such as hyperspectral imagery and environmental covariates, to improve prediction accuracy. This study demonstrates the potential of machine learning and remote sensing for non-invasive soil monitoring and contributes valuable insights for advancing precision agriculture practices.</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

Leveraging Vegetation Indices and Random Forest for Soil Nutrient Monitoring in Winter Wheat

  • 期刊论文

  • 发表日期：

    2025年5月16日

查看全文
  • Shukhrat Shokirov,
  • Ilhom Abdurahmanov,
  • Zokhid Mamatkulov,
  • Zoir Abdurahmanov,
  • Doniyor Zarifboev,
  • Rustam Oymatov,
  • Zoltán Kovács,
  • Judit Csabai,
  • Ye Shiping,
  • Obid Khakberdiev

摘要

Remote sensing and machine learning techniques together could be useful for assessing some of the chemical properties of soils, which are essential for sustainable agriculture and precision farming. This study investigated the application of Sentinel-2 imagery and the Random Forest Regression model (RFR) to predict key soil properties, including Humus (%), Nitrate Nitrogen (N-NO3), and Phosphorus (P), and Potassium (K). Among these, K was predicted with the highest accuracy, achieving R2 values up to 0.66 and an RMSE of 0.62. Feature importance analysis revealed that vegetation indices such as Normalized Difference Red Edge Index (NDRE) and Green Normalized Difference Vegetation Index (GNDVI), along with spectral bands like NIR, were the most influential predictors, highlighting the importance of red-edge and NIR-sensitive regions for potassium estimation. In contrast, the RFR model struggled to predict Humus (%), N–NO3, and P, indicating weaker correlations between these properties and the spectral data used. These findings underline the utility of the RFR model in leveraging remote sensing data for potassium monitoring while highlighting challenges in modeling other soil properties. Future research should explore the integration of additional data sources, such as hyperspectral imagery and environmental covariates, to improve prediction accuracy. This study demonstrates the potential of machine learning and remote sensing for non-invasive soil monitoring and contributes valuable insights for advancing precision agriculture practices.