Scholarlink

Purpose <p>Despite extensive research in digital soil mapping (DSM), the synergy between diverse sensing platforms remains poorly understood. This study evaluates whether proximal and orbital soil sensing are complementarity for mapping on-farm variability.</p> Methods <p>This study evaluated sensor complementarity and redundancy across three agricultural sites totaling 375&#xa0;ha. The sensing platforms comprised on-the-go electromagnetic induction (EMI) for apparent electrical conductivity (ECa) and magnetic susceptibility, laboratory-based magnetic susceptibility, x-ray fluorescence (pXRF), and visible-near-infrared diffuse reflectance (VNIR). Additionally, synthetic soil images (SYSI) from Sentinel-2 provided spectral bands and Hue-Saturation-Value conversions. Statistical assessment employed unsupervised techniques, namely Spearman’s correlation, Variance Inflation Factor, and Condition Index, alongside supervised models, including stepwise Multiple Linear Regression, Random Forest, and Partial Least Squares Regression. These two latter models quantified variable importance for clay, silt, sand, total organic carbon, and a soil chemical quality index.</p> Results <p>Redundancy diagnostics revealed that SYSI bands and electromagnetic sensors provided overlapping information, whereas VNIR and pXRF yielded complementary data. While full models exhibited slightly higher predictive accuracy, reduced models utilizing only complementary sensors maintained comparable performance, confirming that a parsimonious variable set is sufficient for soil property prediction. In the absence of spectroscopy (pXRF and VNIR), a combination of ECa with SYSI-Red and Saturation bands were identified as a viable complementary alternative, albeit with lower accuracy than the spectroscopy-based models.</p> Conclusion <p>This study establishes that the synergy between VNIR and pXRF sensors optimizes DSM, rendering additional sensors redundant and outperforming single-sensor configurations. Where spectroscopy is unavailable, the joint use of ECa with SYSI-Redand Saturation bands serve as a practical surrogate. Despite a slight reduction in accuracy, this combination remains a viable tool for delineating on-farm soil variability and supporting site-specific management decisions.</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

Complementarity of proximal and orbital soil sensing for mapping on-farm soil variability

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月1日

查看全文
  • João Vítor Fiolo Pozzuto,
  • Laura Delgado Bejarano,
  • Henrique Oldoni,
  • Lucas Rios do Amaral

摘要

Purpose

Despite extensive research in digital soil mapping (DSM), the synergy between diverse sensing platforms remains poorly understood. This study evaluates whether proximal and orbital soil sensing are complementarity for mapping on-farm variability.

Methods

This study evaluated sensor complementarity and redundancy across three agricultural sites totaling 375 ha. The sensing platforms comprised on-the-go electromagnetic induction (EMI) for apparent electrical conductivity (ECa) and magnetic susceptibility, laboratory-based magnetic susceptibility, x-ray fluorescence (pXRF), and visible-near-infrared diffuse reflectance (VNIR). Additionally, synthetic soil images (SYSI) from Sentinel-2 provided spectral bands and Hue-Saturation-Value conversions. Statistical assessment employed unsupervised techniques, namely Spearman’s correlation, Variance Inflation Factor, and Condition Index, alongside supervised models, including stepwise Multiple Linear Regression, Random Forest, and Partial Least Squares Regression. These two latter models quantified variable importance for clay, silt, sand, total organic carbon, and a soil chemical quality index.

Results

Redundancy diagnostics revealed that SYSI bands and electromagnetic sensors provided overlapping information, whereas VNIR and pXRF yielded complementary data. While full models exhibited slightly higher predictive accuracy, reduced models utilizing only complementary sensors maintained comparable performance, confirming that a parsimonious variable set is sufficient for soil property prediction. In the absence of spectroscopy (pXRF and VNIR), a combination of ECa with SYSI-Red and Saturation bands were identified as a viable complementary alternative, albeit with lower accuracy than the spectroscopy-based models.

Conclusion

This study establishes that the synergy between VNIR and pXRF sensors optimizes DSM, rendering additional sensors redundant and outperforming single-sensor configurations. Where spectroscopy is unavailable, the joint use of ECa with SYSI-Redand Saturation bands serve as a practical surrogate. Despite a slight reduction in accuracy, this combination remains a viable tool for delineating on-farm soil variability and supporting site-specific management decisions.