Scholarlink

<p>The widespread adoption of precision agriculture hinges on the timely acquisition of low-cost, high-quality soil data. Visible-near infrared (Vis-NIR) spectroscopy has emerged as a promising alternative to traditional methods, enabling rapid estimation of soil properties through spectral libraries calibrated with regression models. While existing studies largely focus on topsoil, the transferability of these models to subsurface layers under tropical Indian soil conditions remains underexplored. Furthermore, comparative evaluation of spectral preprocessing techniques at the profile level using profile-exclusive models has received limited attention. Therefore, the present study aimed to evaluate spectral preprocessing techniques such as Savitzky-Golay (SG) filter, Standard Normal Variate (SNV) and Multiplicative Scatter Correction (MSC) for developing PLSR-based calibration models to predict soil pH, electrical conductivity (EC), organic carbon (OC), sand, silt and clay contents. The study further assessed the effectiveness of topsoil-calibrated models for predicting subsurface soil properties using the optimal preprocessing approach. A total of 256 horizon wise soil samples collected from 93 profiles of Chamarajanagar district of Karnataka, India were used in the study. SG filtering outperformed SNV and MSC, yielding higher prediction accuracy for clay, sand, and pH predictions, whereas predictions for silt were moderate to poor, and EC and OC were poorly predicted. SG-PLSR models calibrated using topsoil data showed reasonable prediction performance for clay (R² = 0.71 − 0.50), sand (R² = 0.68 − 0.60), and pH (R² = 0.58 − 0.52) up to the fourth soil layer, although model accuracy declined with increasing depth. These results suggest that topsoil derived models demonstrated limited but promising transferability for relatively stable physical properties such as clay and sand, whereas prediction performance for dynamic chemical properties remained weak across subsurface layers.</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

Soil Property Prediction with Vis-NIR Spectroscopy in Different Depths: Testing the Robustness of Topsoil-Derived Models

  • 期刊论文

  • 发表日期：

    2026年6月30日

查看全文
  • K. Palani Renuka,
  • S. Dharumarajan,
  • M. Lalitha,
  • R. Vasundhara,
  • B. Kalaiselvi,
  • A. Sathish

摘要

The widespread adoption of precision agriculture hinges on the timely acquisition of low-cost, high-quality soil data. Visible-near infrared (Vis-NIR) spectroscopy has emerged as a promising alternative to traditional methods, enabling rapid estimation of soil properties through spectral libraries calibrated with regression models. While existing studies largely focus on topsoil, the transferability of these models to subsurface layers under tropical Indian soil conditions remains underexplored. Furthermore, comparative evaluation of spectral preprocessing techniques at the profile level using profile-exclusive models has received limited attention. Therefore, the present study aimed to evaluate spectral preprocessing techniques such as Savitzky-Golay (SG) filter, Standard Normal Variate (SNV) and Multiplicative Scatter Correction (MSC) for developing PLSR-based calibration models to predict soil pH, electrical conductivity (EC), organic carbon (OC), sand, silt and clay contents. The study further assessed the effectiveness of topsoil-calibrated models for predicting subsurface soil properties using the optimal preprocessing approach. A total of 256 horizon wise soil samples collected from 93 profiles of Chamarajanagar district of Karnataka, India were used in the study. SG filtering outperformed SNV and MSC, yielding higher prediction accuracy for clay, sand, and pH predictions, whereas predictions for silt were moderate to poor, and EC and OC were poorly predicted. SG-PLSR models calibrated using topsoil data showed reasonable prediction performance for clay (R² = 0.71 − 0.50), sand (R² = 0.68 − 0.60), and pH (R² = 0.58 − 0.52) up to the fourth soil layer, although model accuracy declined with increasing depth. These results suggest that topsoil derived models demonstrated limited but promising transferability for relatively stable physical properties such as clay and sand, whereas prediction performance for dynamic chemical properties remained weak across subsurface layers.