Scholarlink

<p>Soil texture is a critical soil property that influences various physical and chemical characteristics, such as water-holding capacity, cation exchange capacity, soil productivity, and aeration. In this study, 394 surface soil samples (0–10&#xa0;cm) were collected from Loess Soils in Golestan Province. The study is based on Sentinel 2 data acquired between April to July 2020 in west Golestan. The aim was to evaluate the performance of three machine learning methods Random Forest (RF), Support Vector Machines (SVM), and Boosted Regression Trees (BRT) for classifying and mapping soil texture.Auxiliary variables, including primary and secondary derivatives of the digital elevation model (DEM), remote sensing spectral indices (RS), and the World Climate dataset, were utilized. The efficiency of the models was evaluated using precision, accuracy, specificity, sensitivity, and kappa coefficient. The results demonstrated that the RF model yielded higher accuracy than the SVM and BRT models, achieving overall accuracies of 0.92, 0.80, and 0.78 for silt, clay, and sand, respectively. Furthermore, the Kappa values in predicting these particles were 0.90, 0.78, and 0.75, respectively.An analysis of variable importance measures revealed that salinity, vegetation index, and drainage network combined with band 6 were the most influential environmental variables for predicting clay, silt, and sand content. Consequently, the RF model is recommended as a valuable and reliable method for producing spatial maps of soil texture in the study area.</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

Ensemble machine learning approaches for estimating soil texture components in loess soils of Golestan Province

  • 期刊论文

  • 发表日期：

    2025年5月9日

查看全文
  • Soraya Bandak,
  • Abdolhossein Boali,
  • Soraya Yaghobi,
  • Ruhollah Taghizadeh-Mehrjardi

摘要

Soil texture is a critical soil property that influences various physical and chemical characteristics, such as water-holding capacity, cation exchange capacity, soil productivity, and aeration. In this study, 394 surface soil samples (0–10 cm) were collected from Loess Soils in Golestan Province. The study is based on Sentinel 2 data acquired between April to July 2020 in west Golestan. The aim was to evaluate the performance of three machine learning methods Random Forest (RF), Support Vector Machines (SVM), and Boosted Regression Trees (BRT) for classifying and mapping soil texture.Auxiliary variables, including primary and secondary derivatives of the digital elevation model (DEM), remote sensing spectral indices (RS), and the World Climate dataset, were utilized. The efficiency of the models was evaluated using precision, accuracy, specificity, sensitivity, and kappa coefficient. The results demonstrated that the RF model yielded higher accuracy than the SVM and BRT models, achieving overall accuracies of 0.92, 0.80, and 0.78 for silt, clay, and sand, respectively. Furthermore, the Kappa values in predicting these particles were 0.90, 0.78, and 0.75, respectively.An analysis of variable importance measures revealed that salinity, vegetation index, and drainage network combined with band 6 were the most influential environmental variables for predicting clay, silt, and sand content. Consequently, the RF model is recommended as a valuable and reliable method for producing spatial maps of soil texture in the study area.