Scholarlink

<p>Soil texture is a fundamental soil property that influences various ecological and agricultural processes, including water retention, nutrient availability, permeability, and erosion susceptibility. Conventional methods of soil texture analysis are labour-intensive, costly, and spatially limited, often requiring a large number of samples for high-resolution mapping over extensive areas. In response to these limitations, the present study investigates the use of advanced machine learning models specifically Multilayer Perceptron (MLP), Decision Tree (DT), and Random Forest (RF) for the digital mapping of soil texture in the Yazd-Ardakan plain, a desert region in Iran. This research integrates a broad spectrum of environmental covariates, including land use, vegetation indices, groundwater quality, soil salinity, Digital Elevation Model (DEM) derivatives, and proximity to anthropogenic features, to predict the soil texture classes (sand, silt, and clay) at a depth of 0–20&#xa0;cm. A total of 201 soil profiles were systematically sampled using the hypercube sampling method, and the particle size distribution of the soil samples was analysed to determine the sand, silt, and clay content. The performance of the predictive models was evaluated using statistical metrics such as the coefficient of determination (R<sup>2</sup>), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). The results indicated that the MLP model outperformed the Decision Tree and RF models, exhibiting the highest accuracy in predicting soil texture. Specifically, the MLP model achieved R<sup>2</sup> values of 0.53, 0.88, and 0.96 for clay, silt, and sand, respectively, with corresponding RMSE values of 0.11, 0.07, and 0.03. The dominant soil texture in the region was found to be sandy, with the sand and sandy loam classes occupying approximately 51% of the study area (250,570 hectares). This study underscores the potential of machine learning-based approaches for soil texture prediction, highlighting the utility of DSM (Digital Soil Mapping) techniques in providing high-resolution spatial maps of soil properties. The findings demonstrate that machine learning models, particularly MLP, can be effectively employed to generate accurate, scalable, and cost-efficient soil maps, which are crucial for agricultural management, environmental monitoring, and sustainable land use planning in arid regions. Future work should explore the applicability of this methodology across different regions and assess its potential for broader environmental modelling applications.</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

Digital mapping of soil texture with the use of MLP, Decision Tree and RF models in Iran's desert region

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Hassan Fathizad,
  • Mohammad Ali Hakimzadeh Ardakani,
  • Benyamin Eshghi

摘要

Soil texture is a fundamental soil property that influences various ecological and agricultural processes, including water retention, nutrient availability, permeability, and erosion susceptibility. Conventional methods of soil texture analysis are labour-intensive, costly, and spatially limited, often requiring a large number of samples for high-resolution mapping over extensive areas. In response to these limitations, the present study investigates the use of advanced machine learning models specifically Multilayer Perceptron (MLP), Decision Tree (DT), and Random Forest (RF) for the digital mapping of soil texture in the Yazd-Ardakan plain, a desert region in Iran. This research integrates a broad spectrum of environmental covariates, including land use, vegetation indices, groundwater quality, soil salinity, Digital Elevation Model (DEM) derivatives, and proximity to anthropogenic features, to predict the soil texture classes (sand, silt, and clay) at a depth of 0–20 cm. A total of 201 soil profiles were systematically sampled using the hypercube sampling method, and the particle size distribution of the soil samples was analysed to determine the sand, silt, and clay content. The performance of the predictive models was evaluated using statistical metrics such as the coefficient of determination (R2), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). The results indicated that the MLP model outperformed the Decision Tree and RF models, exhibiting the highest accuracy in predicting soil texture. Specifically, the MLP model achieved R2 values of 0.53, 0.88, and 0.96 for clay, silt, and sand, respectively, with corresponding RMSE values of 0.11, 0.07, and 0.03. The dominant soil texture in the region was found to be sandy, with the sand and sandy loam classes occupying approximately 51% of the study area (250,570 hectares). This study underscores the potential of machine learning-based approaches for soil texture prediction, highlighting the utility of DSM (Digital Soil Mapping) techniques in providing high-resolution spatial maps of soil properties. The findings demonstrate that machine learning models, particularly MLP, can be effectively employed to generate accurate, scalable, and cost-efficient soil maps, which are crucial for agricultural management, environmental monitoring, and sustainable land use planning in arid regions. Future work should explore the applicability of this methodology across different regions and assess its potential for broader environmental modelling applications.