Scholarlink

<p>Accurate digital mapping of soil classes remains essential for sustainable land management. The DSMART (Disaggregation and Harmonization of Soil Map Units through Resampled Classification Trees) algorithm is a probabilistic soil subgroup mapping technique that utilizes environmental covariates in conjunction with legacy soil data. The objective of this study is to disaggregate soil subgroups from existing 1:20,000 scale soil maps using DSMART in a topographically diverse region of Adana, Türkiye. A total of 231 legacy map units were harmonized and resampled to 20&#xa0;m resolution. A total of 40 soil profiles were collected using a conditioned Latin Hypercube Sampling (c-LHS) design. Predictors were derived from a 5-meter digital elevation model (DEM), and NDVI was derived from 20-meter Sentinel-2 images and legacy soil information. The DSMART algorithm was implemented in R using the “C5.0” decision tree model, producing multiple realizations and subgroup probability surfaces. The accuracy of the model was evaluated using Kappa statistics and Shannon entropy. The most prevalent subgroups, Typic Xerofluvent (Tx), Typic Calcixerept (Tc), and Typic Xerorthent (To), were reliably predicted, particularly in heterogeneous landscapes. Subgroups such as Fluventic Haploxerept (Fh) and Oxyaquic Xerofluvent (Ox) demonstrated elevated uncertainty, attributable to insufficient data representation. For instance, while Typic Calcixerept (Tc), Typic Xerofluvent (Tx), and Typic Xerorthent (To) achieved high accuracies of 96.4%, 90.9%, and 88.7% respectively, Ox exhibited the highest misclassification rate (~ 40%) and Fh could not be reliably predicted due to extremely limited samples. The most significant predictors were found to be elevation and curvature-based topographic indices. The overall model demonstrated an accuracy of 85.2%, and the probability surfaces exhibited smoother transitions in comparison to traditional polygon-based soil maps. The DSMART methodology facilitates the generation of detailed, probabilistic soil maps, thereby exceeding the precision of classical survey techniques by accounting for uncertainty and spatial heterogeneity. The findings of this study contribute to the development of soil management strategies for Mediterranean agricultural regions.</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 disaggregation of soil subgroups using DSMART in a Mediterranean semiarid landscape

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月12日

查看全文
  • Yavuz S. Turgut,
  • Y. Kenan Koca,
  • Orhan Dengiz

摘要

Accurate digital mapping of soil classes remains essential for sustainable land management. The DSMART (Disaggregation and Harmonization of Soil Map Units through Resampled Classification Trees) algorithm is a probabilistic soil subgroup mapping technique that utilizes environmental covariates in conjunction with legacy soil data. The objective of this study is to disaggregate soil subgroups from existing 1:20,000 scale soil maps using DSMART in a topographically diverse region of Adana, Türkiye. A total of 231 legacy map units were harmonized and resampled to 20 m resolution. A total of 40 soil profiles were collected using a conditioned Latin Hypercube Sampling (c-LHS) design. Predictors were derived from a 5-meter digital elevation model (DEM), and NDVI was derived from 20-meter Sentinel-2 images and legacy soil information. The DSMART algorithm was implemented in R using the “C5.0” decision tree model, producing multiple realizations and subgroup probability surfaces. The accuracy of the model was evaluated using Kappa statistics and Shannon entropy. The most prevalent subgroups, Typic Xerofluvent (Tx), Typic Calcixerept (Tc), and Typic Xerorthent (To), were reliably predicted, particularly in heterogeneous landscapes. Subgroups such as Fluventic Haploxerept (Fh) and Oxyaquic Xerofluvent (Ox) demonstrated elevated uncertainty, attributable to insufficient data representation. For instance, while Typic Calcixerept (Tc), Typic Xerofluvent (Tx), and Typic Xerorthent (To) achieved high accuracies of 96.4%, 90.9%, and 88.7% respectively, Ox exhibited the highest misclassification rate (~ 40%) and Fh could not be reliably predicted due to extremely limited samples. The most significant predictors were found to be elevation and curvature-based topographic indices. The overall model demonstrated an accuracy of 85.2%, and the probability surfaces exhibited smoother transitions in comparison to traditional polygon-based soil maps. The DSMART methodology facilitates the generation of detailed, probabilistic soil maps, thereby exceeding the precision of classical survey techniques by accounting for uncertainty and spatial heterogeneity. The findings of this study contribute to the development of soil management strategies for Mediterranean agricultural regions.