Scholarlink

<p>This study evaluates of the DSMART (Disaggregation and Harmonisation of Soil Map Units Through Resampled Classification Trees) algorithm for the probabilistic updating of legacy soil maps in Bhadesar block, Chittorgarh district, Rajasthan. The objective was to generate high-resolution (30&#xa0;m) soil subgroup maps by disaggregating conventional 1:50,000 scale soil polygons using environmental covariates and DSMART modeling. DSMART outputs included spatial predictions of the probable soil classes, along with associated probability and confusion index (CI) maps to quantify uncertainty. Areas with low CI values indicate high classification confidence, typically where one soil class strongly dominates the prediction probabilities. Results showed that the most probable soil class map provided higher classification accuracy, particularly for subgroups like <i>Typic Haplustepts</i>. Evaluation metrics indicated superior user’s and producer’s accuracy for primary class predictions, while second probable classes offered limited improvement refers to accuracy. Covariate usage analysis revealed that normalised difference vegetation index (NDVI) and topographic ruggedness index (TRI) were the most influential predictors, consistently selected in all model runs. The CI map effectively identified zones of classification ambiguity, useful for guiding field validation to improve soil classification. The study shows that probabilistic disaggregation with DSMART improves the spatial detail and reliability of soil maps. This supports precision land management, resource planning, and evidence-based agricultural decisions in heterogeneous landscapes.</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

Probabilistic disaggregation of legacy soil maps using DSMART: a multi-realization framework for soil classification

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月25日

查看全文
  • Pravash Chandra Moharana,
  • Bappa Das,
  • Mahaveer Nogiya,
  • Ram Prasad Sharma,
  • Ravindra Naitam,
  • Buddaraju Amrutha,
  • Ram Sakal Singh,
  • Hrittick Biswas,
  • Nitin Gorakh Patil

摘要

This study evaluates of the DSMART (Disaggregation and Harmonisation of Soil Map Units Through Resampled Classification Trees) algorithm for the probabilistic updating of legacy soil maps in Bhadesar block, Chittorgarh district, Rajasthan. The objective was to generate high-resolution (30 m) soil subgroup maps by disaggregating conventional 1:50,000 scale soil polygons using environmental covariates and DSMART modeling. DSMART outputs included spatial predictions of the probable soil classes, along with associated probability and confusion index (CI) maps to quantify uncertainty. Areas with low CI values indicate high classification confidence, typically where one soil class strongly dominates the prediction probabilities. Results showed that the most probable soil class map provided higher classification accuracy, particularly for subgroups like Typic Haplustepts. Evaluation metrics indicated superior user’s and producer’s accuracy for primary class predictions, while second probable classes offered limited improvement refers to accuracy. Covariate usage analysis revealed that normalised difference vegetation index (NDVI) and topographic ruggedness index (TRI) were the most influential predictors, consistently selected in all model runs. The CI map effectively identified zones of classification ambiguity, useful for guiding field validation to improve soil classification. The study shows that probabilistic disaggregation with DSMART improves the spatial detail and reliability of soil maps. This supports precision land management, resource planning, and evidence-based agricultural decisions in heterogeneous landscapes.