Scholarlink

<p>Global Positioning System (GPS) based 2216 surface soil (0–15&#xa0;cm) samples were collected across the <i>Kymore Plateau and Satpura hill zone</i> of Madhya Pradesh, India using multi-layer sampling strategy to ensure spatial representativeness. The concentrations of Uranium (<sup>238</sup>U) and Thorium (<sup>232</sup>Th) were analysed using inductively coupled plasma mass spectrometry (ICP-MS/MS) with spectral data acquired through a Spectro-radiometer. The <sup>238</sup>U concentrations ranged from 0.15 to 852.15 ppb (mean: 94.92 ppb), while <sup>232</sup>Th ranged from 0.06 to 1385.06 ppb (mean: 71.73 ppb). Soil pH exhibited a positive correlation with <sup>238</sup>U and a negative correlation with ²³²Th. Several machine learning (ML) models were evaluated for prediction, including Partial Least Squares Regression (PLSR), Support Vector Regression(SVR), Random Forest (RF), Artificial Neural Network (ANN), Extreme Gradient Boosting (XGBoost), Light Gradient-Boosting Machine (LightGBM), Categorical Boosting (CatBoost) and Extreme Learning Machine (ELM), the ANN demonstrated the balanced and robust performance. For <sup>238</sup>U, ANN achieved a testing coefficient of determination (R²) of 0.36 (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\:\sqrt{^{238}\text{U}}\)</EquationSource> </InlineEquation> transformed data), root mean square error (RMSE) of 2.87, and ratio of performance to deviation (RPD) of 1.25, while for ²³²Th it achieved the highest testing R² of 0.56, RMSE of 0.46, and RPD of 1.51. Data transformation square root of <sup>238</sup>U (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\:\sqrt{^{238}\text{U}}\)</EquationSource> </InlineEquation>) and logarithmic of <sup>232</sup>Th (ln <sup>232</sup>Th) improved model performance by stabilizing variance. Geostatistical analysis using an exponential semivariogram model indicated moderate spatial dependence (SD) for <sup>238</sup>U (nugget-to-sill (N/S) ratio ~ 43%, range 18.23&#xa0;km) and moderate to strong SD for ²³²Th (30%, range 11.06&#xa0;km). Validation results root mean square standardised error (RMSS ≈ 1) confirmed reliable uncertainty estimation. Spatial distribution maps revealed <sup>232</sup>Th enrichment in the Panna-Katni region and <sup>238</sup>U enrichment in the Northern districts (Katni-Panna-Satna) and Southern Seoni, with Eastern districts of study area showing relatively lower concentrations.</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

Predictive modelling of Uranium (238U) and Thorium (232Th) in soils of Central India: integrating ICP-MS/MS, spectroscopic, and machine learning models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月14日

查看全文
  • G. S. Tagore,
  • Devid Kumar Sahu,
  • Y. M. Sharma,
  • Sanjay Srivastava,
  • R. K. Nema

摘要

Global Positioning System (GPS) based 2216 surface soil (0–15 cm) samples were collected across the Kymore Plateau and Satpura hill zone of Madhya Pradesh, India using multi-layer sampling strategy to ensure spatial representativeness. The concentrations of Uranium (238U) and Thorium (232Th) were analysed using inductively coupled plasma mass spectrometry (ICP-MS/MS) with spectral data acquired through a Spectro-radiometer. The 238U concentrations ranged from 0.15 to 852.15 ppb (mean: 94.92 ppb), while 232Th ranged from 0.06 to 1385.06 ppb (mean: 71.73 ppb). Soil pH exhibited a positive correlation with 238U and a negative correlation with ²³²Th. Several machine learning (ML) models were evaluated for prediction, including Partial Least Squares Regression (PLSR), Support Vector Regression(SVR), Random Forest (RF), Artificial Neural Network (ANN), Extreme Gradient Boosting (XGBoost), Light Gradient-Boosting Machine (LightGBM), Categorical Boosting (CatBoost) and Extreme Learning Machine (ELM), the ANN demonstrated the balanced and robust performance. For 238U, ANN achieved a testing coefficient of determination (R²) of 0.36 ( \(\:\sqrt{^{238}\text{U}}\) transformed data), root mean square error (RMSE) of 2.87, and ratio of performance to deviation (RPD) of 1.25, while for ²³²Th it achieved the highest testing R² of 0.56, RMSE of 0.46, and RPD of 1.51. Data transformation square root of 238U ( \(\:\sqrt{^{238}\text{U}}\) ) and logarithmic of 232Th (ln 232Th) improved model performance by stabilizing variance. Geostatistical analysis using an exponential semivariogram model indicated moderate spatial dependence (SD) for 238U (nugget-to-sill (N/S) ratio ~ 43%, range 18.23 km) and moderate to strong SD for ²³²Th (30%, range 11.06 km). Validation results root mean square standardised error (RMSS ≈ 1) confirmed reliable uncertainty estimation. Spatial distribution maps revealed 232Th enrichment in the Panna-Katni region and 238U enrichment in the Northern districts (Katni-Panna-Satna) and Southern Seoni, with Eastern districts of study area showing relatively lower concentrations.