Scholarlink

<p>Acid mine drainage (AMD) is one of the most severe environmental threats caused by the oxidation of pyrite (FeS₂) in mine tailings, releasing sulfuric acid and mobilizing potentially toxic elements into surrounding ecosystems. Accurate spatial modeling of iron (Fe) and sulfur (S) -the key constituents of pyrite- is essential for predicting AMD risk and informing remediation strategies. However, conventional geostatistical methods such as simple and ordinary co-kriging (SCK, OCK) struggle with undersampled or heterotopically distributed variables, limiting their effectiveness. This study extends ordinary co-kriging with related means (OCK-RM) by using six regression and machine-learning models-Support Vector Regression, Elastic Net, Lasso, Ridge, Stochastic Gradient Descent, and Linear Regression-to estimate the mean relation between Fe and S in a data-driven way. Using the Haveri Cu–Au tailings dataset with strong Fe-S correlation and partial sampling overlap, model performance is evaluated by cross-validated root mean square error (RMSE) and linear correlation (LC). Lasso Regression provides one of the most stable mean relations (LC = 0.80 for Fe and 0.725 for S) and is used to construct the OCK-RM block model. Compared with simple and ordinary cokriging, this ML-assisted OCK-RM variant gives modest but consistent gains in predictive performance and improves the reproduction of the Fe-S relation and AMD-relevant grade–tonnage behavior.</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

Integrating machine learning regression into ordinary co-kriging with related means for improved spatial modeling of acid mine drainage drivers

  • 期刊论文

  • 发表日期：

    2026年4月15日

查看全文
  • Artur Korniyenko,
  • Nasser Madani,
  • Mohammad Maleki,
  • Annika Parviainen

摘要

Acid mine drainage (AMD) is one of the most severe environmental threats caused by the oxidation of pyrite (FeS₂) in mine tailings, releasing sulfuric acid and mobilizing potentially toxic elements into surrounding ecosystems. Accurate spatial modeling of iron (Fe) and sulfur (S) -the key constituents of pyrite- is essential for predicting AMD risk and informing remediation strategies. However, conventional geostatistical methods such as simple and ordinary co-kriging (SCK, OCK) struggle with undersampled or heterotopically distributed variables, limiting their effectiveness. This study extends ordinary co-kriging with related means (OCK-RM) by using six regression and machine-learning models-Support Vector Regression, Elastic Net, Lasso, Ridge, Stochastic Gradient Descent, and Linear Regression-to estimate the mean relation between Fe and S in a data-driven way. Using the Haveri Cu–Au tailings dataset with strong Fe-S correlation and partial sampling overlap, model performance is evaluated by cross-validated root mean square error (RMSE) and linear correlation (LC). Lasso Regression provides one of the most stable mean relations (LC = 0.80 for Fe and 0.725 for S) and is used to construct the OCK-RM block model. Compared with simple and ordinary cokriging, this ML-assisted OCK-RM variant gives modest but consistent gains in predictive performance and improves the reproduction of the Fe-S relation and AMD-relevant grade–tonnage behavior.