Explainable machine learning and DNN for surrogate modeling of the entropy-weighted water quality index (EWQI) with hydrogeochemical insights in a coal-mining region, India

Surface-water quality in coal-mining regions is threatened by mining, overburden dumping, agricultural runoff, and settlements. This study presents a novel framework integrating hydrochemical analysis, Entropy Weighted Water Quality Index (EWQI), Geographic Information System-based mapping, and machine learning (ML) surrogates to assess surface-water quality in the Barka Sayal coal-mining region, India. Thirteen physicochemical parameters were analyzed in 30 post-monsoon surface-water samples. Pearson correlation, Piper, and Gibbs analyses characterized hydrochemical relationships, facies, and dominant processes. EWQI ranged from 29.62 to 85.54 (mean 56.25), classifying 36.67% of samples as Good, 53.33% Poor, and 10% Very Poor. Mineral dissolution and water–rock interaction dominated, with secondary influences from evaporation and localized anthropogenic inputs. Inverse Distance Weighting maps suggest localized deterioration, though 30 sites limit certainty. Four ML models—Generalized Linear Model (GLM), Random Forest, Gradient Boosting Machine, and Deep Neural Network (DNN)—approximated EWQI from its 13 component variables. DNN achieved the best test-set agreement (root mean squared error = 0.0100; mean absolute error = 0.0088; mean squared error = 0.0001; coefficient of determination, R 2 = 1); differences among models, including DNN versus GLM, were statistically significant in Friedman and Diebold–Mariano tests ( p < 0.05). SHapley Additive exPlanations (SHAP) identified sulphate and fluoride as most influential inputs, partly expected because both are direct EWQI components. Because the models reproduce EWQI from its own components, they are surrogates rather than independent predictors of water quality, and SHAP rankings do not imply causation. The framework supports hydrochemical assessment in mining-affected watersheds; source attribution requires further geochemical evidence.

Authors

Institutions

Publication Details

Journal
Environmental Geochemistry and Health
Published
2026-11-21
DOI
https://doi.org/10.1007/s10653-026-03618-6
Primary Topic
Water Quality and Pollution Assessment
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Explainable machine learning and DNN for surrogate modeling of the entropy-weighted water quality index (EWQI) with hydrogeochemical insights in a coal-mining region, India

Ravi prakash Kumar, Prasoon Kumar Singh
Environmental Geochemistry and Health
Water Quality and Pollution Assessment
article

Explainable machine learning and DNN for surrogate modeling of the entropy-weighted water quality index (EWQI) with hydrogeochemical insights in a coal-mining region, India

Ravi prakash Kumar, Prasoon Kumar Singh
article en

Abstract

Surface-water quality in coal-mining regions is threatened by mining, overburden dumping, agricultural runoff, and settlements. This study presents a novel framework integrating hydrochemical analysis, Entropy Weighted Water Quality Index (EWQI), Geographic Information System-based mapping, and machine learning (ML) surrogates to assess surface-water quality in the Barka Sayal coal-mining region, India. Thirteen physicochemical parameters were analyzed in 30 post-monsoon surface-water samples. Pearson correlation, Piper, and Gibbs analyses characterized hydrochemical relationships, facies, and dominant processes. EWQI ranged from 29.62 to 85.54 (mean 56.25), classifying 36.67% of samples as Good, 53.33% Poor, and 10% Very Poor. Mineral dissolution and water–rock interaction dominated, with secondary influences from evaporation and localized anthropogenic inputs. Inverse Distance Weighting maps suggest localized deterioration, though 30 sites limit certainty. Four ML models—Generalized Linear Model (GLM), Random Forest, Gradient Boosting Machine, and Deep Neural Network (DNN)—approximated EWQI from its 13 component variables. DNN achieved the best test-set agreement (root mean squared error = 0.0100; mean absolute error = 0.0088; mean squared error = 0.0001; coefficient of determination, R 2 = 1); differences among models, including DNN versus GLM, were statistically significant in Friedman and Diebold–Mariano tests ( p < 0.05). SHapley Additive exPlanations (SHAP) identified sulphate and fluoride as most influential inputs, partly expected because both are direct EWQI components. Because the models reproduce EWQI from its own components, they are surrogates rather than independent predictors of water quality, and SHAP rankings do not imply causation. The framework supports hydrochemical assessment in mining-affected watersheds; source attribution requires further geochemical evidence.

Environmental Geochemistry and HealthVol. 48(18)
Indian Institute of Technology Dhanbad (IN)
Water Quality and Pollution Assessment
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
Explainable machine learning and DNN for surrogate modeling of the entropy-weighted water quality index (EWQI) with hydrogeochemical insights in a coal-mining region, India — Ravi prakash Kumar, Prasoon Kumar Singh · Environmental Geochemistry and Health (2026) | TGRS Research Map | TGRS