Disentangling spatial patterns and driving factor interactions of urban soil heavy metals using machine learning with multi-source heterogeneous data: A case study in Nanjing, China

Abstract To elucidate the spatial differentiation characteristics and driving mechanisms of heavy metals in soils across the Nanjing urban area, this study integrated four machine-learning (ML) algorithms combined with ordinary kriging (OK), local indicators of spatial association (LISA), SHapley Additive exPlanations (SHAP), and structural equation modeling (SEM). Predictive modeling, spatial mapping, and mechanistic interpretations were conducted using heavy metal concentration data from 229 urban soil samples and multi-source environmental covariates. The results revealed that the coefficients of determination (R²) for the optimal ML models applied to the test set ranged from 0.7221 to 0.9491 across the four heavy metals. Moran’s I values for Cd, Pb, and Cu were 0.275, 0.218, and 0.212, respectively, indicating varying degrees of positive spatial autocorrelation. In contrast, the Moran’s I value for Zn was only 0.048, suggesting a relatively weak overall spatial autocorrelation with pronounced local patchiness. Spatial predictions derived from two ML–OK coupling strategies revealed a complex spatial pattern characterized by multi-element accumulation in the historic urban core area, localized hotspots in construction-disturbed areas in the southwest, anomalous enrichment in northern areas affected by historical mining or industrial activities, and a transition toward lower concentrations in peripheral areas. SHAP analysis indicated that the spatial distributions of Pb and Zn were primarily associated with multiple heavy metal combined contamination, whereas Cu was influenced by anthropogenic inputs and adsorption-mediated immobilization by soil organic matter and minerals. Cd was more sensitive to competitive interactions among multiple heavy metals, hydrological transport, and construction disturbance. The explained spatial variance based on SEM of the four heavy metals ranged from 76.8% to 96.3%, further validating the differences among the various elemental driving pathways.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-13
DOI
https://doi.org/10.21203/rs.3.rs-11163484/v1
Primary Topic
Heavy metals in environment
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Disentangling spatial patterns and driving factor interactions of urban soil heavy metals using machine learning with multi-source heterogeneous data: A case study in Nanjing, China

李仁英, Taoran Shi, Chunfa Wu, Jing Shao et al.
6 links
Research Square
Heavy metals in environment
preprint

Disentangling spatial patterns and driving factor interactions of urban soil heavy metals using machine learning with multi-source heterogeneous data: A case study in Nanjing, China

李仁英, Taoran Shi, Chunfa Wu, Jing Shao, Hanghang Yue, Hong Wang
preprint en

Abstract

Abstract To elucidate the spatial differentiation characteristics and driving mechanisms of heavy metals in soils across the Nanjing urban area, this study integrated four machine-learning (ML) algorithms combined with ordinary kriging (OK), local indicators of spatial association (LISA), SHapley Additive exPlanations (SHAP), and structural equation modeling (SEM). Predictive modeling, spatial mapping, and mechanistic interpretations were conducted using heavy metal concentration data from 229 urban soil samples and multi-source environmental covariates. The results revealed that the coefficients of determination (R²) for the optimal ML models applied to the test set ranged from 0.7221 to 0.9491 across the four heavy metals. Moran’s I values for Cd, Pb, and Cu were 0.275, 0.218, and 0.212, respectively, indicating varying degrees of positive spatial autocorrelation. In contrast, the Moran’s I value for Zn was only 0.048, suggesting a relatively weak overall spatial autocorrelation with pronounced local patchiness. Spatial predictions derived from two ML–OK coupling strategies revealed a complex spatial pattern characterized by multi-element accumulation in the historic urban core area, localized hotspots in construction-disturbed areas in the southwest, anomalous enrichment in northern areas affected by historical mining or industrial activities, and a transition toward lower concentrations in peripheral areas. SHAP analysis indicated that the spatial distributions of Pb and Zn were primarily associated with multiple heavy metal combined contamination, whereas Cu was influenced by anthropogenic inputs and adsorption-mediated immobilization by soil organic matter and minerals. Cd was more sensitive to competitive interactions among multiple heavy metals, hydrological transport, and construction disturbance. The explained spatial variance based on SEM of the four heavy metals ranged from 76.8% to 96.3%, further validating the differences among the various elemental driving pathways.

Research Square
Nanjing University of Information Science and Technology (CN)
National Natural Science Foundation of China
Sustainable cities and communities
Heavy metals in environment
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