[Comparison of Prediction Accuracy for the Spatial Distribution of Soil Heavy Metal Chromium （Cr） in a Geologically High-background Area of Eastern Chongqing].

Authors

Zheng-Xin Liu, Si-Xiang Ling, Shui-Ming Zhang, Wei Wei, Jun-Jie Feng, Xiao-Ning Li, Chun-Wei Sun, Xi-Yong Wu

Published in

Huan jing ke xue= Huanjing kexue. Volume 47. Issue 7. Pages 4989-5000. Jul 08, 2026.

Abstract

Accurately predicting the spatial distribution of heavy metals in weathered soils derived from black rock series is essential for ecological development and pollution control in geochemical high-background areas. Due to the complex composition of black rock series parent rocks， traditional interpolation methods often fail to capture spatial heterogeneity effectively， while machine learning （ML） models have demonstrated promising performance in mountainous regions but lack systematic comparison. This study focuses on the spatial prediction of soil chromium （Cr） in the black rock series-dominated high-background region of eastern Chongqing. Twelve environmental variables from six categories-geological， topographic， climatic， soil， anthropogenic， and vegetation-were selected as auxiliary predictors. Four models were constructed and compared： ordinary Kriging （OK）， regularized random forest （RRF）， light gradient boosting machine （LightGBM）， and multilayer perceptron （MLP）. Cross-validation was employed to evaluate model performance and identify the optimal model for predicting the spatial distribution of Cr. The results showed that 25.3% of Cr concentrations exceeded the regional background value， but all remained below the agricultural risk screening and control thresholds， indicating localized Cr enrichment within a manageable range. Among the models， RRF achieved the highest predictive accuracy （R²=0.855， RMSE=4.044）， followed by LightGBM （R²=0.823）， MLP （R²=0.801）， and OK （R²=0.323）， demonstrating the superior performance of RRF. While all models captured similar spatial trends， RRF provided more detailed and accurate identification of high-concentration zones， with LightGBM performing moderately well， MLP showing localized overestimations， and OK exhibiting the weakest fit. Variable importance and SHAP analysis based on the RRF model revealed that Cr distribution was mainly influenced by geological factors （Fe₂O₃， MgO， and K₂O）， soil organic carbon （SOC）， distance to rivers， and annual precipitation. Overall， the RRF-based spatial prediction approach demonstrates significant advantages in geologically complex regions and offers methodological support for heavy metal risk assessment and environmental management in geochemical high-background areas.

PMID:
42473411
Bibliographic data and abstract were imported from PubMed on 20 Jul 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.