Unlocking urban soil secrets: machine learning and spectrometry in Berlin's heavy metal pollution study considering...

Area of Science:

• Environmental Science
• Geochemistry
• Data Science

Background:

• Berlin faces significant soil pollution from historical heavy metal (HM) emissions.
• Accurate mapping of HM concentrations is crucial for environmental management and risk assessment.

Purpose of the Study:

• To apply machine learning (ML) techniques for predicting HM concentrations in Berlin's soils.
• To evaluate the impact of incorporating spatial context (land use/land cover) on prediction accuracy.

Main Methods:

• Utilized a dataset of 667 soil samples with spectrometry data for nine HMs (As, Cd, Co, Cr, Cu, Mn, Ni, Pb, Zn).
• Employed four ML algorithms: Partial Least Square Regression (PLSR), Support Vector Machine Regression (SVMR), Random Forest (RF), and Gaussian Process Regression (GPR).
• Stratified soil samples into five land use/land cover (LULC) classes: park, forest, farmland, traffic area, and constructed land.

Main Results:

• SVMR achieved the best overall prediction for Zinc (R² = 0.65) on the full dataset.
• Stratified modeling showed improved performance: PLSR for Nickel in farmland (R² = 0.77) and RF for Nickel in traffic areas (R² = 0.82).
• Incorporating LULC stratification enhanced model accuracy compared to unstratified predictions.

Conclusions:

• Machine learning models are effective tools for predicting soil HM concentrations in urban areas.
• Integrating spatial data, such as LULC, significantly improves the accuracy of HM pollution prediction models.
• This approach provides valuable insights for targeted soil remediation and urban planning in polluted regions.