Predicting heavy soil metal contents and mapping using machine learning algorithms

Background/Question/Methods

Nowadays, due to the rapid urbanization phenomenon, Human-beings are facing global environmental problems. Soil is the largest pool of the terrestrial ecosystem and plays an important role in various functions such as the production of organisms and preservation of diversity, yet it is losing those functions owing to soil contamination. In an effort to minimize that, environmental impact assessment is conducted and the importance of the soil sector is emphasized. Therefore, this study predicted the stock amount of soil heavy metals according to urban development so that an efficient soil management plans could be established. Based on the national inventory, metadata on soil contamination in the Seoul metropolitan area of South Korea was constructed, as well as the distribution of heavy metal capacity in the soil was mapped using three machine learning models.

Results/Conclusions

We predict the spatial distribution of heavy metals concentrations on the soil with all three models for the Seoul metropolitan area of South Korea. The kNN model achieved the highest predictive performance. The use of kNN model enabled the integration of numerical as well as categorical variables into one prediction approach. In general, the model shows similar patterns of the spatial heavy metals distribution that only differ locally from each other. Among the nine heavy metals, the concentration of As, Cd, Cu, F, Hg, and Zn were found to be higher than the median of 1. It was also confirmed that the high heavy metals concentration value was mainly distributed in Seoul, where urbanization was advanced. However, the values of Ni and Cr6+ were relatively low with the median of 0. In conclusion, the higher the urban development area, the higher the concentration of soil heavy metals, which suggests the importance of the soil sector for environmental impact assessment.