Research on the spatial pattern distribution of soil selenium using machine learning methods integrating geographic proximity in complex terrain

• Published in: Journal of soils and sediments 2024-07, Vol.24 (7), p.2776-2790
• Main Authors: Liu, Xiaoyan, Ma, Qianru, Song, Zhaofen, Ye, Zhicheng, Zhai, Xu, Zhang, Miao, Zhang, Lili, Wang, Qiang
• Format: Article
• Language: English
• Subjects: Aggregation; Agricultural products; Correlation coefficient; Correlation coefficients; Earth and Environmental Science; Effectiveness; Enrichment; Environment; Environmental factors; Environmental Physics; Geographical distribution; Geology; Geostatistics; Industrial development; Land use; Learning algorithms; Machine learning; Paleozoic; Predictions; Rivers; Rocks; Sec 5 • Soil and Landscape Ecology • Research Article; Selenium; Soil; Soil analysis; Soil investigations; Soil Science & Conservation; Soil types; Soils; Spatial distribution; Stratigraphy; Topsoil; Trace elements; Trace elements (nutrients)
• ISSN: 1439-0108; 1614-7480
• DOI: 10.1007/s11368-024-03836-4

Purpose Selenium is an essential trace element that offers various health benefits. However, its uneven distribution results in selenium deficiency in many regions. Here, we aimed to investigate the influence of environmental factors on soil Se concentration and predict the spatial distribution of selenium in topsoil. Methods This study used 327 sample points to compare geostatistics and machine learning models for predicting soil selenium, considering five important conditioning factors including the parent material (geology), biology, topography, climate (MAP and ETA), and soil type using the R platform. We analyzed the relationship between these five factors and soil selenium through statistics and Pearson’s correlation coefficient. Results Based on the R 2 , RMSE, CCC, and σ Zscore , the RFdc model demonstrated the best prediction efficacy, with R 2 and RMSE values of 0.656 and 0.387. The results of spatial distribution predicted by the RFdc model and the relationship between environmental factors and soil selenium revealed that the parent rock significantly influenced the total selenium content. Soil type and land use factors exhibited significant correlations with selenium enrichment as well as topographic factors (slope and elevation) and climatic factors (precipitation). Selenium-rich (> 1.6 mg/kg) aggregation zones in Shitai County tended to be distributed in areas with the black rock system, early Paleozoic stratigraphy, and in forest areas with Alisol and Acrisold soils. Selenium-deficient areas (