Mastering the spatial distribution of soil heavy metal content and evaluating the pollution status of soil heavy metals is of great significance for ensuring agricultural production and protecting human health. This study used a machine learning model to study the spatial distribution of soil heavy metal content in a coastal city in eastern China. Having obtained six soil heavy metal contents, including Cr, Cd, Pb, As, Hg, and Ni, environmental variables such as precipitation, soil moisture, and population density were selected. Random forest (RF) was used to model the spatial distribution of soil heavy metal content. The research findings indicate that the RF model demonstrates a robust predictive capability in discerning the spatial distribution of soil heavy metals, and environmental factor variables can explain 60%, 52.3%, 53.5%, 63.1%, 61.2%, and 51.2% of the heavy metal content of Cr, Cd, Pb, As, Hg, and Ni in soil, respectively. Among the chosen environmental variables, precipitation and population density exert notable influences on the predictive outcomes of the model. Specifically, precipitation exhibits the most substantial impact on Cr and Ni, whereas population density emerges as the primary determinant for Cd, Pb, As, and Hg. The RF prediction results show that Cr and Ni in the study area are less affected by human activities, while Cd, Pb, As, and Hg are more affected by human industrial and agricultural production. Research has shown that using RF models for predicting soil heavy metal distributions has certain significance.

Accurately identifying pollution risks and sources is crucial for regional land resource management. This study takes a certain coastal county in eastern China as the object to explore the spatial distribution, pollution risk, and source apportionment of heavy metals in topsoil. A total of 633 samples were collected from the topsoil with a depth ranging from 0 to 20 cm, which came from different topographical and land use types (e.g., farmland, industrial areas, and mining areas), and the concentrations of HMs and As were measured by using atomic fluorescence spectrometry and inductively coupled plasma mass spectrometry. Firstly, the spatial distribution of soil HMs (Cd, Cr, Hg, Ni, and Pb) and arsenic (As) was predicted by incorporating environmental variables strongly affecting soil formation into geostatistical methods and machine learning approaches. Then, various pollution indicators were employed to conduct pollution evaluations, and potential ecological risk assessments were implemented based on the generated soil map. Finally, source apportionment was conducted using random forest (RF), absolute principal component score–multiple linear regression (APCS-MLR), correlation analysis, and spatial distribution of soil HMs and As. Findings in this research reveal that the RF approach yielded the best spatial prediction performance (0.59 ≤ R² ≤ 0.73). The Nemerow and geoaccumulation indices suggest that various pollution levels exist in this area. The average concentrations of As, Hg, and Ni are 7.233 mg/kg, 0.051 mg/kg, and 27.43 mg/kg respectively, being 1.14 times, 1.27 times, and 1.15 times higher than the background levels, respectively. The central–northern region presented a slight potential ecological risk, with Hg and Cd being identified as the primary risk factors. Natural, agricultural, transportation, and industrial and mining activities were identified as the main HMs and As sources. These findings will assist in the design of targeted policies to reduce the risks of HMs and As in urban soil and offer useful guidelines for soil pollution research in similar regions.

Quantifying changes in soil properties greatly benefits our understanding of soil management and sustainable land use, especially in the context of strong anthropogenic activities and climate change. This study investigated the effects of long-term reclamation on soil properties in an artificial oasis region with a cultivation history of more than 50 years. Critical soil properties were measured at 77 sites, and a total of 462 soil samples were collected down to a depth of 1 m, which captures both surface and subsurface processes that are critical for long-term cultivation effects. Thirteen critical soil properties were analyzed, among which four properties—soil organic carbon (SOC), total phosphorus (TP), pH, and ammonium nitrogen (NH₄⁺)—were selected for detailed analysis due to their ecological significance and low intercorrelation. By comparing cultivated soils with nearby desert soils, this study found that semicentennial cultivation led to significant improvements in soil properties, including increased concentrations of SOC, NH₄⁺, and TP, as well as reduced pH throughout the soil profile, indicating improved fertility and reduced alkalinity. Further analysis suggested that environmental factors—including temperature, clay content, evaporation differences between surface and subsurface layers, sparse vegetation cover, cotton root distribution, as well as prolonged irrigation and fertilization—collectively contributed to the enhancement of SOC decomposition and the reduction of soil alkalinity. Furthermore, three-dimensional digital soil mapping was performed to investigate the effects of long-term cultivation on the distributions of soil properties at unvisited sites. The soil depth functions were separately fitted to model the vertical variation in the soil properties, including the exponential function, power function, logarithmic function, and cubic polynomial function, and the parameters were extrapolated to unvisited sites via the quantile regression forest (QRF), boosted regression tree, and multiple linear regression techniques. The QRF technique yielded the best performance for SOC (R² = 0.78 and RMSE = 0.62), TP (R² = 0.79 and RMSE = 0.12), pH (R² = 0.78 and RMSE = 0.10), and NH₄⁺ (R² = 0.71 and RMSE = 0.38). The results showed that depth function coupled with machine learning methods can predict the spatial distribution of soil properties in arid areas efficiently and accurately. These research conclusions will lead to more effective targeted measures and guarantees for local agricultural development and food security.

Quantifying the rates of soil evolution greatly benefits our understanding of soil formation and management, especially in the context of strong anthropogenic activities and climate change. This study investigated soil evolution in an artificial oasis region with a reclamation history of more than 50 years, and critical soil properties were measured at 77 sites. A total of 462 soil samples were collected down to a depth of 1 m. A total of seven critical soil properties were analysed, and four (i.e., soil organic carbon (SOC), total phosphorus (TP), pH, and ammonium nitrogen (NH₄⁺)), which were not closely correlated with each other, were selected for further investigation. Through comparison with desert soils, this investigation found that semicentennial cultivation resulted in significant changes in soil properties, with strong vertical variations, including increases in the C, N and P contents and decreases in pH throughout the whole profile. The temperature, clay content, evaporation rate between the topsoil and subsoil, low vegetation cover, cotton lateral roots, irrigation and fertilization played crucial roles in promoting SOC decomposition and reducing soil alkalinity, thereby contributing to rapid soil evolution. Thus, reclaimed desert soil was scientifically confirmed to be suitable for agricultural use, which will ease the food production crisis, protect the environment, and promote soil evolution. Furthermore, three-dimensional digital soil mapping was performed to investigate the effects of long-term cultivation on the distributions of soil properties at unvisited sites. The soil depth functions were separately fitted to model the vertical variation in the soil properties, including the exponential function, power function, logarithmic function and cubic polynomial function, and the parameters were extrapolated to unvisited sites via the quantile regression forest (QRF), boosted regression tree and multiple linear regression techniques. The QRF technique yielded the best performance for SOC (R²= 0.78 and RMSE = 0.62), TP (R² = 0.79 and RMSE = 0.12), pH (R² = 0.78 and RMSE = 0.10) and NH₄⁺ (R² = 0.71 and RMSE = 0.38). The results showed that depth function coupled with machine learning methods can predict the spatial distribution of soil properties in arid areas efficiently and accurately. These research conclusions will lead to more effective targeted measures and guarantees for local agricultural development and food security.