Scholarlink

<p>Soil salinization threatens soil health, making accurate spatial monitoring vital for efficient soil resource use and quality improvement. However, because multiple soil-forming elements influence soil properties, soil salinization tends to show firm spatial heterogeneity in some areas with complex environmental covariates, with significant extreme differences in pH and soil salt content (SSC) and a severely skewed distribution. This study aimed to analyze the spatial distribution of soil salinization in Daan City and to identify the most suitable predictive model for study area. The spectral indices of the study area were obtained using Sentinel-2 remote sensing images, and multiple environmental covariates were collected. Two machine learning models, extreme gradient boosting (XGB) and random forest (RF), were employed to predict soil salinization, and their uncertainties were quantified for comparative analysis. The mean values obtained from 50 repetitions of the calculation were used as the final results to evaluate the accuracy and stability of the two models for SSC and pH prediction, effectively eliminating the modeling errors due to the severely skewed distribution of soil data in complex environments, and helping to screen the best thematic models and covariates better. For the prediction of SSC and pH,the coefficient of determination (R<sup>2</sup>) values of the RF model were 0.76 and 0.75, with uncertainties of 2.42&#xa0;g/kg and 0.14, respectively. Similarly, the XGB model R<sup>2</sup> were 0.44 and 0.74, with uncertainties of 3&#xa0;g/kg and 0.16, respectively. The experimental results demonstrated that the RF model was more effective and stable in predicting the distribution of SSC and pH under complex environments. The SSC and pH in the study area were predicted to increase gradually from northwest to southeast, and the salinization distribution was the same as in the previous study. Still, there was an inevitable deterioration in the degree, but the extent and the range of moderate and severe salinization increased. The method proposed in this study covers all the steps of soil salinization and alkalization prediction. It maintains high prediction accuracy under complex soil environments and unclear research on soil salinization and alkalization mechanisms, which has considerable research potential. It lays the foundation for future refined salinization and alkalinity mapping in Daan City. Also, it provides ideas for larger-scale salinization and alkalinity mapping, which will further help local governments and farmers to make decisions in planning and utilizing existing soil and water resources.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predictive Mapping of Soil Salinity and Alkalinity in Complex Pedogenic Environments Utilizing Machine Learning Models

  • 期刊论文

  • 发表日期：

    2026年2月7日

查看全文
  • Jingtian Wang,
  • Qing Huang,
  • Pu Huang,
  • Yuhan Shi

摘要

Soil salinization threatens soil health, making accurate spatial monitoring vital for efficient soil resource use and quality improvement. However, because multiple soil-forming elements influence soil properties, soil salinization tends to show firm spatial heterogeneity in some areas with complex environmental covariates, with significant extreme differences in pH and soil salt content (SSC) and a severely skewed distribution. This study aimed to analyze the spatial distribution of soil salinization in Daan City and to identify the most suitable predictive model for study area. The spectral indices of the study area were obtained using Sentinel-2 remote sensing images, and multiple environmental covariates were collected. Two machine learning models, extreme gradient boosting (XGB) and random forest (RF), were employed to predict soil salinization, and their uncertainties were quantified for comparative analysis. The mean values obtained from 50 repetitions of the calculation were used as the final results to evaluate the accuracy and stability of the two models for SSC and pH prediction, effectively eliminating the modeling errors due to the severely skewed distribution of soil data in complex environments, and helping to screen the best thematic models and covariates better. For the prediction of SSC and pH,the coefficient of determination (R2) values of the RF model were 0.76 and 0.75, with uncertainties of 2.42 g/kg and 0.14, respectively. Similarly, the XGB model R2 were 0.44 and 0.74, with uncertainties of 3 g/kg and 0.16, respectively. The experimental results demonstrated that the RF model was more effective and stable in predicting the distribution of SSC and pH under complex environments. The SSC and pH in the study area were predicted to increase gradually from northwest to southeast, and the salinization distribution was the same as in the previous study. Still, there was an inevitable deterioration in the degree, but the extent and the range of moderate and severe salinization increased. The method proposed in this study covers all the steps of soil salinization and alkalization prediction. It maintains high prediction accuracy under complex soil environments and unclear research on soil salinization and alkalization mechanisms, which has considerable research potential. It lays the foundation for future refined salinization and alkalinity mapping in Daan City. Also, it provides ideas for larger-scale salinization and alkalinity mapping, which will further help local governments and farmers to make decisions in planning and utilizing existing soil and water resources.