Scholarlink

<p>Water movement on the surface and at different depths of soil exhibits distinct behaviors that alter over time, adding to the complexity of water-soil-plant systems. The level of soil moisture content is important for plant survival, which stems from the previously mentioned phenomenon. Due to this, in the present study, six algorithms are proposed to model and predict Substrate Water Content (SWC) through a dataset built from three input factors (volumetric water content, time elapsed from the last irrigation, and porosity). This was achieved with seven Machine Learning models (in a mixed type: neural networks, support vector machines and ensemble learners) and one mathematical (regression) model. Following the creation of the necessary database from various sources, the effective input data for the model is chosen using statistical tests. The cross-validation method is then used to separate and prepare the data into training, validation, and testing phases. To analyze the predicted results of the SWC strategy highlighted in this research, error criteria set by the researchers were also included as part of the ML techniques and graphical tools used during the analysis phase, which included marginal-scatter plots, hybrid violin-box plots, filled error plots, and parallel coordinate diagrams. Estimation of substrate moisture using ML models, in particular, ensemble models. In summary, the XGBoost ensemble model produced the best results, with the lowest Root Mean Square Error (RMSE = 0.009 m3.m-3), the highest Nash-Sutcliffe coefficient (NS = 0.987), and Pearson Correlation Coefficient (PCC = 0.994) with the data.</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

Multi-boosting and machine learning for soil substrate water content prediction

  • 期刊论文

  • 发表日期：

    2025年12月17日

查看全文
  • Marwan Kheimi,
  • Abdollah Ramezani-Charmahineh,
  • Mohammad Zounemat-Kermani

摘要

Water movement on the surface and at different depths of soil exhibits distinct behaviors that alter over time, adding to the complexity of water-soil-plant systems. The level of soil moisture content is important for plant survival, which stems from the previously mentioned phenomenon. Due to this, in the present study, six algorithms are proposed to model and predict Substrate Water Content (SWC) through a dataset built from three input factors (volumetric water content, time elapsed from the last irrigation, and porosity). This was achieved with seven Machine Learning models (in a mixed type: neural networks, support vector machines and ensemble learners) and one mathematical (regression) model. Following the creation of the necessary database from various sources, the effective input data for the model is chosen using statistical tests. The cross-validation method is then used to separate and prepare the data into training, validation, and testing phases. To analyze the predicted results of the SWC strategy highlighted in this research, error criteria set by the researchers were also included as part of the ML techniques and graphical tools used during the analysis phase, which included marginal-scatter plots, hybrid violin-box plots, filled error plots, and parallel coordinate diagrams. Estimation of substrate moisture using ML models, in particular, ensemble models. In summary, the XGBoost ensemble model produced the best results, with the lowest Root Mean Square Error (RMSE = 0.009 m3.m-3), the highest Nash-Sutcliffe coefficient (NS = 0.987), and Pearson Correlation Coefficient (PCC = 0.994) with the data.