Scholarlink

<p>Accurate modeling of soil infiltration rates is essential for sustainable water management, flood mitigation, and erosion control. However, traditional empirical models often fall short in capturing the complexity and variability of infiltration processes across diverse soil and climatic conditions<i>. This study addresses this limitation by</i> evaluating the performance of empirical models (Kostiakov, Modified Kostiakov, Horton, Philip, SCS, and a Novel model) and machine learning models (GPR, GBR, XGBoost, CATBoost, LightGBM, RF) using infiltration data from 70 sites in the Mitidja Plain, north of Algeria. Model performance was assessed using RMSE, NRMSE, r, NSE, and KGE. The Novel model outperformed all other empirical models (NSE = 0.921, KGE = 0.843, NRMSE = 0.089), Machine learning models, particularly GPR, demonstrated superior predictive capability (NSE = 0.938, KGE = 0.938, NRMSE = 0.065). Sensitivity analysis identified cumulative time as the most influential factor in infiltration prediction, followed by soil texture and initial water content. These findings highlight the potential of machine learning to capture complex infiltration dynamics more reliably than traditional empirical approaches<i>.</i> The results offer valuable insights for improving hydrological modeling, optimizing water resource management, and supporting sustainable agricultural practices in the Mitidja Plain and similar environments.</p> Graphical Abstract <p>This graphical abstract outlines a workflow for predicting infiltration rates in the Mitidja plain, north of Algeria using empirical and machine learning models. It begins with experimental data collection and data preprocessing using Python (Spyder). Various empirical models (Kostiakov, Modified Kostiakov, Philip, Horton, SCS, and a novel model) and machine learning models (CatBoost, XGBoost, GBR, LightGBM, RF, and GPR) are evaluated. The performance comparison reveals that Gaussian Process Regression (GPR) is the most accurate model, effectively capturing infiltration dynamics. These findings support sustainable water management strategies. GPR’s accuracy supports improved water management strategies.</p> <p></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

Advancing Infiltration Rate Prediction in Algeria’s Mitidja Plain: A Machine Learning and Empirical Model Comparison

  • 期刊论文

  • 发表日期：

    2025年5月9日

查看全文
  • Amina Mazighi,
  • Mohamed Meddi,
  • Hind Meddi,
  • Ishak Abdi,
  • Abdelkader Boucefiane,
  • Aqil Tariq

摘要

Accurate modeling of soil infiltration rates is essential for sustainable water management, flood mitigation, and erosion control. However, traditional empirical models often fall short in capturing the complexity and variability of infiltration processes across diverse soil and climatic conditions. This study addresses this limitation by evaluating the performance of empirical models (Kostiakov, Modified Kostiakov, Horton, Philip, SCS, and a Novel model) and machine learning models (GPR, GBR, XGBoost, CATBoost, LightGBM, RF) using infiltration data from 70 sites in the Mitidja Plain, north of Algeria. Model performance was assessed using RMSE, NRMSE, r, NSE, and KGE. The Novel model outperformed all other empirical models (NSE = 0.921, KGE = 0.843, NRMSE = 0.089), Machine learning models, particularly GPR, demonstrated superior predictive capability (NSE = 0.938, KGE = 0.938, NRMSE = 0.065). Sensitivity analysis identified cumulative time as the most influential factor in infiltration prediction, followed by soil texture and initial water content. These findings highlight the potential of machine learning to capture complex infiltration dynamics more reliably than traditional empirical approaches. The results offer valuable insights for improving hydrological modeling, optimizing water resource management, and supporting sustainable agricultural practices in the Mitidja Plain and similar environments.

Graphical Abstract

This graphical abstract outlines a workflow for predicting infiltration rates in the Mitidja plain, north of Algeria using empirical and machine learning models. It begins with experimental data collection and data preprocessing using Python (Spyder). Various empirical models (Kostiakov, Modified Kostiakov, Philip, Horton, SCS, and a novel model) and machine learning models (CatBoost, XGBoost, GBR, LightGBM, RF, and GPR) are evaluated. The performance comparison reveals that Gaussian Process Regression (GPR) is the most accurate model, effectively capturing infiltration dynamics. These findings support sustainable water management strategies. GPR’s accuracy supports improved water management strategies.