Scholarlink

<p>Soil water retention curve, or SWRC, is a vital tool in the assessment of soil–water relations, but the determination of SWRC is a very tedious and costly process. To develop an easier approach, this study aims to improve the accuracy of SWRC prediction through the use of advanced machine learning algorithms. Four models, Random Forest (RF), K-Nearest Neighbors (KNN), Kernel Ridge Regression (KRR), and CatBoost with Artificial Ecosystem Optimization (CatBoost-AEO) were used to predict the SWRC using datasets from 108 soil samples collected in central Finland. The Boruta algorithm was applied to the feature selection process, and three input sets were obtained. Three different input scenarios were developed based on different soil characteristics. Quantitative measures that were used to assess the model's performance include the correlation coefficient (R), Root Mean Square Error (RMSE), and Mean Absolute Percent Error (MAPE). Thus, the findings of the current study revealed that the proposed CatBoost-AEO hybrid model performed better than other models. By using optimal inputs, CatBoost-AEO produced an R score of 0.9996 with an RMSE of 0.4400 and MAPE of 2.54% for training sets and attained an R-value of 0.9876 with an RMSE of 2.6515 and MAPE of 17.02% for testing data. SHAP (SHapley Additive exPlanations) values demonstrated that soil suction stood out as the key variable, with a SHAP mean nearly four times greater than the porosity. This study shows that using AEO and SHAP analysis with CatBoost enhances SWRC prediction accuracy.</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

Enhancing forest soil water retention curve prediction accuracy and interpretability through CatBoost, artificial ecosystem optimization, and SHAP analysis

  • 期刊论文

  • 发表日期：

    2026年7月21日

查看全文
  • Hassan Ojaghlou,
  • Masoud Karbasi

摘要

Soil water retention curve, or SWRC, is a vital tool in the assessment of soil–water relations, but the determination of SWRC is a very tedious and costly process. To develop an easier approach, this study aims to improve the accuracy of SWRC prediction through the use of advanced machine learning algorithms. Four models, Random Forest (RF), K-Nearest Neighbors (KNN), Kernel Ridge Regression (KRR), and CatBoost with Artificial Ecosystem Optimization (CatBoost-AEO) were used to predict the SWRC using datasets from 108 soil samples collected in central Finland. The Boruta algorithm was applied to the feature selection process, and three input sets were obtained. Three different input scenarios were developed based on different soil characteristics. Quantitative measures that were used to assess the model's performance include the correlation coefficient (R), Root Mean Square Error (RMSE), and Mean Absolute Percent Error (MAPE). Thus, the findings of the current study revealed that the proposed CatBoost-AEO hybrid model performed better than other models. By using optimal inputs, CatBoost-AEO produced an R score of 0.9996 with an RMSE of 0.4400 and MAPE of 2.54% for training sets and attained an R-value of 0.9876 with an RMSE of 2.6515 and MAPE of 17.02% for testing data. SHAP (SHapley Additive exPlanations) values demonstrated that soil suction stood out as the key variable, with a SHAP mean nearly four times greater than the porosity. This study shows that using AEO and SHAP analysis with CatBoost enhances SWRC prediction accuracy.