Scholarlink

<p>Soil organic carbon (SOC) is a key indicator of soil fertility and environmental resilience. Accurate estimation of the SOC content is necessary for assessing carbon sequestration and the effectiveness of land management practices. This study aimed to develop an effective machine learning-based model for predicting the spatial variability of SOC across Udon Thani Province in Northeast Thailand. This study considered a total of 96 remotely sensed environmental variables along with measured soil data from 60 locations. Four machine learning algorithms: artificial neural network (ANN), boosted regression tree (BRT), Cubist, and random forest (RF) algorithms were applied to develop SOC prediction models. To avoid overfitting, statistically significant variables were identified using Spearman’s rank correlation and Student’s <i>t</i>-test at <i>P</i> &lt; 0.05. The results demonstrated that the significance of the environmental variables differed among the machine learning models. The key variables for the ANN model included Channel Network Base Level (CNBL), Elevation, Valley Depth, Standardized Height, and Topographic Wetness Index. Standardized Height, Valley Depth, and Relative Slope Position were the most significant for the BRT model. CNBL predominantly influenced the Cubist model, whereas the RF model identified Modified Catchment Area, Elevation, Valley Depth, Slope Length, CNBL, and Standardized Height as the most important predictors. Among all the models, the ANN showed the best performance during the testing phase ®<sup>2</sup> = 0.64, normalized root mean square error = 0.06%, mean absolute error = 0.05%). This study highlights the complex SOC-terrain-climate interactions and the need for location-specific management to enhance soil health.</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

Spatial Modeling of Soil Organic Carbon in Northeast Thailand Using Machine Learning Techniques

  • 期刊论文

  • 发表日期：

    2026年7月6日

查看全文
  • Nitiwat Wongchanla,
  • Subhankar Debnath,
  • Jai Govind Singh,
  • Avishek Datta,
  • Sushil Kumar Himanshu

摘要

Soil organic carbon (SOC) is a key indicator of soil fertility and environmental resilience. Accurate estimation of the SOC content is necessary for assessing carbon sequestration and the effectiveness of land management practices. This study aimed to develop an effective machine learning-based model for predicting the spatial variability of SOC across Udon Thani Province in Northeast Thailand. This study considered a total of 96 remotely sensed environmental variables along with measured soil data from 60 locations. Four machine learning algorithms: artificial neural network (ANN), boosted regression tree (BRT), Cubist, and random forest (RF) algorithms were applied to develop SOC prediction models. To avoid overfitting, statistically significant variables were identified using Spearman’s rank correlation and Student’s t-test at P < 0.05. The results demonstrated that the significance of the environmental variables differed among the machine learning models. The key variables for the ANN model included Channel Network Base Level (CNBL), Elevation, Valley Depth, Standardized Height, and Topographic Wetness Index. Standardized Height, Valley Depth, and Relative Slope Position were the most significant for the BRT model. CNBL predominantly influenced the Cubist model, whereas the RF model identified Modified Catchment Area, Elevation, Valley Depth, Slope Length, CNBL, and Standardized Height as the most important predictors. Among all the models, the ANN showed the best performance during the testing phase ®2 = 0.64, normalized root mean square error = 0.06%, mean absolute error = 0.05%). This study highlights the complex SOC-terrain-climate interactions and the need for location-specific management to enhance soil health.