Sign in Subscribe
Machine learning

Estimation of Soil Organic Carbon Stocks Utilizing Machine Learning Algorithms and Multi-source Geospatial Data in Coastal Wetlands of Tianjin and Hebei, China

Strategic Insights

— 2 min read

Researchers in Tangshan, People's Republic of China, have conducted a study on the estimation of soil organic carbon stocks in coastal wetlands using machine learning algorithms and multi-source geospatial data. The study, which utilized 160 soil samples and 35 remote sensing features, found that coastal wetlands are crucial for regulating climate change. The research team, led by Weidong Man from the North China University of Science and Technology, employed Pearson correlation analysis, Boruta, and recursive feature elimination to optimize features and construct a soil organic carbon density prediction model using multivariate adaptive regression splines, extreme gradient boosting, and random forest algorithms. The results showed that random forest had the best prediction accuracy, and that optical features were more important than radar and geo-climatic features in the prediction model.

Key Takeaways:

  • The study utilized 160 soil samples and 35 remote sensing features to estimate soil organic carbon stocks in coastal wetlands of Tianjin and Hebei, China.
  • The research team employed machine learning algorithms, including multivariate adaptive regression splines, extreme gradient boosting, and random forest, to develop a prediction model for soil organic carbon density.
  • The results showed that random forest had the best prediction accuracy, with an R2 value of 0.587, RMSE of 0.798 kg/m2, and MAE of 0.660 kg/m2.
  • The study found that optical features were more important than radar and geo-climatic features in the prediction model.
  • The size of soil organic carbon stocks was related to soil organic carbon density and the area of each wetland type, with aquaculture pond having the highest soil organic carbon stocks.
  • The research has been peer-reviewed and published in Chinese Geographical Science.

Some of the specific names mentioned in the study include: Weidong Man, Rui Yang, Mingyue Liu, Yongbin Zhang, Qingwen Zhang, Caiyao Kou, Xiang Li, Di Tian, Xuan Yin, Jiannan He, Jingfen Tong, Dong Liu, and Yahui Liu.

Statistics:

  • The study utilized 160 soil samples and 35 remote sensing features.
  • The research team employed machine learning algorithms to develop a prediction model for soil organic carbon density, which had an R2 value of 0.587, RMSE of 0.798 kg/m2, and MAE of 0.660 kg/m2.
  • The size of soil organic carbon stocks varied among different wetland types, with aquaculture pond having the highest soil organic carbon stocks.
  • The study found that optical features were more important than radar and geo-climatic features in the prediction model.

Sources:

  • Estimation of Soil Organic Carbon Stocks Utilizing Machine Learning Algorithms and Multi-source Geospatial Data In Coastal Wetlands of Tianjin and Hebei, China. Chinese Geographical Science, 2025;35(4):707-721.
  • North China University of Science and Technology, College of Mining Engineering, Tangshan 063210, People's Republic of China.
  • Weidong Man, Rui Yang, Mingyue Liu, Yongbin Zhang, Qingwen Zhang, Caiyao Kou, Xiang Li, Di Tian, Xuan Yin, Jiannan He, Jingfen Tong, Dong Liu, and Yahui Liu.

Read more