Comparison of Digital Mapping Methods for the Thickness of Black Soil Layer of Cultivated Land in Typical Black Soil Area of Songnen Plain

被引：0
|
作者：
Guo J. [1 ,2 ]
Liu F. [1 ,2 ]
Xu S. [1 ,2 ]
Gao L. [1 ]
Zhao Z. [1 ,4 ]
Hu W. [2 ,3 ]
Yu D. [1 ,2 ]
Zhao Y. [1 ,2 ]
机构：
[1] State Key Laboratory of Soil and Sustainable Agriculture, Institute of Soil Science, Chinese Academy of Sciences, Nanjing
[2] University of Chinese Academy of Sciences, Beijing
[3] Key Laboratory of Soil Environment and Pollution Remediation, Institute of Soil Science, Chinese Academy of Sciences, Nanjing
[4] School of Geomatics, Anhui University of Science and Technology, Huainan
来源：
Journal of Geo-Information Science | 2024年 / 26卷 / 06期
基金：
中国国家自然科学基金;
关键词：
digital soil mapping; machine learning; optimal covariates; random forest-regression kriging; spatial differentiation characteristics; stacking generalization model; thickness of black soil layer; variable selection;
D O I：
10.12082/dqxxkx.2024.230682
中图分类号：
学科分类号：
摘要：
The thickness of black soil layer is closely related to the soil fertility and quality of agricultural soils. Accurately describing the spatial distribution of the thickness of the cultivated black soil layer in the typical black soil area of Songnen in Northeast China is of great significance. It contributes to the protection of black soil and promotes the sustainable development of agriculture. However, the commonly used predictive models are difficult to apply when trying to map digital soils in flat areas. How to accurately predict the spatial distribution characteristics of the thickness of black soil layer is an urgent problem that needs to be solved. The typical black soil area of Songnen in Northeast China was selected as the research area. Based on the basic data of 106 profile points and 45 environmental factors in the study area, the variables were screened by factor importance ranking and correlation elimination method. Multiple Linear Regression (MLR), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting (XGBoost), Random Forest-Regression Kriging (RF-RK), and Stacking methods were used to predict the thickness of black soil layer. The predictive accuracy of different models was evaluated and the optimal covariates influencing the spatial distribution of the thickness of black soil layer were studied. Based on the best model, the black soil layer thickness classification map of farmland in the black soil area of northeast China was drawn. Our results showed that the Stacking method combined the advantages of several models, and its prediction performance was the best (R2=0.47, MAE=21.02 cm, RMSE=27.12 cm), followed by RF- RK and RF. After eliminating the environmental variables with low contribution through variable screening, the R2 of different models increased by an average of 0.11, with a maximum increase of 0.32 in MLR. The spatial distribution trend of the thickness of black soil layer predicted by different models was generally consistent. The black soil layer above 60 cm was mainly distributed in the northeast and southeast of the study area, while the black soil layer below 30 cm was mainly distributed in the southwest of the study area. In the plain area, RF- RK and Stacking can be used as effective methods for predicting the thickness of black soil layer. Gross Primary Productivity (GPP), Slope Length and Steepness Factor(LS), and Land Surface Temperature Maximum (LSTm) were the most important explanatory variables in the model. The spatial distribution information of the thickness of black soil layer can provide data support for black soil protection and agricultural sustainable development in the black soil region. © 2024 Science Press. All rights reserved.
引用
收藏
页码：1452 / 1468
页数：16
相关论文
共 62 条
  • [41] He X.L., Yang L., Li A.Q., Et al., Soil organic carbon prediction using phenological parameters and remote sensing variables generated from Sentinel- 2 images[J], Catena, 205, (2021)
  • [42] Peng S.Z., Ding Y.X., Liu W.Z., Et al., 1 km monthly temperature and precipitation dataset for China from 1901 to 2017[J], Earth System Science Data, 11, 4, pp. 1931-1946, (2019)
  • [43] Zhang X., Liu L.Y., Wang Y.J., Et al., A SPECLib-based operational classification approach: A preliminary test on China land cover mapping at 30 m[J], International Journal of Applied Earth Observation and Geoinformation, 71, pp. 83-94, (2018)
  • [44] Wang S.H., Lu H.L., Zhao M.S., Et al., Assessing soil pH in Anhui Province based on different features mining methods combined with generalized boosted regression models[J], Chinese Journal of Applied Ecology, 31, 10, pp. 3509-3517, (2020)
  • [45] Breiman L., Random forests[J], Machine Learning, 45, pp. 5-32, (2001)
  • [46] Liang W.Z., Luo S.Z., Zhao G.Y., Et al., Predicting hard rock pillar stability using GBDT, XGBoost, and LightGBM algorithms[J], Mathematics, 8, 5, (2020)
  • [47] Chen T.Q., Guestrin C., XGBoost: A scalable tree boosting system[C], Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 785-794, (2016)
  • [48] Li X.C., Luo J.H., Jin X.L., Et al., Improving soil thickness estimations based on multiple environmental variables with stacking ensemble methods[J], Remote Sensing, 12, 21, (2020)
  • [49] Karnieli A., Zhang Z.T., Wang H.F., Karnieli A., Et al., Inversion of soil moisture content from hyperspectra based on ridge regression[J], Transactions of the Chinese Society for Agricultural Machinery, 49, 5, pp. 240-248, (2018)
  • [50] Rushing C., Bulusu A., Hurwitz H.I., Et al., A leave-one-out cross-validation SAS macro for the identification of markers associated with survival[J], Computers in Biology and Medicine, 57, pp. 123-129, (2015)
1234567