Spatial interpolation of cropland soil bulk density by increasing soil samples with filled missing values.

Area of Science:

• Soil Science
• Geospatial Analysis
• Data Science

Background:

• Accurate spatial interpolation of soil properties requires large datasets, but historical data is often incomplete.
• The effectiveness of imputing missing soil bulk density (BD) values for improving spatial interpolation accuracy remains uncertain.

Purpose of the Study:

• To evaluate the accuracy of different models in filling missing soil BD data.
• To assess the impact of using filled BD data on the accuracy of spatial interpolation methods.
• To improve soil property mapping in data-scarce regions.

Main Methods:

• Compared pedotransfer function (PTF), multiple linear regression (MLR), random forest (RF), and radial basis function neural network (RBFNN) for imputing missing BD values.
• Utilized 2,883 soil BD samples from the Sichuan Basin, China, to train and validate models.
• Applied ordinary kriging (OK) and inverse distance weighting (IDW) for spatial interpolation using both original and filled BD data.

Main Results:

• RBFNN demonstrated the highest accuracy in filling missing BD values, significantly improving R² and reducing errors (MAE, MRE, RMSE).
• Incorporating RBFNN-filled BD data reduced the spatial interpolation uncertainty for both OK and IDW methods.
• The methodology effectively enhanced the accuracy of soil BD mapping.

Conclusions:

• The RBFNN model is a reliable method for imputing missing soil BD data, outperforming traditional methods.
• Using imputed data substantially improves the accuracy of spatial interpolation, reducing uncertainty in soil property mapping.
• This approach offers a valuable solution for soil mapping in regions with incomplete historical soil data.