Predicting Soil Bulk Density in Boreal Podzolic Soil using Ground-Penetrating Radar and Electromagnetic Induction

Soil compaction is one of the major challenges in sustainable agriculture, primarily due to the use of heavy farming machinery. Tillage and soil compaction influence soil properties, state variables, and processes, ultimately affecting soil health, crop growth, and yield. Traditional methods to estimate soil compaction level, like bulk density (BD) and penetration resistance, are laborious, destructive, time-consuming and provide point-scale measurements only. Near-surface geophysical techniques like Ground-Penetrating Radar (GPR) and Electromagnetic Induction (EMI) are being increasingly utilized to estimate soil properties and state variables in the agricultural landscape since GPR and EMI can address some of drawbacks of traditional methods. However, there is a lack of studies with GPR and EMI examining the BD change associated with tillage and soil compaction. We hypothesize that proxies from GPR and/or EMI can be used to predict BD as an indicator of soil compaction. The objectives were to: 1) evaluate the impact of BD change on dielectric constant (K_r) and direct ground wave amplitude (A) measured from GPR, and apparent electrical conductivity (EC_a) measured by EMI; and 2) assess the predictive capability of GPR and EMI for BD determination. The experiment was conducted on a loamy sand textured soil at a boreal podzolic site in Newfoundland, Canada. Proxy data (i.e., K_r, A and EC_a) were collected using a 500 MHz center frequency GPR system and an EMI sensor representing three compaction treatments (i.e., after tillage, after 4- and 10-time roller passes). Treatment effects and relationships between proxies and the average BD of 0-30 cm soil depth were tested using analysis of variance (ANOVA) and correlation analysis. A Random Forest (RF) regression approach was employed to identify the most significant variables for predicting BD. Subsequently, simple, and multiple linear regression models (LRM) were developed. The accuracy of these LRMs was assessed by comparing predicted and measured BD values. ANOVA results reveal that the measured BD and proxies are significantly different at all three compaction levels. The average BD strongly correlated with soil proxies; K_r(r=0.72), A (r=0.71), and EC_a(r=0.89). Based on RF, EC_a and K_r are the most important variables to predict BD for the studied data set. Therefore, EC_aand K_r were used to develop simple and multiple LRMs. The simple LRM developed with EC_a showed a higher coefficient of determination, R²=0.80, compared to K_r (R²=0.63), while the multiple LRM showed the highest R² (R²=0.83). The model predicted BDs did not deviate from 1:1 line with a root mean square error of <0.14 g/cm³. This study highlights the potential of using GPR and EMI to predict BD non-destructively while covering a larger sample volume. Further research must be conducted to assess the applicability and limitations of this approach under different water contents, electrical conductivities, and soil types.