Predicting the relative density from on-the-go horizontal penetrometer measurements at some arable top soils in Northern Switzerland

• Soil density was predicted from penetrometer resistance (PR) and water content.
• A strong correlation was found between soil bulk density and water content.
• Relative density (ρrel) allowed comparison of soil compaction between soils.
• A texture-independent model accurately predicted ρrel from PR and effective stress (σ').

This study sought to develop empirical models to predict soil relative density (ρrel) from measurements of horizontal penetrometer resistance (PR) and soil water content (θg) in a wide range of soil textures. This permits the comparison of the state of soil compactness in different soil textures. It was hypothesised that model coefficients would be texture-dependent when soil compactness was expressed as bulk density (ρd) and that a model with constant coefficients could be obtained when soil compactness was expressed in terms of ρrel (obtained as the ratio of ρd to reference bulk density (ρref)). Field measurements were conducted in 2014 using a horizontal penetrometer at 0.25 m depth in 10 fields in Switzerland with a wide range of soil textures covering sandy loam, silt loam, loam, clay loam and clay (clay concentration, (CC) = 153–585 g kg−1 and organic matter concentration, (OM) = 9–168 g kg−1). At selected locations along the penetrometer measurement transects, cylindrical soil cores were sampled for determination of soil texture, OM, θg and ρd. Soil water potential and effective stress (σ') were also estimated for each location. Standard Proctor tests were performed on eight soils with variable textures. Proctor density was well described as a function of CC and OM (R2adj = 0.97, RMSE = 0.046 Mg m−3) and was used as reference density to obtain ρrel. From this we developed a model for prediction of ρrel from PR and σ′ that allows comparisons between soils without changes in model coefficients. However, σ' cannot be obtained from on-the-go measurements and the model is therefore of limited value for soil compaction mapping. A model for estimating ρrel from PR and θg yielded satisfactory predictions (R2adj = 0.66, RMSE = 3.3%), although θg is a texture-dependent measure of soil water that cannot be compared across soils. Moreover, ρd was well predicted from PR and θg (R2adj = 0.93, RMSE = 0.05 Mg m−3), possibly because all our measurements were carried out at similar soil water potential, which implies that θg carries soil textural information. Future research should test the proposed equations for a wide range of soil water potential values. The findings presented can be of use in developing measurement systems for mapping soil compactness that combine the proposed prediction functions with horizontal penetrometer and water content sensor systems.