Apparent electrical conductivity predicts physical properties of coarse soils

Precision agriculture informed by electromagnetic induction surveys could reduce groundwater withdrawals and nitrogen leaching from coarse soils. However, coarse, nonsaline soils often have extremely narrow ranges of mapped apparent electrical conductivity (ECa) and the efficacy of ECa for predicting soil physical properties is uncertain in this context. For this reason, it is also uncertain as to whether electromagnetic induction surveys are valuable for guiding precision agriculture on coarse, nonsaline soils. Additionally, the need to ground-truth electromagnetic induction surveys for individual agricultural fields with soil sampling and statistical model development hampers adoption of precision agriculture at the regional scale. Our research objectives were to quantify the variation in mapped ECa and develop statistical relationships between ECa and soil physical properties both within and across several agricultural fields in the Wisconsin Central Sands, a distinct hydropedological region with coarse, glaciolacustrine soils. We used nonparametric correlation analyses to identify associations and quantile regression, a statistical approach with no assumptions of normality or homoscedasticity, to identify predictive relationships between ECa and soil physical properties. We found strong, significant (p < 0.05) correlative and predictive relationships between ECa and topsoil (0-0.3 m) particle size fraction, organic matter content, and field capacity within and across several fields. Yet, we did not observe many significant relationships between ECa and subsoil (0.5-0.6 m) physical properties, which we attribute to heterogeneous soil layering and the low depth resolution of our soil sampling approach. Our findings demonstrate that proximal sensing of ECa can identify intrafield variability in soil properties under extremely narrow observed ECa ranges (0-11 mS m−1). Moreover, we found that interfield quantile regression models predicted soil physical properties across several agroecosystems. Heteroscedasticity was present in interfield ECa relationships with physical properties, which resulted in the need for different quantile regression models across the conditional distribution. The flexibility for accommodating heteroscedasticity in soils and simplicity of modeled functions make quantile regression a promising approach for developing interfield or regional models of ECa to predict soil physical properties in distinct, hydropedological regions with coarse soils.