Ground Penetrating Radar (GPR) has been widely studied as a tool for extracting soil parameters relevant to agriculture and horticulture. When combined with Machine-Learning-based (ML) methods, high-resolution Stepped Frequency Countinuous Wave Radar (SFCW) measurements hold the promise to give cost effective access to depth resolved soil parameters, including at root-level depth. In a first step in this direction, we perform an extensive field survey with a tractor mounted SFCW GPR instrument. Using ML data processing we test the GPR instrument's capabilities to predict the apparent electrical conductivity (ECaR) as measured by a simultaneously recording Electromagnetic Induction (EMI) instrument. The large-scale field measurement campaign with 3472 co-registered and geo-located GPR and EMI data samples distributed over ~6600 square meters was performed on a golf course. The selected terrain benefits from a high surface homogeneity, but also features the challenge of only small, and hence hard to discern, variations in the measured soil parameter. Based on the quantitative results we suggest the use of nugget-to-sill ratio as a performance metric for the evaluation of end-to-end ML performance in the agricultural setting and discuss the limiting factors in the multi-sensor regression setting. The code is released as open source and available at https://opensource.silicon-austria.com/xuc/soil-analysis-machine-learning-stepped-frequency-gpr.