Large-scale soil mapping using multi-configuration EMI and supervised image classification

Abstract

Reliable and high-resolution subsurface characterization beyond the field scale is of great interest for precision agriculture and agro-ecological modelling because the shallow soil (~1–2 m depth) is responsible for the storage of moisture and nutrients that are accessible to crops. This can potentially be achieved with a combination of direct sampling and Electromagnetic Induction (EMI) measurements, which have shown great potential for soil characterization due to their non-invasive nature and high mobility. However, only a few studies have used EMI beyond the field scale because of the challenges associated with a consistent interpretation of EMI data from multiple fields and acquisition days. In this study, we performed a detailed EMI survey of an area of 1 km2 divided in 51 agricultural fields where previous studies showed a clear connection between crop performance and soil properties. In total, nine apparent electrical conductivity (ECa) values were measured at each location with a depth of investigation ranging between 0–0.2 to 0–2.7 m. Based on the combination of ECa maps and available soil maps, an a priori interpretation was performed and four sub-areas with characteristic sediments and ECa were identified. Then, a supervised classification methodology was used to divide the ECa maps into areas with similar soil properties. In a next step, soil profile descriptions to a depth of 2 m were obtained at 100 sampling locations and 552 samples were analyzed for textural characteristics. The combination of the classified map and ground truth data resulted in a 1 m resolution soil map with eighteen units with a typical soil profile and texture information. It was found that the soil profile descriptions and texture of the EMI-based soil classes were significantly different when compared using a two-tailed t-test. Moreover, the high-resolution soil map corresponded well with patterns in crop health obtained from satellite imagery. It was concluded that this novel EMI data processing approach provides a reliable and cost-effective tool to obtain high-resolution soil maps to support precision agriculture and agro-ecological modelling.