Delineation of Management Zones in Precision Agriculture by Integration of Proximal Sensing with Multivariate Geostatistics. Examples of Sensor Data Fusion

Fundamental to the philosophy of Precision Agriculture (PA) is the concept of matching inputs to needs. Recent research in PA has focused on use of Management Zones (MZ) that are fi eld areas characterised by homogeneous attributes in landscape and soil conditions. Proximal sensing (such as Electromagnetic Induction (EMI), Ground Penetrating Radar (GPR) and X-ray fl uorescence) can complement direct sampling and a multisensor platform can enable us to map soil features unambiguously. Several methods of multi-sensor data analysis have been developed to determine the location of subfi eld areas. Modern geostatistical techniques, treating variables as continua in a joint attribute and geographic space, off er the potential to analyse such data eff ectively. Th e objective of the paper is to show the potential of multivariate geostatistics to create MZ in the perspective of PA by integrating fi eld data from diff erent types of sensors, describing two study cases. In particular, in the fi rst case study, cokriging and factorial cokriging were employed to produce thematic maps of soil trace elements and to delineate homogenous zones, respectively. In the second case, a multivariate geostatistical data-fusion technique (multi collocated cokriging) was applied to diff erent geophysical sensor data (GPR and EMI), for stationary estimation of soil water content and for delineating within-fi eld zone with diff erent wetting degree. Th e results have shown that linking sensors of diff erent type improves the overall assessment of soil and sensor data fusion could be eff ectively applied to delineate MZs in Precision Agriculture. However, techniques of data integration are urgently required as a result of the proliferation of data from diff erent sources.