Field partition by proximal and remote sensing data fusion

There is a growing interest in the application of remote and proximal sensing technologies to precision agriculture for the estimation of soil and crop variability. The objective of this research was to jointly analyse data from proximal and remote sensors through geostatistics and a non-parametric clustering approach to delineate homogenous zones. The study was carried out on tomato crop in an experimental field in southern Italy. The field was split into two blocks in order to differentiate the two irrigation treatments: Optimal water supply conditions (OP), Deficit irrigation conditions. The collected multi-sensor data were: 1) bulk electrical conductivity from electromagnetic induction (EMI) sensor, 2) vegetation indices (normalised difference vegetation index, RNIR/RGreen index and normalised difference red edge) calculated from two remote sensing images of WorldView-2 satellite; 3) radiance data from one GeoEye satellite image. Multivariate geostatistics and a clustering approach were applied to the overall multi-sensor dataset reported in the above points 1 and 2, whereas the data of the point 3 were clustered to validate the field delineation. The approach allowed us to integrate the data of the different sensors and to identify three homogenous sub-field areas related to the intrinsic properties of soil and the crop response. The comparison between the previous delineation and the one obtained with GeoEye data, after water treatment differentiation, showed that the plant response was more affected by water management than soil properties. The approach has the potential to define prescription maps in precision agriculture.