An approach for delineating homogeneous zones by using multi-sensor data

Spatial heterogeneity in soil properties has an impact on crop response. There is a growing demand for rapid and non-invasive acquisition of fine-scale information on soil and plant variation for site-specific management. Proximal sensing (Electromagnetic Induction (EMI), Ground Penetrating Radar (GPR), hyperspectral spectroscopy (HS)) and remote sensing (RS) can complement direct sampling. However, sensor data fusion techniques, jointly analysing data from different sources, are still being developed.The objective of this work was to define a multivariate and multi-sensor approach by combining EMI, GPR, RS and HS data, without any previous weighing, in order to differentiate an 1.5-ha arable field into homogenous zones.The multi-sensor data were split into four groups: 1) bulk electrical conductivity (EC) from EMI data, 2) amplitude of GPR signal data, 3) the first principal components relating to five bands (green, yellow, red, rededge, near-infrared (NIR) PCs) of hyperspectral reflectance data and 4) the vegetation indices (NDVI, NDRE and NIR/Green) calculated from the remote sensing image. The data of each group were separately analysed and interpolated at the nodes of a same grid by using cokriging or kriging. To obtain spatially contiguous clusters, a combined approach was used, based on multivariate geostatistics and a non‐parametric density function algorithm of clustering, applied to the overall multi-sensor data set of the estimates.The full approach allowed to identify three homogenous areas. In particular cluster 1, in the NW part of the field, with the lowest values of bulk electrical conductivity and GPR amplitude, and the highest red PC values. The other two clusters were delineated in the SE part of the field, with the highest values of green, yellow, red edge and NIR PCs for cluster 2, and the highest values of bulk electrical conductivity and vegetation indices for cluster 3. The delineation might be related to the intrinsic spatial variability of soil and the health status of plants and be used to produce a prescription map for site-specific management.

► Proximal and remote sensors integrated to delineate homogeneous areas.
► Multi-sensor data integrated using multivariate geostatistics and clustering.
► Clusters reflect intrinsic variability of soil and the health status of plants.
► Homogeneous areas necessary for site-specific management.