A geostatistical fusion approach to treat vineyard sample and remote sensing data by deconvoluting kriging

Remote Sensing (RS) data offer a considerable source of information for implementing site-specific management in agriculture. The main limitation to their application lies in their support that may prove too coarse for applications that instead require very fine spatial resolutions. In this work, a multi-step method of data fusion using PLANET satellite data (3 m spatial resolution) is proposed in integration with measurements made on the canopy of the rows of a vineyard of very fine thickness (less than 1 m). The vineyard has an area of 5.3 ha and has rows facing south/south-west. An Italian black grape variety, Sangiovese, is cultivated in it to produce ‘Chianti Classico’. The approach consists in downscaling the different radiometric variables to 1 m-scale by deconvoluting kriging and then in estimating the linear coregionalization model, considering only the directional variograms in the row direction of all (both sample and raster) data. Finally, multi-collocated cokriging was used to produce thematic maps and multi-collocated factor cokriging to partition the vineyard into homogeneous zones. The approach allowed for using satellite imagery at a resolution too coarse for the particular configuration of the vineyard under study and can be applied to any type of row crop.