Assessing the spatial variation of soil and crop properties is the basis for site specific management of crop practices in precision agriculture applications. To this aim, proximal and remote spectral vegetation indices are increasingly replacing soil physical-chemical analysis. In this study the spatial variation of soil properties, proximal and remote spectral vegetation indices were compared in a winter wheat (Triticum aestivum L.) crop grown in a 4.15 ha field in northern Italy. Soil analysis (particle size distribution, pH, carbonates, C, total N, available P, exchangeable cations and electrical conductivity) was geo-referentially carried out; the proximal indices chlorophyll content by N-Tester and normalized difference vegetation index (NDVI) through GreenSeeker were determined in three dates during stem elongation; the remote indices PurePixelTM Chlorophyll Index and PurePixelTM Vegetation Index were determined through the Landsat 8 satellite in three dates during the same wheat stage. Dry biomass yield (DBY), grain yield (GY) and yield components were determined at harvest. Soil, proximal and remote data were submitted to principal component analysis (PCA), and the retained PCs were clustered to delineate areas at low, intermediate and high yield potential, based on soil parameters (CLUsp), proximal (CLUpi), and remote vegetation indices (CLUri). DBY and GY were significantly correlated with several soil parameters and vegetation indices. Spatial distribution of soil and crop data consistently depicted a low performing area (GY < 3 Mg ha-1) and a high performing one (GY > 8 Mg ha-1). CLUsp determined a lower GY difference between low and high performing area (+60%), compared to CLUpi and CLUri (almost + 100%). In CLUsp and CLUpi the low and high performing area were of similar size (25 and 29% for the two respective areas in CLUsp; 25 and 33% in CLUpi), whereas in CLUri they were quite different (16 and 46%). Lastly, yield potential levels determined by vegetation indices (CLUpi and CLUri) exhibited a better degree of agreement with DBY and GY levels, than soil parameters (CLUsp). In exchange for this, the above referred soil parameters are quite consistent in time, allowing soil data to be used for more years. On concluding, PCA followed by clustering resulted in a robust delineation of field areas at different yield potential. This is the premise for developing research driven strategies of practical use.
Precision agriculture; Soil properties; Remote vegetation indices; Wheat; Yield mapping