COMBINING SPECTROMETRIC AND MORPHOMETRIC DATA WITH MULTIVARIATE AND GEOSTATISTICAL METHODS AND IMAGE CLASSIFICATION FOR-HIGH RESOLUTION DIGITAL SOIL MAPPING

This works aims to assess the potential of soil visible-Near Infrared (vis-NIR) reflectance spectroscopy, representing soil characteristics, and of a detailed digital elevation model (DEM) and derivate morphometric attributes (slope, aspect, curvatures, ...) representing soil forming factors, coupled with multivariate and geostatistical methods and image classification for a rapid and cost-effective realisation of a high-resolution digital soil map (DSM) of a vineyard (7 ha) from southern Italy. Reflectance measurements were carried out in the laboratory, on air-dried and sieved soil samples collected at two depths with reference to a 15x15 m greed. A continuum removal technique was used to normalize the reflectance spectra. The normalized spectra were mean centred. A PCA was performed on the normalized spectra from which the means were subtracted. For each soil layer, the scores of the first two PCs (explaining most of the variance) were selected and the strength of their spatial auto-correlation was assessed. The four PCs data-set were then submitted to the geostatistical analysis and mapped by using ordinary and regression kriging. The reliability of the final PCs raster maps was tested by means of the cross-validation procedure coupled with a set of error indices. The DEM was produced from the processing of data acquired by an Unmanned Aerial Vehicle and processed for the production of morphometric maps (MM). All PCs maps, DTM and MM were classified using k-means clustering coupled with silhouette method to produce a DSM. Each digital map unit was characterized through the description and analysis of a representative soil profile. The boundaries of the DSM units were successfully compared with those of a conventional soil map. The results achieved demonstrated the potential of the approach followed for a low cost, rapid and effective DSM in the specific pedo-environmental condition of the study area.