Estimating Spatial Variability of Crop Yields Using Satellite Vegetation Indices

The research reports an analysis on satellite vegetation indices (VIs) derived from Landsat 5, 7 and 8 imagery at various crop stages compared to yield monitoring data, to estimate the final yield. Five years georeferenced grain harvest data were archived in a 11.07 ha experimental field located in a flat area in Northern Italy. Multispectral images were downloaded, and various Vis were computed: simple ratio (SR), normalized difference vegetation index (NDVI), soil adjusted vegetation index (SAVI), enhanced vegetation index (EVI), green normalized difference vegetation index (GNDVI), and green chlorophyll index (GCI). Final GY (t/ha) and selected VIs were compared using geostatistics (ArcGIS version 10.3) by computing the empirical semivariograms and interpolating with the spherical model. Dates during crop growth showing highest correlations between VIs and GY were recognized as the best times/crop stages for final GY prediction. Then, geostatistics was applied to the best VIs to assess the correspondence with GY at same pixel level. Three correspondence levels were set, and final agreement was determined based on their comprehensive assessment in all pixels. This study demonstrated a final agreement between VIs and GY in the range 64-86%. Final agreement was shown to be adversely related to its CV, meaning that more consistent correspondence levels were conductive to higher overall agreement, and vice versa. Landsat satellite imagery proved a good potential for estimating final GY over different crops in a rotation, at a relatively small field scale (11.07 ha)