Spatial analysis of early wheat canopy normalized difference vegetative index

Efficient use of real-time canopy sensors requires knowledge of the scale (resolution) of variation in the measured canopy property. Knowing the amount of needed optical data requires estimation of the optimal combination of physical sensor density (number of sensors along the applicator boom) and sensor output density (sensor readings per unit distance along the travel path). The objective of this study was to determine the sampling grid size that would adequately describe field variation in canopy normalized difference vegetative index (NDVI) by varying either physical sensor density or sensor output density.

Wheat (Triticum aestivum L.) canopy NDVI data were collected at Feekes growth stage 3 in five fields in central and western Kentucky in February of 2004 or 2005. Spatial structure of NDVI was characterized by variogram analysis across grid sizes ranging from 0.56 (high-density) to 5.1 m2 and both semivariance and spatial structure parameters for high-density data sets were compared to those obtained with decreasing numbers of sampling points (greater grid size). Nugget, range, and sill values were maintained across evaluated grid sizes in four of five site-years. Correlations between each field`s high-density semivariance values and those for the `low-density` data sets were generally high (1.0 < R2 < 0.8) for all site-years, but there were many cases where intercepts deviated significantly from 0.0 and slopes deviated significantly from 1.0. Observed differences in individual sensor performance did not influence the pattern of NDVI spatial structure. Grid size could be increased from 0.56 to 5.1 m2 without significantly affecting the measured spatial structure of canopy NDVI in most fields. Wheat growers might achieve spatially optimal N applications with lower data resolution and less capital intense machinery.