Numerous studies have demonstrated the patchy distribution of weeds within
fields. The majority of these studies have used discrete sampling, recordin
g weed densities at the intersections of regular grids. In this study, Aven
a spp. seedlings were recorded on square grids at four sites. The data were
then divided into test and real data sets using the whole, two-thirds and
one-half of the data to evaluate the consistency of global variogram models
and accuracy of ordinary kriging estimates. Kriging provided poor weed den
sity estimates at both very low and high densities, i.e. data were smoothed
when compared with true values. Grid sampling took considerable time and,
therefore, money to complete, whereas continuous sampling with multispectra
l imagery (performed at one site) was much quicker and at a finer resolutio
n. It is suggested that sampling systems that collect continuous rather tha
n discrete data are currently more appropriate for site-specific weed manag
ement.