Probability Maps for Risk Asessment in a Patchy Weed Control

Abstract

Weeds were monitored during April 1996 along a 36 m×36 m grid within a 21 ha winter wheat field to detected whether black bindweed (Fallopia convolvulus L.), cleavers (Galium aparine L.), which are important weeds in cereals, and total weed densities reached economical thresholds for control. The geostatistical method of block kriging with indicator values (1 if below the threshold and 0 otherwise) was applied to estimate probabilities that weed intensity in 324 m2 areas was below the threshold. The probabilities could be used to assess the risk that would occur if those areas were not sprayed with herbicides. Because indicator variograms showed a autocorrelation structure of the data the influence of a reduction of sampling points on the reestimation of the probabilities was tested. The coincidence of the probabilities obtained by the reduced data with those estimated with the complete data set was evaluated by correlation (r) and mean relative error (MRE). The loss of information due to the reduction of sampling points to one half was not substantial. Correlations coefficients were high (black bindweed, 0.78; cleavers, 0.82; total weeds, 0.77) with low MRE's. A satisfactory reestimation with one-fourth of the sample points was not possible.