Incorporation of weed spatial variability into the weed control decision-making process

Floral surveys were carried out on a field of 28 m × 100 m on the nodes of a regular 2 m × 2 m grid, using a rectangular sampling area of 25 cm × 30 cm. In total, 765 units were sampled, each one characterized by the spatial co-ordinates and the number of seedlings of different weed species. The spatial representation of the weeds was obtained with kriging. Simulations were carried out for Amaranthus spp., which had the highest frequency and density (221 plants m^?2), and Portulaca oleracea L., a species that combined a more aggregated distribution with a medium–high density (27 plants m^?2). The results obtained clearly indicated that the usefulness of geostatistical procedures depends on the type of question posed by the user. If the goal is to estimate weed density and, consequently, crop yield loss, kriging appears to overburden the decision-making process, without improving the estimates obtained. This procedure becomes useful for obtaining weed infestation maps to be used for intermittent spraying applications. The reliability of these maps increases with the number of samples used for kriging. With the more aggregated species, at least 50 samples are required to obtain an infestation map. The reduction in the area to be treated depends on the threshold level adopted and on the number of samples used for kriging. With a threshold around the break-even point for most post-emergence treatments, this reduction varies from 10% to 40% with maps obtained from 50 and 175 samples respectively. The usefulness of infestation maps obtained with kriging for improving the decision-making process is strictly dependent on the weed patch dynamics: if these patches remain relatively stable over time, kriging can be carried out periodically without overburdening the decision-making process, whereas, if they are not stable, maps need to be drawn up each year, with a significant increase in costs.