Predicting rice yield losses caused by multispecies weed competition
MetadataShow full item record
Permanent link to this item: http://hdl.handle.net/10568/43970
Monoculture of irrigated rice (Oryza sativa L.) in Latin America has led to serious weed problems and intensive herbicide use. Yield loss prediction enables the economic analysis of weed control, providing a basis for strategic use of herbicides and diversified weed management, but site differences restrict predictions to the environments for which the models are calibrated. We developed an algorithm for predicting rice yield losses based on early assessments of multispecies weed infestations emerging in successive flushes within variable crop stands. Rice was drill-seeded near Palmira, Colombia, at 75,150, 200, 250, and 300 kg ha?1, and grown under nonflood intermittent irrigation for two growing seasons. Five common weeds were allowed to emerge at 15, 24, and 30 days after rice emergence (DAE), when farmers usually decide about early, intermediate, and late herbicide applications. Yield losses were predicted using hyperbolic models with independent variables describing the mixed-weed infestations in terms of density (no. of plants m?2), leaf area index, dry matter m?2, relative density [weed/(weed + rice)], relative leaf area (RLA), and a visual estimate of relative ground cover (RCV). With early weed emergence (15 DAE), weed density accounted poorly (r2 = 0.77) for yield loss. Regression fits improved when RLA (r2 = 0.86) and RCv (r2 = 0.90) were used as independent variables to describe weed infestations in terms of the light apportionment between rice and weed canopies. Measured in terms of RCv, late-emerging (24 and 30 DAE) weeds had lower effects on yield loss variability (r2 = 0.69 and 0.59, respectively). Although subjective, RCv is easier to estimate than RLA, and was the best single variable to describe the competitiveness of a mixedweed infestation. An additional variable was needed only when yield losses were predicted from weed density. Predictions based on RLA were further improved (by 36%) when each species' RLA was measured separately and the model was extended for additive effects of all species. Yield loss predictions using empirical equations cannot be extrapolated widely across different locations; however, these data suggest that meaningful independent variables can strengthen the usefulness of hyperbolic equations for predicting rice yield losses over a range of situations, such as mixtures of weed species at various densities and times of emergence, and different rice seeding rates.