Weed management through herbicide application in direct-seeded rice and yield modeling by artificial neural network

Abstract

In direct seeded rice (DSR) cultivation, weed is the major constraint mainly due to absence of puddling in field. The yield loss due to weed interference is huge, may be up to 100%. In this perspective, the present experiment was conducted to study the efficacy of selected herbicides, and to predict the rice yield using artificial neural network (ANN) models. The dry weight and density of weeds were recorded at different growth stages and consequently herbicide efficacy was evaluated. Experimental results revealed that pre-emergence (PRE) herbicide effectively controlled the germination of grassy weeds. Application bispyribac-sodium as postemergence (POST) following PRE herbicides (clomazone or pendimethalin) or as tank-mixture with clomazone effectively reduced the density and biomass accumulation of diverse weed flora in DSR. Herbicidal treatments improved the plant height, yield attributes and grain yield (2.7 to 5.5 times) over weedy check. The sensitivity of the best ANN model clearly depicts that the weed control index (WCI) of herbicides was most important than their weed control efficiency (WCE). Besides, the early control of weeds is a better prescription to improve rice yield. Differences in sensitivity values of WCI and WCE across the crop growth stages also suggest that at 15, 30 and 60 days after sowing, herbicides most effectively controlled sedges, broad leaves and grasses, respectively. Based on the grain yield and herbicidal WCE, it can be concluded that the combined application of pendimethalin or clomazone as PRE followed by bispyribac-sodium as POST or tank-mixture of clomazone + bispyribac sodium can effectively control different weed flushes throughout the crop growth period in DSR.