Evaluation of different empirical models to estimate safflower yield loss from redroot pigweed (Amaranthus retruflexus L.) under water stress conditions

To evaluate yield loss models in safflower–redroot pigweed systems, two field experiments were conducted during the 2007 and 2008 growing seasons. The relative yield loss of the crop was recorded in plots laid out in a split-plot design with three irrigation treatments as main plots [100, 75 and 50% FC (field capacity)] and five weed densities (0, 3, 6, 9 and 12 m^?2) as subplots in three replicates. Among Cousens, Spitters, and Kropff & Lotz models, Spitters was the best predictor of yield loss and showed the lowest RMSE (0.001–0.002), high Adj-R ² (0.824–0.929), highest precision (0.962–0.981) and accuracy (constant and systematic biases of 0.000–0.103 and 0.863–1.040, respectively), and random residuals plot. The Cousens, and Kropff & Lotz models showed relatively good fit, although they had higher RMSE values, larger constant and systematic biases, and nonrandom residual plots. The reliability of all models was influenced by irrigation treatment, with generally larger constant and systematic biases at 50% irrigation, causing yield loss under/overprediction under 100 and 50% irrigation, respectively. The parameters of the Spitters model correctly explained the biological function of water shortage on yield response; however, the parameters of other models failed to represent a biologically acceptable yield response function, particularly at 50% irrigation.