his study was carried out in 2013 and 2014 to compare the potential of artificial neural networks and logistic regression to predict dominant weed presence on dryland chickpea and winter wheat fields in Kurdistan province, Iran. In both models, climatic and soil characteristics were defined as independent variables and presence/absence of the dominant weeds as the dependent variable. The geographical coordinates of each field was overlaid on georeferenced map of the province for producing the distribution of weed species maps in ArcGIS. Also, the zonation maps developed by using GIS based on LR models. Demographic indices of weed species were calculated, and the dominant weeds were determined. In the area under study, 61 and 74 weed species were identified on chickpea and winter wheat fields, respectively. The results indicated that Galium aparine L., Convolvulus arvensis L., Scandix pectin-veneris L. and Tragopogon graminifolius DC. at three-leaf stage (99, 81, 71 and 70, respectively), Convolvulus arvensis and Tragopogon graminifolius at podding stage of chickpea (96 and 77, respectively); and Convolvulus arvensis, Tragopogon graminifolius, Turgenia latifolia (L.) Hoffm. and Carthamus oxyacantha M. B. at heading stage of winter wheat (95, 80, 78 and 72, respectively) were the dominant weeds with the highest abundance indices. The logit models did not show good fitness and could not fit any models for Galium aparine at three leaf stage and dominant weeds at podding stage of chickpea. However, ANN could develop the best suited models for prediction all dominant weeds with high MSE values. Sensitivity analysis on the optimal networks revealed that altitude and rainfall were the most significant parameters. The results demonstrates the potential of ANN as a promising tool for survey of weed population dynamics.

Keywords