Artificial Neural Networks (ANN) are models inspired by biological neural networks which can accurately predict complex and non-linear processes at a desirable level. A survey was conducted in 2014 to compare the potential of ANN and a logistic regression equation (LR) to predict weed presence in 33 dryland winter wheat fields in the province of Kurdistan in Iran. In both models, climatic and soil properties were defined as independent variables and weed abundance as the dependent variable. Field bindweed (Convolvulus arvensis L.) and salsify (Tragopogon graminifolius DC.) were the dominant weeds. Predictions based on the LR model failed to predict accurately the abundance of field bindweed (R2= 0.25) and salsify (R2=0.39) whereas ANN resulted in the best predictions. R2 values ranged from 0.62 to 0.96 for salsify and field bindweed, respectively. Sensitivity analysis revealed that altitude and rainfall were the most significant parameters for modelling weed abundance under dry land cropping conditions. This study demonstrates the potential for ANN as a new tool for the study of weed population dynamics.

Keywords