Sugar beet root rot loss: ANN and Regression models

Abstract Root rot diseases of sugar beet resulted by Rhizoctonia solani, Pythium aphanidermatum can arouse significant losses in sugar beet crops. Simple and accurate statistical techniques are essential to predict sugar beet yield loss as an industrial crop for agricultural management specialists. The objectives of the present study were to compare: (1) the prediction capabilities of different statistical models at several regions and cultivars, (2) the effectiveness of artificial neural network (ANN) models to predict root rot yield loss, and (3) the effectiveness of simple and multiple linear regression models compare to ANN models. In order to measure crop loss, during two consecutive growing seasons, eleven different sugar beet cultivars were planted in the infected fields in Qazvin, Torbate Jam, Urmia and Kermanshah regions in Iran. To calculate the crop loss, the correlation between Disease incidence (DI) and the important components was considered. The highest correlation (P ≤ 0.01) was observed between DI and Root yield (RY), so we developed different yield loss models. Consequently, the best-fitted simple regression model obtained for RC01 and Urmia with 94.51 % and 90.44 % R-squared, respectively. Then the proper parameters specified in the regression models considered to develop an ANN model using Matlab software to predict the yield loss. Interestingly, obtained ANN models resulted in R2, RMSE of 87.07 %, 5.01 and the lowest AIC vs. 86.78 % and 6.033 for final linear regression, respectively. Our results suggest that, based on the simplicity of the model, simple linear regression is the best model although the artificial neural network models are more accurate for prediction.