Predicting Yield of Rainfed and Irrigated Barley (Hordeum vulgare L.) in Kermanshah by Artificial Neural Network Approach (Case study Kermanshah, Iran)

  Predicting the yield of strategic crops such as wheat and barley can provide liquidity to purchase products from farmers, suitable space for storage, sufficient pricing, and estimating the import and export needs of the agricultural products. Crop yield forecasting can be estimated using different methods. Artificial neural network is one of the methods recently has been considered and has high accuracy in crop yield estimation. In this approach the relationship between independent and dependent variables as well as their complex interactions will be studied and hidden correlatioins between the variables is discovered. Crop yield, particularly in case of rain-fed crops, severely affected by climatic factors such as rainfall, humidity, temperature fluctuations and solar radiation during the growth season. On this basis, it seems that in case of relationships between climatic data (as independent variables) and crop yield (as dependent variable) it would be possible to predict agricultural production. Considering the importance of forecasting agricultural production, an experiment was conducted in Kermanshah province, with the aim of estimating rainfed and irrigated barley yield using artificial neural network approach,. Barley yield data of 25 years (1991 to 2015) as well as raw and corresponding standardized meteorological data (total annual rainfall, average annual temperature, humidity annual average total sunshine hours, average annual evaporation and the number of frost days) were used as input data networks. To find the best network, performance of different types of neural networks was tested to barley yield evaluation. To evaluate models the statistics indices of correlation coefficient (R), coefficient of determination (R2), mean squared error (MSE) and root mean square error (RMSE) were used. The results showed that the best neural network built for rainfed barley was Modular Neural Network with Momentum learning law. The network had a correlation coefficient of 0.96 and 0.92, and coefficient of determination of 0.85 and 0.92 for raw and standard data, respectively. The best neural network built for irrigated barley was also Modular Neural Network with Momentum learning law. The correlation coefficient for the raw and standard input data was 0.78 and 0.72 with coefficient of determination of 0.60 and 0.51, respectively. The results showed the less efficiency of artificial Neural Network in predicting irrigated compared to rainfed barley yield. Comparison of MSE and RMSE between the models also revealed that networks related to rainfed barely with more accuracy had more correlation coefficient compared to irrigated barely. It seems that the managing operations such as watering in irrigated barley has reduced the effect of climate factors on barley yield. So that the sensitivity of irrigated barley yield was much less than the rainfed barley. In fact, in irrigated cultivations, depending on the amount of available water, crops were irrigated regularly. This would be in addition to supplying the needed water of crops, also reduce the impact of other climatic factors such as temperature. Generally based on the results of this study, the accuracy of Artificial Neural Network to predict the yield of rainfed barley was more acceptable than irrigated barley in Kermanshah province. Rainfed cultivation are more affected by climatic factors such as rainfall and temperature, which would be the reason of the achieved results. Hence, the accuracy of neural network for rainfed barley was more than irrigated barely, which represents more relationship between yield of rainfed barley with climatic factors as inputs of the model. So the sensitivity test of the yield to climatic factors revealed more sensitivity in rainfed than irrigated barley. Moreover, the accuracy of neural network showed that neural network was built for barley than wheat.