Soybean Yield Prediction Using Artificial Neural Network (ANN) as Function of Nitrogen Fertilizer and Plant Density

Many factors, including climatic conditions, planting date, planting pattern, plant populations, and nutrition can cause a variety of yield through the impacts on the plant. Also, since Iran is located in arid and semi-arid regions, the amount of organic matter in its soils is low, for this reason, it has low levels of nitrogen. Most plants are faced with a lack of nitrogen in these regions, which are compensated through organic and chemical fertilizers; in this regard, nitrogen fertilizers play an important role in plant production.
Also, the increasing demand for agricultural products and problems of access to field data, reveals the necessary of using the appropriate models to predict crop yield.
The aim of this research was to study the effect of nitrogen fertilizer and plant density on yield and yield components of soybean (Variety Gorgan 3) and also prediction this parameter by using the artificial neural network. Two major factors in randomized complete block design were investigated in this research in three replications on soybean (variety Gorgan 3); nitrogen fertilizer in three levels (100, 200 and 300 kg per hectare) and plant density in three levels (100,000, 150,000 and 200,000 plants per hectare). Ten plants were randomly selected from the middle row in each plot to measure traits such as plant height, pods number per plant, pods weight per plant, plant weight, the number of branches and shoot diameter. Data analysis was conducted using SAS software and LSD test as a factorial experiment. For prediction yield and yield components in the artificial neural network, the Levenberg-Marquardt algorithm was used to train the ANN. In order to develop ANN's models, plant density and nitrogen fertilizer were used as input vectors and yield and yield components were used as the output. Shoot height increased by increasing the amount of nitrogen fertilizer and plant density, but increased pod number, plant weight, the number of branches and shoot diameter were a result of increased nitrogen and reduced density. Pods weight increased by reducing the density. Network with 2-20-7 topology could predict the parameters with R2 of 0.99987 and MSE of 0.2497. . Pod weight was significantly higher with the density of 100,000 plants per hectare, while these amounts were similar statistically in 150,000 to 200,000 plants per hectare. Soybean yield is greatly influenced by the weight and pod number, although the pod weight was much more of low density; but this problem may be resolved in a high density due to the larger number of plants. No significant difference was statistically observed in shoot diameter between 100 and 200 kg N per hectare. Accordingly, 100 kg N per ha is suitable for bring down the cost and using fewer fertilizers. Network with 2-20-7 topology had the most performance in soybean yield prediction and had the least performance in a number of branch prediction.