Artificial intelligence approach to estimating rice yield

After wheat, rice is one of the most important agricultural products in the world, and Iran has a special position here with annual production of more than 2 million t of rice. Evaluation of crop yield has an important role in agricultural policy making due to different conditions and restrictions. Estimating rice yield is a key factor in food security. Any change in the effective parameters can cause changes in rice yield and therefore the food security of the population will be affected. In this study, rice crop yield was estimated by artificial neural networks (ANNs) and ANN-genetic programming (GP) in 2011 and 2015. Rainfall, permeability, soil texture, land type, evapotranspiration and inlet and inflow and outflow water to paddy lands were used as inputs. The results showed that the ANN-GP with a root mean square error (RMSE = 80.8 kg ha?(1)) and a correlation coefficient (CC = 0.91) was more accurate than the stand-alone ANN (with RMSE = 139 kg ha?(1) and CC = 0.67). Finally, the effect of each input parameter on rice yield was evaluated. Irrigation, drainage and soil type parameters had the best impact rank, with 36, 28 and 31%, respectively. Therefore, the proposed method can act as an efficient tool in estimating rice yield and help decision makers to manage and develop the agricultural system.

Automated summary

The study estimated rice crop yield in Iran using artificial neural networks (ANNs) and ANN-genetic programming (GP). Results showed that ANN-GP was more accurate than stand-alone ANN, with irrigation, drainage, and soil type parameters having the most significant impact on rice yield at 36%, 28%, and 31% respectively. This proposed method can be an efficient tool for estimating rice yield and aiding decision makers in managing and developing agricultural systems.