learning algorithms

In order to plan and make correct policies in agricultural sector in terms of sustainable application of water resources and ensuring food security and self-sufficiency in the production of strategic products such as wheat, it is necessary to make an accurate prediction of dryland wheat yield. Recently, use of artificial intelligence methods for this purpose has increasingly attracted researchers' attention. In this study, the performance of three ANN algorithms, BR, GD, and LM were evaluated and compared to estimate dryland wheat yield. Meteorological data (2004-2018) from 10 meteorological stations, located in Kermanshah province, were used as input data in the proposed models. After determining the correlations between meteorological parameters and dryland wheat yield, relative humidity (RH) and precipitation (P) variables under three different input data combinations were used in the proposed models. Comparison of the predicted and observed data of dryland wheat yield showed acceptable performance of all three models. The R2 values of training step for the best combination of input data for the ANN algorithms (BR, GD and LM) were 0.85, 0.64 and 0.74, respectively, and the RMSE values were 0.09, 0.16 and 0.18 tons per hectare, respectively. Comparison of the results of different input data combinations showed that the P parameter has the most importance in predicting the yield of dryland Wheat, however, the use of P and RH data simultaneously as the third scenario leads to the highest accuracy. Finally, the BR algorithm by combining the inputs of P and RH with R2 and RMSE values for the test data equal to 0.85 and 0.09 ton/ha respectively, as the optimal model in estimating the drtyland Wheat Yield compared to other algorithms and input combinations were known.