Prediction of Saffron Yield based on Soil properties Using Regression and Artificial Neural Networks Models in the Vamenan Region of Golestan Province

Saffron (Crocus sativus L.) is one of the most expensive crop which is grown in restricted areas of the world. Due to its economic values, some farmers, based on similarities of climatic conditions have cultivated it in some regions of country regardless of land capability and suitability, which sometimes the result was not satisfactory. Saffron yield prediction based on soil properties enables us to assess the land suitably for cultivation of this valuable plant. For this purpose, 100 soil samples were collected from Vamenan Saffron fields in Golestan province and the soil chemical and physical properties including the percentage of constituents of the mineral part of soil texture (Sand, Silt, Clay), Phosphorus, potassium, Nitrogen, pH, Electrical Conductivity (EC), Organic matter and Calcium Carbonate Equivalent were measured. In addition, the weight of Saffron wet flower (kg.Ha-1) was measured. In the present study, various combinations of soil properties as input were applied and nine models were developed using artificial neural networks and multiple linear regression models for predicting the saffron yield. Performance of the models was validated using Root Mean Square Error (RMSE), Correlation Coefficient (R) and Geometric Mean of Error Ratio (GMER) methods. The results of the correlation analyses showed phosphorus and organic matter are most effective factors in the production of Saffron. Results showed that performance of the models is much different where R2 value varies from 0.45 to 0.89. Comparing the performance of Saffron yield estimation models indicated the optimal model was obtained from the combination of phosphorous, organic matter, potassium and calcium carbonate equivalent as input and values of R2 and RMSE equal to 0.874 and 0.996 kg.ha-1, respectively.Evaluation of model results indicated that the coefficient varied was obtained from 0.45 to 0.89. The best model in saffron yield estimation was obtained when phosphorous, organic matter, potassium and electrical conductivity were as the input, so that values of R2 and root mean square error (RMSE) were obtained 0.891 and 0.89 kg.ha-1, respectively.