Prediction of soil texture using artificial neural networks

Soil texture is one of the most important soil properties that affect many physico-chemical properties such as water storage, cation exchange capacity (CEC), soil fertility and soil ventilation. Today, artificial intelligence technology such as neural and neuro-fuzzy networks is used to solve problems in modeling systems and processes. For this purpose, 150 soil samples from a depth of 0 - 15 cm of Gavshan Dam watershed in the Kurdistan province were collected. The geographic locations, height and slope percent of every sampling point were recorded. The particle size distribution of samples was measured in the laboratory using hydrometer method. The longitude and latitude, height, slope percent and soil texture particles of training points were introduced to artificial neural networks to estimate soil texture particles by MATLAB software. The accuracy of model was evaluated by scoring, using statistical indicators such as root mean square error (RMSE), the ratio of geometric mean error (GMER) and correlation coefficient (R2). According to the results, the values for estimating sand and clay are approximately the same and for predicting the silt, less than sand and clay, and 37.0, although less error. The accuracy and accuracy of the model show that the neural network does not have any accuracy and accuracy in estimating the percentage of soil texture components and the soil texture mapping.