Studies in recent decades have demonstrated that artificial neural networks possess strong capabilities for modeling complex and nonlinear systems. This research investigates the applicability of radial basis function (RBF) and multilayer perceptron (MLP) neural networks for determining land suitability. Data from 300 soil profiles located in a northern agricultural plain were used. The evaluated parameters included soil texture, pH, EC, ESP, CaCO₃ content, gypsum content, soil depth, topography, groundwater depth, and segment depth.
Land suitability classes were initially determined using the FAO classification method. These values were used as target outputs for training neural networks with various structural configurations. Of the total dataset, 69% was used for training and 31% for testing. Next, the trained neural networks were applied to predict land suitability classes using the unseen portion of the data (31%).
A comparison of the two neural network models revealed that the MLP network produces more accurate predictions of land suitability classes than the RBF network. Additionally, the MLP model required less training time compared to the RBF model. These results indicate that MLP neural networks offer superior performance for land suitability estimation.

Login Required

Please login to access the full PDF of this article.

🔑 Author Login