Comparison of Three Geostatistics Methods for Prediction of Soil Texture Classes in Crop and Orchard Lands of Guilan Province

Soil texture is a static soil property that has great effects on soil physico-chemical properties. Therefore, global demands are increasing for a high spatial resolution map of soil texture. Lack of intrinsic soil data can lead to wrong policies regarding management and degradation of soil and water resources. Iran has many scattered soil data that have been collected at great cost. These data can be useful in a wide range of applications if presented accurately in digital map format. In this study, Ordinary Kriging, Pixel-Based Classification (PBC), and Inverse Distance Weighted (IDW) methods were investigated using 4665 soil surface samples collected from croplands and orchards to map Guilan soil texture groups (fine, medium and coarse) and soil mineral particles. MBE, NRMSE, KIA, R2 and Pa statistics were used for verification. The results indicated that IDW could provide higher accuracy for clay (R2 = 0.64 and NRMSE = 0.22) and sand (R2 = 0.67 and NRMSE = 0.25) particles prediction, but PBC had higher accuracy for predicting fine, medium and coarse soil texture groups according to KIA and Pa of 0.46 and 0.73, respectively. However, superiority of PBC was minor (KIA = 0.43 and Pa = 0.71) compared to Ordinary Kriging. PBC used auxiliary soil data as inputs for Artificial Neural Network to predict soil mineral particles of unvisited pixels. For more certainty regarding efficiency of PBC in predicting soil texture groups, it is recommended to test the mentioned methods in areas with more physiographic diversity.