Three-dimensional mapping of soil texture using spline depth functions and artificial neural networks

Quantitative, continuous and three-dimensional soil data at appropriate scales are prerequisites for modeling of natural resources and environment. Despite the importance of soil texture, its legacy soil maps are often provided for the surface layers in which vertical and lateral variations of soil properties are not considered. The combination of digital soil mapping (DSM) and soil depth functions is an alternative tool to cope with these problems, especially in countries with limited data such as Iran. Therefore, equal-area spline depth function and DSM techniques were employed to assess the vertical and lateral distribution of soil texture in Silakhor Plain, located in Lorestan province, western Iran. By fitting the depth function to the measured clay, silt and sand percent in 103 sites to a depth of one meter, their estimated percents were obtained at five standard soil depths of Global Soil Map project (0-5, 5-15, 15-30, 30-60 and 60-100 cm). Also artificial neural network model was employed to predict lateral distribution of soil texture fractions using the auxiliary variable derived from satellite image and digital elevation model (DEM) in the standard depths. The results of the sensitivity analysis showed although the relative importance of auxiliary variables in predicting soil texture was different according to the depth and texture fractions, the performance of artificial neural network in upper layers was more than lower layers. R2 values for clay, silt and sand and from the top to the bottom were 0.73 to 0.49, 0.43 to 0.76 and 0.26 to 0.68, respectively. Results also showed, for estimating soil texture, auxiliary variables derived from satellite image were more important in surface layers and of DEM were more important in subsurface layers.