Estimation of soil quality indices and its uncertainty using Bootstrap-based Artificial Neural Networks (BANNs)

In this study the slope of soil water retention curve at its inflection point(Si) as a soil physical quality index and its correlation with soil convenient properties and with information on vegetation cover from satellite images(SAVI) and digital elevation model (DEM) were studied. For this purpose, 176 disturbed and undisturbed soil samples were collected from East Azarbaijan and Gilan provinces. The test sites were chosen as such to provide wide variety in terrain, land use characteristics, vegetation, soil types and soil distribution patterns. Particle size distribution, total porosity, bulk density, organic matter, EC, pH, CCE, mean weight diameter(MWD), geometric mean and standard deviation of particle diameter, water content at -30 kPa, DEM and SAVI were used as pedotransfer function (PTFs) inputs. Since reliable hydrologic prediction is essential for planning, developing and rational management of the soils, therefore, in this study the uncertainty involved in Si prediction using artificial neural network (ANN) models was quantified. The uncertainty associated with Si was investigated using the bootstrap based artificial neural networks (BANNs). The performance of PTFs was evaluated using the root mean square error (RMSE) between the observed and the predicted values and the Morgan-granger-newbold test ( MGN). Although variability exists within bootstrapped replications, improvements were achieved with certain input combinations of basic soil properties, topography and vegetation information compared with using only the basic soil properties as inputs.