Improved Index Points of Soil Moisture Retention Curve Estimation Using Remote Sensing Data and the Use of Bayesian Networks and Artificial Neural Network

With advances in remote sensing technology, vast efforts have been carried out recently for predicting difficult-to measure soil properties. This study explores the use of information on vegetation cover from satellite images (SAVI) and digital elevation model (DEM) in addition to pedologic attributes to develop pedotransfer functions (PTFs) for estimating three coefficients of soil moisture retention curve (PWP, FC, θs). For this purpose 176 samples from East Azarbyjan and Guilan provinces were collected consisting of 10 various texture classes. Particle size distribution, total porosity, bulk density, organic matter, macro and micro porosity, EC, pH, CCE, geometric mean and standard deviation of the particle diameter, water content at -1 kPa, DEM and SAVI were used as PTFs inputs. Artificial neural networks (ANNs) and Bayesian Networks were used to predict PWP, FC, θs. The performance of the developed PTFs was evaluated using the root mean square error (RMSE) and the MGN test between the observed and the predicted values. Good improvement (based on RMSE) in the PTF’s ability to estimate the three coefficients was achieved with certain input combinations of basic soil properties, topography and vegetation information comparing with using only the basic soil properties as inputs. In comparing Bayesian Network and ANNs method, the results indicated that Bayesian Network estimated the three soil moisture retention curve coefficients more accurately and with greater reliability than the ANNs method.