Artificial neural network (ANN) and multivariable regression (MVR) models are known as two best and most common models to predication and estimation of environmental parameters. One of the most effective neural network modeling is the multilayer perceptron (MLP) that known as a supervised network. Multivariate regression (MVR) model is usually used to find the relevant coefficients in the model equation. Generalization is a big limitation for
MVR models. Soil saturated hydraulic conductivity (Ks) parameter is one of the most widely used soil hydraulic properties for modeling the movement of water in soils and for designing drainage and irrigation water management practices and depend on soil structure, soil texture, chemical properties (EC, pH, CEC, content of ions), organic matter content and bulk density. Measurement of
Ks is relatively time-consuming and become infeasible when hydrologic estimates are needed for large areas.
Minoo Island is an area in the Khuzestan province, in southwestern Iran and is close to the city of Abadan. There are not enough information about
Minoo soils. Aims of this research are including comparison between
ANN and
MVR models, create, modify and select suitable model to estimate
Ks by a few easily measured soil properties and investigation on whether or not application
MVR and
ANN models in study area to estimate
Ks parameter. Sampling operations in order to measurement chemical properties of soil were conducted at 40 points, these properties including content of Na+, Ca2+, Mg2+, EC, pH and CO32-. Also from 20 points were sampled in order to determine content clay of soil. Then
Ks measured in 25 points by Ernst method, because high level of groundwater in
Minoo Island, Ernst method is a suitable method. Three parameters including pH, Na+ and CO32- were selected as desirable variables to use in
MVR and
ANN modeling process. In a comparison between performance of
MVR and
ANN models to predication of
Ks it is clearly has presented that efficiency of
ANN model is better of
MVR model (R2=36 vs R2=15), in other words,
ANN model could be estimate
Ks 2.4 times better in comparison to
MVR model. According to the results, authors not recommended to use of
MVR (as a mathematical model) and
ANN (as an empirical model) models to estimate of Ks. authors believe because of dynamic ecosystem
Minoo Island (recent layers of young soils (Entisol and Inceptisol soil orders), groundwater level oscillating, heterogeneous properties of soil and tidal irrigation), ecosystem of this area is so complex to interpret soil properties of
Minoo Island just by modeling. We propose to use these model just as a complementary tool in addition to field and laboratorial measurements. Finally authors believe because of dynamic ecosystem of study area and unclear boundaries between environmental parameters (soil and water properties of
Minoo Island), a system based on Fuzzy Logic or expert system (with a knowledge database based on experiences of experts who worked or will work in this area) is a better selection as a complementary tool to help future researchers.