Performance of Ordinary least Squares (OLS) and Bayesian network (BN) in Exchange sodium percentage prediction based on Sodium adsorption Ratio

Two different criteria are exist in the soil science as indices of Alkality. These are the Sodium Adsorption Ratio (SAR) and the Exchangeable Sodium Percentage (ESP). As shown for measured of ESP, it is essential to have soil Cation Exchange Capacity (CEC). But, For CEC determined using laborious and time consuming tests, it be more appropriate and economical to develop a model that predict ESP indirectly from a easily-measured properties. Researches showed a relationship between ESP and SAR. So, SAR can be allocated to predict of ESP. For this reason, many attempts have been made to predict ESP from soil. The specific goal of the research develop model to determining ESP based on SAR by OLS and BN models for Bonab soils in East Azarbaijan province, Iran. For arrived presented research, 209 soil samples were taken by grid survey (250˟250) of Bonab, Iran. The site is located at mean 1300 m above mean sea level, in semiarid climate in the Northwest of Iran. Then, some soil chemical properties such as Na+, Ca2+, Mg2+, SAR and ESP of the soil samples were measured using laboratory experiments. Then, two model was developed by OLS and BN. OLS estimators are linear functions of the values of the dependent variable which are linearly combined using weights that are a non-linear function of the values of the explanatory variables. So the OLS estimator is respect to how it uses the values of the dependent variable only, and irrespective of how it uses the values of the explanatory. So A Bayesian network is a probabilistic graphical model that represents a set of variables and their conditional dependencies via a directed acyclic graph (DAG). Given symptoms, the network can be used to compute the probabilities of the presence of various diseases. Efficient algorithms can perform inference and learning in Bayesian networks. Generalizations of Bayesian networks that can represent and solve decision problems under uncertainty. The Coefficient of Determination (R2) and Root Mean Square error (RMSE) of the soil ESP-SAR model is reported 0.99, 0.71 and 0.98, 1.63 by OLS and BN respectively. Based on the statistical result, both of soil ESP-SAR model was judged acceptable. T-test were used to compare the soil ESP values predicted using the soil ESP-SAR model with the soil ESP values measured by laboratory tests. The paired samples t-test results indicated that the difference between the soil ESP values predicted by the model and measured by laboratory tests were not statistically significant (P>0.05). Therefore, the soil ESP-SAR model can provide an easy, economic and brief methodology to estimate soil ESP. The GMER index also indicated low estimation of two selected land evaluation method. The results of present study illustrated that OLS and BN models can predict ESP with acceptable limits. OLS and BN are mathematical models between input and output variables and has the ability of modeling between ESP and SAR.