Prediction of Transmissivity of Malikan Plain Aquifer Using Random Forest Method

Transmissivity is an important factor in identifying the characteristics of aquifers, so, estimating its value and distribution by modeling is necessary for aquifer management. Estimation of this parameter using field experiments such as pumping test is costly and time-consuming. For suitable management of Malikan plain, as one of active agricultureal areas in north-west of the country, understanding the hydrogeological parameters, such as transmissivity is required. In this study the random forest (RF) algorithm, which is a learning method based on ensemble of decision trees, is proposed for predicting transmissivity that has not been used in this field, yet. The RF technique has advantages over other methods due to having high prediction accuracy, ability to learn nonlinear relationships, non-parametric natural and ability to determine the important variables in the prediction process. Increasing the number of trees decrease the error, so 500 trees were selected to reap high efficiency of the model. The model results were evaluated by OOB error estimating method and in addition, to reduce the dimensions, increase the accuracy and better interpretation of the model process, the FS method was used. The most important variables in the prediction were also identified by the FS method. Based on the results of RF modeling with AUC=0.96 and MSE=0.036, electrical conductivity, aquifer media and hydraulic gradient variables were the most important parameters in predicting transmissivity, respectively. Also the accuracy of RF model and determining the important parameters in transmissivity prediction showed the advantages of this model over other models in prediction issue.