Investigating the performance of random forest algorithm in predicting water table fluctuations Compared with two models of decision tree and artificial neural network (Case study: unconfined aquifer of Birjand plain)

Today, due to uncontrolled withdrawal of groundwater resources and declining water table, especially in arid and semi-arid regions, planning and management in the consumption of these valuable resources are of great importance, which requires a study of the behavior of the aquifer in relation to the changes made on it. The purpose of this study is to investigate the efficiency of random forest algorithm in predicting the water table of the unconfined aquifer of Birjand plain and to compare the results with two models of decision tree and artificial neural network. In this regard, first, the input data to the model was collected on a monthly basis during 2010-2011 until 2016-2017 water years, and after checking the trend and removing it, to create the mentioned models, the rattle software package in the statistical software R was used. The results of simulation using the random forest algorithm based on evaluation criteria of R2=0.714, RMSE=0.003 and NS=0.598 (m) show that this algorithm has a relatively high ability to simulate the aquifer water table. Comparing the results of this algorithm with two decision tree and artificial neural network models, it can be seen that the results of the random forest algorithm compared to the decision tree model with R2 = 0.5409, RMSE = 0.0072 and NS = -0.0187 (m) is more consistent with the actual water table of the aquifer and is in line with results of the artificial neural network with R2 = 0.7055, RMSE = 0.003 and NS = 0.6046 (m).