Simulation of groundwater head using LS-SVM and comparison with ANN & MLR

Nowadays, in order to implement management scenarios, choosing appropriate practical solutions for managing groundwater resources as well as determining the appropriate groundwater harvesting rate requires a simplified aquifer model and its simulation. On the other hand, modeling of groundwater aquifers is very important for simulating and predicting the water level. The first step in groundwater management is the simulation of groundwater level which is followed by its prediction based on the factors affecting the groundwater level. In this study, three models, namely, least-square support vector machine (LS-SVM), multivariate linear regression (MLR) and artificial neural networks (ANN) were used to simulate the groundwater level. The study was carried out on Imam-Zadeh Jafar aquifer in Kohgilouye and Boyerahmad province, Iran. To do this, several input features including groundwater level in previous month, precipitation, temperature, aquifer exploitation, and evaporation in the current month were considered to predict groundwater level at the end of the current month. The target time period was monthly data of 20 years from 1997 to 2016. About 75% of the data were used for model training and the remaining for test. The results revealed that all three models were capable of simulating groundwater level with acceptable performance. Among them, LS-SVM having groundwater level in previous month, aquifer exploitation, and precipitation as input features resulted in the highest accuracy with RMSE, MAPE, and R2 equal to 0.61, 0.069, and 0.99, respectively. These models can be used as alternatives for numerical models to manage and predict groundwater level in groundwater resources management problems.