Groundwater level prediction using an improved SVR model integrated with hybrid particle swarm optimization and firefly algorithm

Abstract

The demand for water resources has increased due to rapid increase of metropolitan areas brought on by growth in population and industrialisation. In addition, the groundwater recharge is being afftected by shifting land use pattern caused by urban development. Using precise and trustworthy estimates of groundwater level is vital for the sustainable groundwater resources management in the face of changing climatic circumstances. In this context, machine learning (ML) methods offer a new and promising approach for accurately forecasting long-term changes in the groundwater level (GWL) without computational effort of developing a comprehensive flow model. In order to simulate GWL, five data-driven (DD) models, including the hybridization of support vector regression (SVR) with two optimisation algorithms i.e., firefly algorithm and particle swarm optimisation (FFAPSO), SVR-FFA, SVR-PSO, SVR and Multilayer perception (MLP), have been examined in the present study. Spatial clustering was utilised to choose four observation wells within Cuttack district in order to study and assess the water levels. Six scenarios were created by incorporating numerous variables, such as GWL in the previous months, evapotranspiration, temperature, precipitation, and river discharge. The goal was to identify the variables that were most efficient in predicting GWL. The SVR-FFAPSO model performs best in GWL forecasting for Khuntuni station, according to the quantitative analysis with correlation coefficient (R) = 0.9978, Nash–Sutcliffe efficiency (NSE) = 0.9933, mean absolute error (MAE) = 0.00025 (m), root mean squared error (RMSE) = 0.00775 (m) during the training phase. It is advised that groundwater monitoring network and data collecting system are strengthen in India for ensuring effective modelling of long-term management of groundwater resources.