Application of gravity recovery and climate experiment data and ensemble modeling to assess saltwater intrusion in the Miandoab coastal aquifer, Iran, under climate change

Abstract

Study region: The Miandoab aquifer, northwest of Iran, which is located in a sub basin of the Urmia Lake. Study focus: To model the groundwater (GW) quantity and quality, shallow learning (Feed Forward Neural Network (FFNN), Adaptive neuro fuzzy inference system (ANFIS), Support Vector Regression (SVR)), their ensemble and deep learning models were applied. Projections by General Circulation Models (GCMs) for the Shared Socio-economic Pathways (SSP585) scenario, after bias correction, and changes of the model inputs including Normalized Difference Vegetation Index (NDVI), Gravity Recovery and Climate Experiment (GRACE), GRACE Follow-On (GRACE-FO) and GW level (GWL) obtained via Markov Chain model were employed for future climate change projections. To project GW quality (GWQ) parameters for future climate conditions, relationships between GWL and GWQ were established via the Fourier model. New hydrological insights for the region: Results revealed that ensemble learning could outperform individual methods up to 23 %. The Hydro-chemical Facies Evolution (HFE) diagrams for 2050 and 2100 indicated that clusters near the shoreline may exhibit severe declining trend in GWL up to 1.53 m and potential intrusion of saltwater. In the higher altitude lands GWL may exhibit declining trend up to 11.74 m. In addition, HFE diagram indicated that the Ca-Cl water type will be more common in 2050.

Keywords

• Groundwater
• Climate change
• Clustering
• Ensemble learning
• Hydro-chemical facies evolution
• Miandoab