Hybrid Machine Learning–Analytical Hierarchy Process Model for Groundwater Recharge Zoning in the Neyshabur Aquifer, Northeastern Iran

Hajian, Farhad

doi:10.22077/jaaq.2026.11587.1146

Hybrid Machine Learning–Analytical Hierarchy Process Model for Groundwater Recharge Zoning in the Neyshabur Aquifer, Northeastern Iran

Document Type : Original Article

Author

farhad hajian

azad university neyshabur

https://doi.org/10.22077/jaaq.2026.11587.1146

Abstract

Declining groundwater levels in semi-arid regions threaten long-term water resource security and agricultural sustainability. In this study, two hybrid machine learning–Analytical Hierarchy Process (ML–AHP) models, including the Random Forest–Analytical Hierarchy Process (RF–AHP) and Gene Expression Programming–Analytical Hierarchy Process (GEP–AHP) models, were developed to map groundwater recharge potential in the Neyshabur aquifer in northeastern Iran. Groundwater level data from 38 piezometers over the period 2000–2018 were used to train both the GEP and RF models, and the importance values of independent variables were extracted and incorporated into the AHP framework to reduce reliance on expert judgment in weight assignment. The results showed that the GEP model performed better in temporal groundwater level prediction, achieving an NSE of 0.407 in the testing phase, while the RF model obtained an NSE of 0.193. In contrast, the RF–AHP model showed superior performance in spatial groundwater recharge zoning, with an AUC of 0.797 compared to 0.629 for the GEP–AHP model. The Topographic Wetness Index (TWI) and wet-season NDVI were identified as the most influential controlling factors of groundwater recharge in both models. The estimated annual recharge was 88.2 × 10⁶ m³ (15.6% of annual precipitation) for the RF–AHP model and 74.0 × 10⁶ m³ (13.1% of annual precipitation) for the GEP–AHP model.

Keywords

Machine learning

Analytical Hierarchy Process

groundwater recharge zoning

Neyshabur aquifer

Subjects