Artificial Intelligence and Qanat Indigenous Strategy for Adapting to Climate Change (Case Study of Evaluating the Efficiency of XGBoost in Estimation of Qain Qanat Discharge)

Fouladi Nasrabad, Mohamad; Farzaneh, Mohammad Reza; Zeraati Neyshabouri, Sepideh

doi:10.22034/judpm.2026.579182.1096

Artificial Intelligence and Qanat Indigenous Strategy for Adapting to Climate Change (Case Study of Evaluating the Efficiency of XGBoost in Estimation of Qain Qanat Discharge)

Articles in Press

Document Type : Original Article

Authors

Mohamad Fouladi Nasrabad ¹

mohammad reza farzaneh ²

Sepideh zeraati neyshabouri ³

¹ Ph.D. Student, Water Resources, Department of Water Science and Engineering, University of Birjand, Iran.

² RCESD

³ Ph.D. , Water Resources, Department of Water Science and Engineering, University of Birjand, Iran

10.22034/judpm.2026.579182.1096

Abstract

Climate change has fundamentally altered the hydrological balance in arid regions, necessitating robust predictive tools to inform urban adaptation strategies. Qanats, as vital strategic water infrastructures in Iran’s arid zones, face complex management challenges due to climate-induced groundwater depletion, highlighting the need for AI-driven decision-support systems. This study evaluates the performance of the XGBoost machine learning model in predicting qanat discharge in the Qaen Plain as a tool for evidence-based water resource management. The dataset comprised 12 years of discharge records (2007–2018), alongside climatic variables and topographical elevation. The data were partitioned into training (70%) and testing (30%) sets, with hyperparameters optimized via RandomizedSearchCV targeting RMSE minimization. The results demonstrate that the XGBoost model exhibits high reliability for policy-making, achieving an $R^2 \approx 0.88$ and a Nash-Sutcliffe efficiency of $NS \approx 0.76$. Sensitivity analysis identified the optimal configuration as n_estimators=500, max_depth=6, learning_rate=0.08, and min_child_weight=4. However, the model showed limitations in predicting peak discharge (exceeding 20 L/s), likely due to the RMSE loss function and data imbalance. These findings suggest that while machine learning significantly enhances the predictability of water resources for smart governance, policy frameworks must account for predictive uncertainties in extreme flow scenarios to ensure sustainable climate adaptation and infrastructural resilience.

Keywords

Machine Learning in HydrologyوClimate Change Adaptation

Sensitivity Analysis

Qanat Management

XGBoost Modeling

Subjects