Muthanna Journal of Engineering and Technology
Volume (14), Issue (4), Year (2026), Pages (165-179)
DOI:10.52113/3/eng/mjet/2026-14-04-/165-179
Research Article By:
Munadhil Ali Hasan
Corresponding author E-mail:munadhilalihasan@gmail.com
ABSTRACT
The conventional deterministic slope-stability analyses of earth dams cannot quantify failure probability arising from natural variability and measurement uncertainty in geotechnical parameters. In this study, a probabilistic framework is devised that combines Monte Carlo simulation (MCS) using limit-equilibrium analysis (GeoStudio SEEP/W-SLOPE/W) and machine-learning surrogate models to assess the reliability indices and failure probabilities for Mosul Dam and Dohuk Dam northern Iraq under their normal functioning rapid drawdown seismic loading and the maximum flood occurrence. Due to the absence of recent field investigations or laboratory tests, we compiled geotechnical parameters from published reports and technical sources. We derived statistical moments from reported ranges using the method of moments. Further, we justified probability distributions from geotechnical reliability literature and available goodness-of-fit checks. Ten thousand MCS realizations were generated for every dam and loading condition. The predictive accuracy for the XGBoost resulted in the highest score for the study with a computed R² = 0.984, MAE = 0.018 and RMSE = 0.024. Furthermore, the reliability update was fully achieved by enabling the model in only 6.5 s. The seismic reliability of Mosul Dam has been found to be critically low with a value of β = 1.95 and Pf = 2.56 × 10⁻². This low value is mainly controlled by variability in foun-dation hydraulic conductivity. Under all scenarios, the Dohuk Dam was still satisfactory but rapid drawdown β reduced from 4.85 to 3.05. The framework enables screening-level, risk-informed monitoring and prioritization of operations, but it calls for site-specific validation prior to design-level decision-making.
Keywords:
Earth dams; Geotechnical uncertainty; Machine learning surrogate models; Monte Carlo simulation; Probabilistic reliability.