A review of uncertainty quantification for dam deformation prediction

Abstract Dams are critical infrastructure for flood control and water resource regulation, and their safe operation is essential to public safety and socioeconomic development. Among various monitoring indicators, deformation is widely regarded as a key metric reflecting structural behavior, making its accurate prediction crucial for structural health assessment. Currently, data-driven deformation prediction models have been widely applied owing to their strong nonlinear fitting capabilities. However, these models typically provide deterministic point predictions only and fail to characterize uncertainties arising from the coupled effects of material deterioration, hydrological and environmental loading disturbances, model approximation, and monitoring noise during long-term operation. This limitation reduces the reliability of deformation predictions and weakens their support for structural safety assessment. Consequently, uncertainty quantification (UQ) has emerged as a critical approach to improving the credibility of deformation predictions. This review analyzes the primary sources of uncertainty in dam deformation prediction. Existing UQ approaches are classified into two main groups, non-Bayesian and Bayesian methods, according to whether uncertainty estimation relies on Bayesian posterior inference. Their methodological characteristics, strengths, and limitations are discussed, and key challenges and future research directions are identified, with the aim of supporting more reliable deformation prediction and structural safety assessment under uncertainty.