A Review on Nonstationary Flood Frequency Analysis under Climate and Catchment Change

Abstract

Flood frequency analysis has traditionally relied on the assumption that hydrological extremes fluctuate within a stationary probabilistic regime, allowing past observations to serve as a basis for estimating future flood risk. That assumption has become increasingly questionable in many river basins affected by climate change, land-use transformation, urbanization, river regulation, and evolving hydroclimatic variability. In response, nonstationary flood frequency analysis has emerged as a major research area in modern hydrology. This review critically examines the conceptual basis, statistical frameworks, covariate strategies, uncertainty sources, and practical implications of nonstationary flood frequency analysis. It discusses the hydrological mechanisms that generate nonstationarity, evaluates methodological developments including time-varying extreme value models, generalized additive models for location, scale and shape, Bayesian approaches, and regional methods, and assesses the opportunities and limitations of machine learning and hybrid frameworks. The review argues that nonstationary analysis should not be treated as a universal replacement for stationary frequency methods, but as a context-dependent approach that must be justified by hydrological evidence, physically meaningful covariates, and decision relevance. Particular attention is given to uncertainty, because flexible nonstationary models often improve in-sample fit while weakening extrapolative reliability. The review concludes that future progress depends on stronger process-statistics integration, more disciplined covariate selection, clearer reporting standards, and a shift from fixed return periods toward design-life risk and reliability-based planning. These directions are essential if nonstationary flood frequency analysis is to become a robust tool for hydrologic design and flood risk management.