Comparative analysis of GAMLSS modeling approaches for nonstationary runoff dynamics in the Yellow River Basin of China

Abstract

Quantifying the driving effects of climate change and human activities on nonstationary runoff dynamics is essential. However, the systematic assessments of nonstationary characteristics and their multiple driving mechanisms at the basin scale remain insufficient. This study compared two Generalized Additive Model for Location, Scale, and Shape (GAMLSS) modeling approaches–Continuous-series modeling (Mode1) and monthly-segmented modeling (Mode2)–to analyze the hydrological nonstationarity characteristics of the Yellow River Basin in China and elucidate the driving mechanisms of runoff processes by incorporating covariates of time, circulation, climate, and human factors. The results indicated that: (1) Taking time as a covariate, Mode2 significantly enhanced model robustness by isolating seasonal dynamics. Mode2 raised the average correlation of the location parameter (μ) with the monthly runoff series to R = 0.82 (range 0.70–0.90), a 30–50% increase over Mode1. Furthermore, the nonstationary standardized runoff index (NSRI) aligned more accurately with actual hydrological fluctuations. (2) Taking circulation indices as covariates, circulation indices (AMO, PDO and NINO3) predominantly governed large-scale hydrological trends, with PDO exerting a more pronounced regulatory effect on extreme events in the downstream region under Mode2. (3) The climate-human composite-driven model had the best fitting effect on runoff, particularly at lower timescales (1-, 3-, and 6-month scales) in the middle and lower reaches, where the interactions among total precipitation (TP), snowmelt (SMT), and soil water storage capacity (SWC) explained over 80% of runoff variations (This model yielded the lowest AICc—15% lower than the single climate model—and the highest explanatory power, with R2 = 0.85 at 1-month, and 0.70 at 3-month scales). This study suggests that Mode2, with its precise characterization of seasonal differentiation and human dynamics, is more suitable for refined water resource management and extreme drought-flood prediction, whereas Mode1 remains efficient for analyzing interdecadal circulation effects. By addressing three key challenges—capturing monthly runoff nonstationarity, integrating multi-factor drivers, and validating runoff simulations—this study greatly improves runoff modeling and drought detection accuracy, laying a scientific foundation for adaptive management under the combined pressures of climate change and human activities.