Quantifying the contribution of SWAT modeling and CMIP6 inputting to streamflow prediction uncertainty under climate change

Abstract

Assessing the impacts of climate change on hydrologic systems is essential for establishing adaptive strategies for water resources management, flood risk control, mitigation, and ecological protection. Such assessments are typically performed by obtaining future levels of climate factors such as precipitation and temperature from climate models and coupling them with hydrological models to simulate future hydrological processes. Different algorithm choices for the hydrological cycle module lead to non-negligible uncertainties in the simulation results. Based on a case study of a small watershed located in the upper Huaihe River, China, we develop a SWAT model-based analyzing workflow that integrates two evapotranspiration calculation methods (Haegeraves and Penman-Monteith), two river routing models (Variable Storage and Muskingum), and four calibrated parameters sets. This workflow can identify the uncertainties from algorithms of the internal modules of the hydrological model, model parameterization, and climate model (i.e., General Circulation Models, GCMs, and climate scenarios in Shared Socioeconomic Pathways, SSPs) in the future runoff projections, and quantify their relative contributions. The results show that (1) the monthly runoff will increase in June and August and have little change in other months in 2021–2060, and increase in all months in 2061–2100; (2) annual mean runoff (Qm) and annual maximum runoff (Qp) will increase by 1.8% and 2.6% in the 2040s, and 14.7% and 18.6% in the 2080s; (3) the results of Bayesian model averaging (BMA) analysis indicates the probability of increase in Qm and Qp in all emission scenarios save the SSP 1–2.6 in the 2080s, and especially the probability of increase in Qp will be up to 97%; (4) for Qm, the evapotranspiration calculation method is the main source of uncertainty, accounting for more than half of the total uncertainty; and for Qp, GCM, SSP, and their interaction are the main sources of uncertainty with the contribution of about 70%. In contrast, the uncertainty arising from the parameterization of the hydrological model is small and can be ignored.