Unraveling the impacts of GCM input averaging and hydrological model output averaging on runoff projections under...

Averaging GCMs' inputs (InA) and averaging hydrological model outputs (OutA) are two prevalent strategies for runoff projection under climate change. In the InA, the average of multiple GCMs is employed to force a hydrological model, whereas the OutA involves driving the hydrological model with each GCM and subsequently averaging the hydrological outputs. However, the differences in runoff projections between these two strategies remain insufficiently investigation. In this study, we conducted a case study to systematically compare the two strategies. Furthermore, we investigated how emission scenarios and hydrological models influence these differences. To this end, five hydrological models (ABCD, GR2M, SVM, RF, and LSTM) were employed, forced by thirteen GCMs under four emission scenarios. Our findings revealed: (1) the differences between the two strategies are more pronounced for high and low runoff than for mean annual runoff. Mean runoff differences were under 7%, while Q5 exceeded 20% and Q75 around 10% across all scenarios for each model; (2) the differences between the InA and OutA strategies in projecting intra-annual variability are smaller than those in inter-annual variability; (3) differences in projected runoff changes between the two strategies vary with both the emission scenarios and the hydrological model. Mean runoff differences were under 5%, while Q5 ranged from 24.7% to 36.9% and Q75 from 8.0% to 24.2% across all models and scenarios. Overall, substantial discrepancies exist between these two strategies in projecting future runoff. Given the methodological limitations of the InA strategy, we recommend using the OutA strategy for assessing the impacts of climate change on runoff.