Large-basin hydrological response to climate model outputs: uncertainty caused by the internal atmospheric variability

Abstract

Abstract. An approach is proposed to assess hydrological simulation uncertainty originating from internal atmospheric variability. The latter is one of three major factors contributing to the uncertainty of simulated climate change projections (along with so-called "forcing" and "climate model" uncertainties). Importantly, the role of the internal atmospheric variability is the most visible over the spatial–temporal scales of water management in large river basins. The internal atmospheric variability is represented by large ensemble simulations (45 members) with the ECHAM5 atmospheric general circulation model. The ensemble simulations are performed using identical prescribed lower boundary conditions (observed sea surface temperature, SST, and sea ice concentration, SIC, for 1979–2012) and constant external forcing parameters but different initial conditions of the atmosphere. The ensemble of the bias-corrected ECHAM5-outputs as well as ensemble averaged ECHAM5-output are used as the distributed input for ECOMAG and SWAP hydrological models. The corresponding ensembles of runoff hydrographs are calculated for two large rivers of the Arctic basin: the Lena and the Northern Dvina rivers. A number of runoff statistics including the mean and the SD of the annual, monthly and daily runoff, as well as the annual runoff trend are assessed. The uncertainties of runoff statistics caused by the internal atmospheric variability are estimated. It is found that the uncertainty of the mean and SD of the runoff has a distinguished seasonal dependence with maximum during the periods of spring-summer snowmelt and summer-autumn rainfall floods. A noticeable non-linearity of the hydrological models' response to the ensemble ECHAM5 output is found most strongly expressed for the Northern Dvine River basin. It is shown that the averaging over ensemble members effectively filters stochastic term related to internal atmospheric variability. The simulated trends are close to normally distributed around ensemble mean value that indicates that a considerable portion of the observed trend can be externally driven.