Global river discharge projections from 250 years of routed runoff from 20 CMIP6 climate models

Large-ensemble global river flow projections are crucial for assessing future changes in extremes of river discharge in light of internal climate variability, model uncertainty, and anthropogenic climate change. However, river discharge simulations from global hydrology models often consider only a limited number of climate projections, while global climate models usually focus on grid-cell level runoff only.

To bridge this gap, we present a new global river discharge dataset covering 250 years derived by routing runoff from 20 global climate models from CMIP6 along the river network. Specifically, we consider daily runoff from both the historical CMIP6 experiment (1850–2014) as well as the most extreme future scenario (SSP5-8.5, 2015–2099). Routing is computed at a 0.1 degree horizontal resolution using the multi-scale routing model mRM, which implements the kinematic wave equation and is adaptable to a wide range of spatial scales. For the validation of the new dataset, we compare distributional properties of annual maximum (1-day) and annual minimum (7-day) river flow to observations at almost 2000 GRDC-Caravan gauge stations. To this end, we use mean squared error (MSE) decomposition to additionally examine the contribution of different error sources. We find that the squared bias is the most important MSE component at each gauge station for both annual extreme statistics while the shape of the distribution is simulated more accurately. Building on these validated river discharge simulations, we project changes in high, low, and mean flows and evaluate the agreement between the 20 ensemble members. This way, the robustness and range of the projections can be estimated considering uncertainties from both global climate models as well as internal climate variability.