Evaluating <scp>CORDEX</scp> ‐ <scp>CORE</scp> Climate Projections for Simulated Hydrological Processes in a Mesoscale Catchment, South Africa

Abstract

ABSTRACT Understanding future streamflow and changes to hydrological processes is one of the common uses of hydrological models. But the uncertainty of different warming scenarios, boundary conditions and extent of Earth System Models (ESMs) impacts how future conditions might translate into changes in hydrological processes and streamflow. We hypothesise that the coarse spatial resolution of readily available CORDEX‐CORE simulations (25 km) will either under‐ or overestimate streamflow in small, mesoscale, heterogeneous catchments (&lt; 700 km 2 ), relative to hydrological models driven by local station data, due to the inability of projections to capture steep headwater‐valley climate gradients. Additionally, we test whether the selection of ESMs exerts a stronger influence on the sensitivity of hydrological model parameters than the choice of downscaling method. We focus on the near future (2024–2053) because projections for the far future (e.g., 2100) carry even greater uncertainty. To evaluate these hypotheses, we applied the J2000 rainfall‐runoff model to the Eerste River catchment using the Representative Concentration Pathways (RCPs) 2.6 and 8.5 scenarios for the Coupled Model Intercomparison Project Phase 5 (CMIP5) for 2006–2053. The simulations were driven by both CORDEX‐CORE ESMs as well as local climate station data to allow for performance benchmarking. The results of the application highlight the reduced skill of J2000 when driven by projected input data compared to using local station data, particularly in simulating high flows. The most likely cause of this is the inability of the coarse resolution of CORDEX‐CORE (25 km) to capture the vast differences in climate between the headwater and valley within the Eerste, which is spatially variable. Using the CORDEX‐CORE projections as input into the hydrological model, there were, however, decent representations of simulated low flows. The sensitivity analysis showed that J2000 model parameters were more influenced by the selection of the ESMs than downscaling using the Regional Spectral Models, highlighted by more sensitivity to scaling of the air capacity of the soil module. Exploratory analyses of future hydrological conditions suggest a mean annual flow reduction of 10%–22% (Q1–Q3) under RCP 2.6 compared to 8%–26% (Q1–Q3) under RCP 8.5 for the near future (2024–2053). While most (66%) projections suggest that the high flows will likewise be reduced, there is a low percentage (33%) of projections that suggest an increase in high flows under RCP 8.5 for the near future. Many of the projections suggest a further extension of the dry season (66%) and the wet season could be moved by 1–2 months later, compared to the historical conditions where June typically represents the mid‐winter period. The simulated hydrological flows suggest a reduction in baseflow and interflow, but high‐resolution projections (≈5–10 km) are required to improve the agreement in the historical projections compared to observed data.