Evaluation of CMIP6 GCM Performance in Simulating Historical Rainfall and Temperature Climatology of the Tana and North Gojjam Sub‐Basins in the Upper Blue Nile Basin, Northwestern Ethiopia

The Upper Blue Nile Basin, a vital water tower for northeastern Africa, faces growing climate risks that threaten its water resources and agricultural systems. While CMIP6 General Circulation Models (GCMs) represent the state‐of‐the‐art in climate projection, their performance in tropical highlands remains uncertain due to complex topography and localised climate processes. This study evaluates five CMIP6 models (ACCESS‐ESM1‐5, GFDL‐CM4, GFDL‐ESM4, INM‐CM4‐8, MRI‐ESM2‐0) against observed data (1990–2014) using a multi‐metric framework combining statistical indices (MR, KGE) and extreme climate indicators (RX1day, SU25, CSDI). Results reveal distinct model strengths: GFDL‐ESM4 and INM‐CM4‐8 showed superior rainfall simulation (KGE > 0.6 for daily scales), while ACCESS‐ESM1‐5 exhibited systematic wet biases (> 30% in main rainy season). Temperature simulations demonstrated strong elevation dependence, with GFDL‐ESM4 achieving the lowest errors (RMSE < 2°C) but persistent cold biases at high elevations. The ensemble mean outperformed individual models, reducing precipitation biases by 25%–40% and improving extreme event detection (20% higher skill for R95p heavy rainfall). Three key findings emerge: (1) Model performance varies substantially by temporal scale and elevation zone, necessitating context‐specific selection; (2) Ensemble approaches effectively mitigate individual model biases, particularly for temperature extremes; (3) Current models systematically underestimate rainfall intensity (RX1day errors up to 40 mm) and drought duration (CDD detection < 50% accuracy). These results advance global understanding of GCM limitations in tropical highlands while providing actionable insights for the Nile Basin. The demonstrated elevation‐dependent biases have implications for other mountainous regions, suggesting CMIP7 should prioritise orographic process representation. For practitioners, we establish a transferable framework for model evaluation and recommend ensemble‐weighted projections for climate adaptation planning in data‐scarce highland regions.