Quantile Delta Mapping Downscaling of CMIP6 Daily Precipitation Under SSP1-2.6 to SSP5-8.5: Model Structure Dominates Projection Uncertainty at a Tropical High-Altitude Station

Daily precipitation projections for tropical high-altitude stations are critical for water resource management and flood risk assessment yet remain limited in data-scarce mountain environments. This study applies a chronology-preserving quantile delta mapping framework to the same four CMIP6 general circulation models under three Shared Socioeconomic Pathways (SSP1-2.6, SSP2-4.5, SSP5-8.5), comparing a historical baseline (1981–2010) with a future period (2051–2080) at a high-altitude tropical station (2160 m a.s.l.) in southern Ecuador. Observations derive from the SC-PREC4SA homogenized daily precipitation product. Pre-correction evaluation across nine performance metrics shows that all raw CMIP6 outputs yield negative Nash-Sutcliffe Efficiency when compared against point-scale observations over complex terrain, consistent with the systematic wet biases intrinsic to coarse-resolution GCM grids. Composite ranking by KGE, NSE, and Pearson r identified MPI-ESM1-2-LR as the best-performing model (r = 0.830). Independent split-sample validation (calibration 1981–2000, evaluation on the withheld 2001–2010 period) confirms out-of-sample skill, with Kling–Gupta efficiency improving from negative values for every raw model to 0.60–0.76 after correction. Ensemble mean annual precipitation changes are modest and directionally mixed (−1.3% to +2.2%), while extreme-event indices show consistent tail intensification: 50-year daily return levels rise from 67.6 mm/day to 71.5–115.2 mm/day across models and scenarios. A two-way variance decomposition shows that inter-model structural uncertainty exceeds scenario uncertainty in every calendar month, averaging 71% of total variance against 5% for scenario choice and 24% for their interaction. A Mann–Whitney comparison pooling bootstrap replicates across models detects a statistically significant difference in return levels between SSP2-4.5 and SSP5-8.5, in contrast with the inconclusive result obtained from the four-model point estimates alone, illustrating the sensitivity of such tests to sample construction in small multi-model ensembles. The fully documented computational pipeline constitutes a transferable methodological framework for probabilistic precipitation projection in data-scarce tropical highland stations across the Andes, directly supporting local water-resource planning, agricultural scheduling, and flood-risk infrastructure design.