Anthropogenic climate change, driven by the escalating concentration of greenhouse gases, has posed fundamental and unprecedented challenges to both natural and human systems in recent decades (IPCC, 2021). Among the most pronounced consequences of global warming are shifts in the frequency, intensity, spatial distribution, and timing of extreme hydro-meteorological events, particularly precipitation. As the moisture-holding capacity of the atmosphere increases by approximately 7% per 1°C of warming—governed by the Clausius-Clapeyron relationship—numerous studies warn that spatial and seasonal fluctuations in precipitation regimes will inevitably intensify. Addressing this critical challenge, the present study investigates precipitation dynamics across Iran during the historical period of 1980–2014. The primary methodological innovation lies in first evaluating the performance of individual downscaled NEX-GDDP-CMIP6 models, followed by the development of a non-linear, exponentially weighted multi-model ensemble. Fundamentally, this research demonstrates how formulating a non-linear ensemble framework—one that accounts for the spatial variance in modeling errors by assigning geographically dependent weights—can substantially enhance the accuracy of precipitation simulations. We also examine how individual models perform against this proposed framework. Rigorous assessment of the capacity of NEX-GDDP-CMIP6 datasets alongside the proposed exponential ensemble (Exp-Ensemble) reveals substantial limitations in the individual models' ability to reliably capture precipitation patterns, particularly over complex mountainous terrain and coastal margins. When evaluated against GPCC observations, these individual models exhibit a pervasive dry bias of -25 to -50 mm yr⁻¹, coupled with Root Mean Square Error (RMSE) values of 127 to 155 mm yr⁻¹. Notably, GFDL and EC-Earth3 display anomalous deviations in precipitation variance; consequently, their spatial outputs across Iran's diverse topography exhibit high spatial uncertainty. Implementing the exponentially weighted multi-model ensemble strategy yielded a statistically significant reduction in these systematic errors. By calibrating grid-cell weights based on each model's historical skill across distinct climate zones, the overall RMSE was reduced to 108 mm yr⁻¹, while the dry bias decreased to -23 mm yr⁻¹. Despite these improvements, standard Bias-Correction Spatial Disaggregation (BCSD) algorithms continue to exhibit limitations. They fail to fully compensate for the unresolved sub-grid-scale processes in model outputs along the southern coastlines of the Caspian Sea. Intense precipitation maxima exceeding 1000 mm yr⁻¹ within this narrow coastal strip are systematically underestimated, with a peak of only 600 mm yr⁻¹ even within the integrated ensemble framework. This pronounced discrepancy highlights a simulated deficit exceeding 50% for this specific geographic sub-region. Ultimately, the arid and semi-arid expanse of Iran represents a formidable challenge for parameterizing precipitation schemes within the sixth phase of the Coupled Model Intercomparison Project (CMIP6). While it cannot eliminate the underlying dry bias ubiquitous across this latest generation of models, the weighted ensemble proposed herein successfully minimizes error variance. Integrating this methodology into broader climate impact evaluations and macro-level water resource policymaking effectively mitigates structural uncertainties, thereby establishing a rigorous baseline for future hydrological projections in Iran and geographically analogous regions worldwide.