As a strategic crop in Iran's economy, saffron is highly vulnerable to climate change. Analyses indicate that extreme events such as frost during flowering periods and prolonged droughts are key factors in reducing saffron yield. This study aims to analyze trends in extreme temperature and precipitation indices and model their impact on saffron yield using multivariate regression. Data on daily low and high temperatures, as well as rainfall from 1990 to 2020, were collected from weather stations in Torbat-e Heydariyeh and Kashmar to calculate extreme climate measures using the ETCCDI method, which was applied using RClimDex software. Saffron yield data was obtained from the Agricultural Jihad Organization. Trend analysis indicated that annual precipitation has had a significant decline. Extreme temperature indices (TNm, TMm, TXm, WSDI) and heavy precipitation indices (R99p at both stations and R95p in Torbat-e Heydariyeh) exhibited significant decreasing trends (p < 0.05). For Kashmar, the multivariate regression model incorporated four extreme indices (R² = 0.70, RMSE = 0.49, NRMSE = 16.4%). Key predictors included ID (β = -0.14, the strongest yield-reducing factor), extreme low temperatures or TNn (β = +0.10), tropical nights (TR20), and consecutive dry days (CDD) negatively impacted yield. For Torbat-e Heydariyeh, the model demonstrated higher accuracy (R² = 0.83, RMSE = 0.43, NRMSE = 15.9%). Significant predictors were TXn (β = +0.18) and frost days or ID (β = -0.12) as the most positive and negative drivers. Overall, temperature variables—particularly nighttime temperatures—dominated yield variability. These findings provide a foundation for climate-smart saffron cultivation planning in similar semi-arid regions. Mitigation strategies should prioritize thermal regulation (e.g., altitude selection, mulching) and water management to offset warming-induced stress.