Alpine grasslands on the Qinghai-Tibetan Plateau, a critical carbon sink, face escalating water stress under climate change, yet the thresholds of compound soil-atmospheric drought triggering autumn phenology shifts remain unknown. This study identified critical soil moisture and vapor pressure deficit thresholds regulating grassland autumn phenology in different water stress regions and investigated changes in grassland autumn phenology sensitivity from a threshold-based perspective using piecewise linear regression and sensitivity analysis. The grassland autumn phenology was delayed in 53% (significant delay: 13%) of the study area from 2001 to 2020, while advanced autumn phenology accounted for 47% of the grassland zone, with 5% showing significant advancement. Considering the different combinations of soil moisture and vapor pressure deficit change trends during 2001−2020, we retrieved four water stress types across the Qinghai-Tibetan Plateau grassland. The differences in soil moisture and vapor pressure deficit thresholds were both statistically significant (p < 0.001) across the four water stress type zones. We quantified critical soil moisture (0.21–0.24 m³·m⁻³) and vapor pressure deficit (0.23−0.29 kPa) thresholds across four water stress zones. In different water stress type regions, the maximum changes in sensitivity of grassland autumn phenology exhibited 1.63 days·10⁻²·(m³·m⁻³)⁻¹ (to soil moisture) and −2.18 days·10⁻²·kPa⁻¹ (to vapor pressure deficit), respectively. Our threshold-based framework reveals how alpine grasses adapt to compound drought, offering actionable metrics for early-warning systems and targeted conservation in global water-limited ecosystems.