Mountain rangelands are vital for forage supply, biodiversity, and the carbon cycle. This research aimed to forecast the long-term grazing capacity of summer rangelands in the Arshad Chaman Kandovan Sahand area and assess climate change effects on vegetation production and cover. Data were gathered from six transects over 16 years (2006–2021). Vegetation measurements included canopy cover and forage yield across three palatability categories (I, II, III), along with soil surface conditions. Climate data—covering minimum, maximum, and average temperatures and precipitation—were sourced from Tabriz, Sahand, and Maragheh stations and analyzed with SPI and SPEI drought indices. GCM data were downscaled using SDSM under RCP scenarios. Machine learning models such as XGBoost, Random Forest, and Linear Regression then simulated grazing capacity. Results showed forage production decreased from 1250 kg/ha in 2006 to 1054 kg/ha in 2021, a 15.6% decline. Similarly, available forage reduced from 408.2 to 316.7 kg/ha, a 22.4% drop. Consequently, short-term grazing capacity fell from 7.3 to 5.7 Animal Unit Months per hectare (AUM/ha). For future projections, the XGBoost model (R² = 0.96) estimated a 3.5% decline by 2051, while the Random Forest model (R² = 0.85) predicted a 3.0% decrease. Based on these outcomes, reducing livestock numbers by about 20% by 2051 is recommended. Additionally, vegetation improvement programs—such as seeding and grazing management—should be implemented with local community participation, along with ongoing monitoring.