Rangeland dieback is an important manifestation of land degradation in arid and semi-arid regions and results from interactions among climatic, moisture-related, vegetation, and physiographic factors. This study aimed to model and predict dieback percentage in Chaharmahal-va-Bakhtiari Province, Iran, by comparing multiple linear regression (MLR) and Random Forest (RF) using field data and remote-sensing indices. Dieback percentage was measured in 1,100 2 × 2 m plots and used as the response variable. Predictors included annual precipitation, actual evapotranspiration, Vegetation Condition Index (VCI), Temperature Condition Index (TCI), OPTRAM soil-moisture index, elevation, and slope. VCI and TCI were derived from Landsat-8 imagery acquired during the 2023 growing season and interpreted as within-year relative indicators. Land surface temperature was excluded after variance inflation factor (VIF) screening because of severe collinearity. RF outperformed MLR, with R² = 0.70 and RMSE = 3.44 compared with R² = 0.40 and RMSE = 4.80, respectively. Annual precipitation, VCI, elevation, and OPTRAM were the most important predictors. Because of their conceptual overlap with observed dieback, VCI and OPTRAM were interpreted primarily as diagnostic indicators of dieback status rather than independent drivers. An exploratory path analysis showed inadequate fit and was therefore reported only as assumed structural associations. Overall, rangeland dieback was multifactorial and largely nonlinear, and RF can provide an effective tool for spatial prediction and monitoring of areas prone to rangeland dieback.