Reliable vegetation forecasting requires capturing both complex spatiotemporal patterns and inherent uncertainties, yet prevailing deterministic approaches focus solely on accuracy while neglecting the uncertainties in vegetation evolution. To address this limitation, we present HyDiffVeg, a novel hybrid framework that advances uncertainty-aware vegetation forecasting by synergistically integrating deterministic prediction with diffusion-based generation. Our key insight is to leverage a diffusion bridge for modeling the distribution of future states, complemented by a multimodal deterministic stream that extracts robust environmental patterns and provides structural guidance. This architecture enables the generation of diverse yet coherent forecasts that capture both dominant trends and meaningful uncertainties. Extensive evaluations on the EarthNet2021 benchmark demonstrate that HyDiffVeg substantially advances the state-of-the-art, achieving improvements of 10.8% in standard scenarios and 15.2% in geographic generalization tests, validating its effectiveness for uncertainty-aware vegetation forecasting.