BACKGROUND: Heterogeneity in mosquito exposure at a fine spatial scale is well documented but often neglected in mathematical models of arbovirus transmission, particularly for real-time forecasting. We aimed to assess whether accounting for heterogeneity in mosquito exposure improves the ability of transmission models to reproduce past chikungunya outbreaks and to forecast an ongoing outbreak in real time. METHODS: We extended an existing transmission model, originally developed to forecast dengue outbreaks in Reunion Island and incorporating seasonality, to include heterogeneity in mosquito exposure. We compared this model to versions assuming homogeneous exposure, evaluating their ability to reproduce the 2005-06 Reunion Island outbreak and the 2023 Paraguay Metropolitana subregion outbreak. We then used the heterogeneity model to forecast the 2025 Reunion Island outbreak in real time. FINDINGS: Accounting for heterogeneity in mosquito exposure at a fine spatial scale was crucial to accurately reproduce Reunion Island's 2005-06 outbreak. Homogeneity models either overestimated final seroprevalence (80% vs 38% observed) or required a reduction of approximately 50% in the basic reproduction number following interventions, which is implausible. The heterogeneity level estimated from this outbreak transferred successfully to the Metropolitana outbreak, accurately forecasting epidemic trajectory and final seroprevalence (∼40% vs ∼75% for the homogeneity model). Under the same assumption, the heterogeneity model successfully forecasted the peak of the 2025 Reunion Island outbreak (∼400 emergency room visits per week) and timing (April), nearly two months ahead. A post-outbreak serosurvey indicated that the heterogeneity model underestimated final seroprevalence (46·1% vs 61·6% observed), whereas the homogeneity model overestimated final seroprevalence (78·5%), suggesting that heterogeneity in 2025 might have been lower than that in 2005-06. INTERPRETATION: These results highlight how incorporating heterogeneity in mosquito exposure can improve chikungunya outbreaks forecasting. Our approach showed strong predictive performance in settings with climatic seasonality and holds promise for forecasting future chikungunya and other arbovirus outbreaks.