Forests play a crucial role in the global carbon cycle and climate change mitigation. However, regional-scale assessments of forest carbon storage and hotspot identification remain challenging. In this study, we employed a k-fold random forest (K-RF) algorithm to classify forest types across the Loess Plateau and developed a spatial optimization simulation method for multi-climate scenario projections. This framework was applied to simulate forest distribution under different climate scenarios for 2030 and 2060, while also estimating carbon storage and identifying hotspots from 1985 to 2060. Our findings demonstrate that the accuracy of forest remote sensing classification extraction overall accuracy (OA) and Kappa increased over time, and OA reached above 0.90, indicating the reliability of the classification results. The forest area increased by 45% from 1985 to 2020. Future forest simulation OA and Kappa were above 0.87, indicating a good simulation model. Intriguingly, elevated atmospheric CO2 concentrations were projected to exert inhibitory effects on future forest distribution. Specifically, under SSP126 and SSP370 scenarios, carbon storage manifested diffused spatial patterns, whereas the SSP585 scenario predicted a northeastward displacement of 131.62 Tg C in forest carbon storage. These evidence-based projections provide critical scientific support for regional carbon neutrality strategies and climate change adaptation planning.