In response to the global imperative of achieving carbon neutrality, increasing attention is being directed toward procuring Frequency Containment Reserve (FCR) not from conventional thermal power plants, but via Virtual Power Plants (VPPs) that aggregate thousands of distributed energy resources (DERs) such as photovoltaics (PV), battery energy storage systems, and fuel cells. These VPPs, coordinated by aggregators, harness the controllable surplus capacity of DERs—excluding on-site consumption—to participate in day-ahead FCR markets. For effective market participation, aggregators must schedule FCR provision by optimizing expected market revenue while accounting for PV generation, electricity consumption, and the dispatchable capacity of each prosumer. On the day of operation, the VPP is required to deliver dynamic power responses to grid frequency deviations, which arise from imbalances between supply and demand. In systems with high penetration of variable renewable energy sources, grid frequency exhibits pronounced volatility and uncertainty. Consequently, relying on a single deterministic frequency forecast may lead to scheduling errors, resulting in inadequate controllable capacity during real-time operation. To address this issue, we propose a scenario-based FCR scheduling framework that incorporates probabilistic frequency forecasting using Natural Gradient Boosting (NGBoost) and a copula-based model. NGBoost provides probabilistic forecasts that capture predictive uncertainty, while copulas enable the modeling of complex temporal dependencies in frequency fluctuations. By integrating these probabilistic scenarios into the scheduling optimization as constraints, the proposed method enhances the reliability of VPP-based FCR provision. Simulation results based on real data from 2000 prosumers confirm the validity of the proposed approach.