We consider the problem of scheduling trucks to pick up fully-loaded containers at manufacturer warehouses in export logistics. Due to factors such as the location and type of the previous task as well as the in-transit behavior of the drivers with different carriers, truck arrival time deviations often show a significant mixed pattern which makes it nontrivial to set their appointment schedules. In this context, we first design a prediction model for classifying the punctuality distribution of each truck to be scheduled, based on which a stochastic optimization problem can be solved for minimizing the expected waiting times. In addition, we propose a distributionally robust optimization approach to address possible classification errors. Both simulation analysis and real-data based case experiment verify the effectiveness of the proposed approach in reducing inefficiencies in truck appointment scheduling.