To reduce the environmental impact of short sea shipping, this study introduces a two-stage propulsion power allocation method aimed at enhancing ship operational efficiency in various weather environments. The first stage utilizes a metocean score-based pruned explicit linear time (MS-PELT) algorithm to segment the trajectory into several legs based on metocean conditions, thereby minimizing frequent engine setting adjustments and simplifying the optimization process. In the second stage, a parallel coupling Dynamic Programming (PCDP) method is introduced to optimize power allocation in each leg using machine learning-based ship performance models. The proposed approach is evaluated using three years of full-scale operational data from a case study chemical tanker. Results show that the MS-PELT method outperforms the state-of-the-art multivariate clustering algorithm by providing practical and efficient segmentation. The optimized power allocation strategy demonstrates a promising average of 8 % emission and environmental impact reductions for case study short sea voyages with good computational efficiency. It is suitable for real-time applications, providing the maritime industry with tools to optimize ship engine settings, reducing emissions and environmental impact.