A hybrid modelling approach for estimating emissions from ship manoeuvres in port areas: Integrating pilot experience and machine learning

Emissions generated during ship manoeuvring in ports represent a critical source of local air pollution although this has received limited attention due to its short duration. In this research 45 berthing manoeuvres were conducted in a full-mission bridge simulator by 15 maritime pilots for bulk-carrier, Ro-Ro and container-ship scenarios. Using a hybrid bottom-up approach integrating Entec and EPA methodologies, emissions from both the own-ship and tugs were quantified. Total emissions varied between 112.91 and 306.98 kg, 127.01–195.06 kg, and 307.35–369.98 kg for the three vessel types, corresponding to 2.71, 1.54, and 1.20-fold differences, respectively. To explain this variability two experiential coefficients C1-C2 representing pilot familiarity with similar vessels were introduced and used as inputs in the MATLAB Regression-Learner-Toolbox. Linear and stepwise regression models yielded the most reliable predictive performance with R2 values of 0.70, 0.79 and 0.76 respectively. Complementary regression equations further enable straightforward estimation of emissions as a simplified and operational extension of the simulation results by substituting C1-C2 values, with negative coefficients confirming that greater pilot familiarity and experience reduce emissions. The framework demonstrates a cost-effective and transferable method for predicting manoeuvring emissions that can support the development of emission conscious training practices and greener port operations.