Applying LightGBM to Enhance the Accuracy of Emission Inventory for the Hai An Port: A Premise for Developing Emission Mitigation Strategies

Abstract

Sustainable port management is becoming a challenge due to the fact that maritime transportation is a significant GHG emitter and source of atmospheric pollutants. This study aims to offer an integrated framework which integrates conventional emission inventory approach with the Light Gradient Boosting Machine (LightGBM) algorithm and SHapley Additive exPlanations (SHAP) to enhance the accuracy and interpretability of ship emission estimation. The framework has been built based on operational data from 33 vessels (2197 vessel calls) in Hai An Port (Vietnam). Greenhouse gas (CO₂, CH₄, N₂O) and atmospheric pollutant (NOₓ, SO₂, CO, VOC, PM₁₀, PM₂.₅) emission inventories were created, and the predictive performance of Linear Regression, Random Forest, XGBoost, and LightGBM was compared. The results demonstrate that the predictive performance of LightGBM was the best with coefficients of determination (R²) of 0.957 to 0.986 and mean absolute percentage error (MAPE) of 3.85% to 6.78%, which were obtained for all the pollutants investigated. SHAP analysis also found that the most important features that drive the emission estimates are engine power, engine load, vessel speed, operating mode and engine operating time. According to the annual emission inventories, CO₂ and NOₓ were the major GHG and AP, respectively, and it is important to improve fuel efficiency and the operation management to reduce emission. The proposed framework will be accurate and interpretable for the ship emission inventory development, and will facilitate the data-driven environmental management, emission reduction strategies and sustainable green port development for port authorities and maritime stakeholders.