Leveraging vehicular test databases to develop a national emissions prediction model: a case study for Sri Lanka

Abstract

This study presents a data-driven framework for predicting national vehicular emissions in Sri Lanka using the existing emissions testing database maintained by the Department of Motor Traffic (DMT), Sri Lanka. Focusing on passenger cars with internal combustion engines, the study adopts a Tier 2 bottom-up estimation methodology, incorporating annual mileage and vehicle category-technology composition to calculate emissions for pollutants such as CO, NMVOC, NOₓ, N₂O, and NH₃. A custom data pre-processing pipeline, was implemented to assign emission factors based on engine power and vehicle vintage. Average mileage per vehicle per annum was derived by cross-checking vehicle test entries over successive years, on a monthly basis. A Long Short-Term Memory (LSTM) neural network was trained with features including GDP, vehicle fleet data, and cyclical encoding of months to forecast mileage. The model achieved strong predictive accuracy (R² = 0.899). Emission forecasts for the next 12-month ranges were computed by integrating predicted mileage with fleet composition and emission factors. The framework offers scalability across vehicle types and serves as a policy-support tool for emission regulation. Despite limitations related to hybrid and electric vehicle data availability, this work demonstrates the potential of leveraging underutilized vehicle test databases for national-scale emissions forecasting in developing countries.