Improving the accuracy of emission inventories with a machine-learning approach and investigating transferability across cities

Abstract

This study presents a novel method for integrating the output of a microscopic emission modeling approach with a regional traffic assignment model in order to achieve an accurate greenhouse gas (GHG, in CO 2-eq ) emission estimate for transportation in large metropolitan regions. The CLustEr-based Validated Emission Recalculation (CLEVER) method makes use of instantaneous speed data and link-based traffic characteristics in order to refine on-road GHG inventories. The CLEVER approach first clusters road links based on aggregate traffic characteristics, then assigns representative emission factors (EFs), calibrated using the output of microscopic emission modeling. In this paper, cluster parameters including number and feature vector were calibrated with different sets of roads within the Greater Toronto Area (GTA), while assessing the spatial transferability of the algorithm. Using calibrated cluster sets, morning peak GHG emissions in the GTA were estimated to be 2,692 tons, which is lower than the estimate generated by a traditional, average speed approach (3,254 tons). Link-level comparison between CLEVER and the average speed approach demonstrates that GHG emissions for uncongested links were overestimated by the average speed model. In contrast, at intersections and ramps with more congested links and interrupted traffic flow, the average speed model underestimated GHG emissions. This proposed approach is able to capture variations in traffic conditions compared to the traditional average speed approach, without the need to conduct traffic simulation.Implications: A reliable traffic emissions estimate is necessary to evaluate transportation policies. Currently, accuracy and transferability are major limitations in modeling regional emissions. This paper develops a hybrid modeling approach (CLEVER) to bridge between computational efficiency and estimation accuracy. Using a k-means clustering algorithm with street-level traffic data, CLEVER generates representative emission factors for each cluster. The approach was validated against the baseline (output of a microscopic emission model), demonstrating transferability across different cities .

Predicting Chemical End-of-Life Scenarios Using Structure-Based Classification Models

“…30 Huang et al 31 explored the use of deep learning algorithms for air pollutant emission estimation. Tu et al 32 presented a hybrid modeling approach, named "cluster-based validated emission recalculation," for predicting the emissions of greenhouse gases from the Great Toronto Area.…”