Predicting carbon footprint in stochastic dynamic routing using Bayesian Markov random fields

Abstract

Evaluating carbon emissions in last-mile logistics is critical for achieving climate goals, yet current models lack integration of spatiotemporal traffic dynamics and climate factors. This study aims to (1) develop a Bayesian Markov random field model integrating spatiotemporal traffic data and speed scenarios influenced by precipitation, (2) quantify carbon dioxide emissions from last-mile logistics illustrated by maintenance dispatches in power distribution systems using a widely recognized traffic speed to CO2 conversion method, and (3) provide actionable strategies for reducing emissions in last-mile logistics. Achieving a traffic speed prediction accuracy with an approximate error of 2%, the proposed model quantified carbon emissions under dynamic routing conditions. Simulation results from maintenance dispatches in power distribution systems indicate that, under average failure conditions, the annual carbon emissions from two teams operating in São Paulo are equivalent to the carbon dioxide absorbed by approximately five hectares of trees. These findings underscore the critical importance of incorporating environmental considerations into reliability assessments. While the study focuses on power distribution systems, the proposed framework is broadly applicable to any last-mile logistics problem, offering actionable insights—such as optimizing dispatch frequencies—to minimize emissions. By addressing the cumulative environmental impact of routine operations, this research supports the transition to carbon-neutral last-mile services and promotes responsible logistics practices across industries worldwide.