Carbon emissions from urban takeaway delivery in China

The rapid expansion of online food delivery in China, facilitated by the Online-to-Offline (O2O) business model, presents a significant environmental challenge. While the convenience and economic benefits are undeniable, the associated carbon footprint from transportation and packaging remains largely unexplored. This study addresses this gap by providing a comprehensive assessment of GHG emissions from urban takeaway delivery across China. The Paris Agreement's goals necessitate substantial emission reductions across various sectors, and the food system, including its increasingly digitalized delivery aspects, is a critical area of focus. China’s O2O food delivery market, driven by platforms like Meituan, Eleme, and Baidu, shows exponential growth, with adoption rates exceeding 50% by 2021. This rapid expansion highlights the urgent need to understand and mitigate the environmental impact of this sector. Existing research offers limited national-level assessments of carbon emissions from food delivery transportation in China, often relying on incomplete or limited data. This research aims to provide a more complete national assessment, using a robust model that integrates multiple data sources to accurately estimate GHG emissions and identify key impact factors.

Existing literature has estimated carbon emissions from food delivery in specific cities or regions, but these studies often suffer from data limitations. While some research has addressed carbon emissions from packaging waste in China's takeaway sector, a detailed national assessment of transportation-related emissions has been lacking. This research fills this gap by providing the first detailed national-level estimate for transportation-related carbon emissions from China's takeaway delivery system, using a model that integrates multiple data sources for improved accuracy and broader spatial coverage.

This study employed a two-pronged methodological approach. First, a Food Delivery Carbon Emission (FDCE) model was developed, based on the IPCC standard emissions factor method, to estimate transportation-related GHG emissions in 56 cities using Meituan order data. This model considered various delivery vehicles (electric bikes, electric motorcycles, fuel motorcycles of different engine sizes), calculating emissions based on vehicle type, energy consumption, and distance traveled. The FDCE model provided initial data for correlation and sensitivity analyses of key factors influencing emissions (order volume, delivery distance, proportion of electric bikes). Second, to expand the analysis to 270 cities where direct order data were unavailable, a combined machine learning and FDCE model (ML-FDCE) was developed. Machine learning models (primarily XGBoost) were trained using eight dimensions of city-level indicators (economy, population, traffic, climate, area, catering industry, communication facilities, and takeaway search volume) to predict annual takeaway order volume (ATOV) and delivery distance per order (DDPO). These predictions were then input into the FDCE model to estimate GHG emissions for the 270 cities. Ten-fold cross-validation was used to assess the predictive models’ accuracy, and SHAP values were employed to determine feature importance. A Monte Carlo simulation was used to propagate uncertainties in key parameters throughout the analysis, improving the robustness of the results. Finally, the study employed scenario analysis to simulate the impact of several GHG mitigation strategies (replacing fuel motorcycles with electric vehicles, fully promoting electric bikes, and optimizing delivery routes).

The study's key findings include: In 2019, urban takeaway delivery in China generated approximately 1.67 MtCO2e, with transportation accounting for a substantial portion (44.67%). Transportation-related emissions were 745 KtCO2e, averaging 0.057 kg CO2e per order and 0.011 kg CO2e per capita. The study identified strong positive correlations between GHG emissions and order volume (R=0.834), and between GHG emissions and a city’s GRP (R=0.625). Conversely, a negative correlation was observed between GHG emissions per order and the proportion of electric bikes used for delivery (R=-0.39). Sensitivity analyses revealed that order volume, delivery distance, and the proportion of electric bikes were the most influential factors. The ML-FDCE model successfully estimated GHG emissions for 270 cities, expanding the study's scope significantly. Scenario analysis indicated that switching to electric bikes and optimizing delivery routes could significantly reduce cumulative GHG emissions (4.39-10.97 MtCO2e between 2023 and 2035). Cities with high delivery distances and a high proportion of motorcycles showed the greatest carbon mitigation potential.

The findings highlight the significant environmental impact of China's rapidly growing online food delivery industry. The strong correlation between emissions and order volume underscores the need for strategies to manage demand growth. The significant impact of electric bike adoption suggests that promoting their use is a crucial mitigation strategy. The success of the ML-FDCE model demonstrates the potential of combining mechanistic modeling with machine learning for estimating emissions where direct data are scarce. The scenario analysis results provide actionable insights for policymakers, showing how targeted interventions can substantially reduce emissions. The study’s findings contribute to a broader understanding of the environmental impact of digitalized food systems, providing valuable information for sustainable urban planning and policymaking in China and potentially other rapidly developing nations.

This study provides a comprehensive assessment of GHG emissions from China's urban takeaway delivery industry, highlighting the significant environmental impact of this rapidly expanding sector. The development of the FDCE and ML-FDCE models provides a robust framework for estimating emissions at both local and national scales, even in the absence of comprehensive data. Scenario analysis indicates the significant potential for emission reduction through policy interventions focusing on electric vehicle adoption and route optimization. Future research should explore the integration of urban planning strategies to further reduce delivery distances and the potential for initiatives promoting household social responsibility to moderate takeaway consumption.

The study acknowledges several limitations. The accuracy of the machine learning models relies on the quality and completeness of the input data. While efforts were made to account for uncertainty, there remains inherent uncertainty in model projections. The study focuses primarily on transportation-related emissions, while acknowledging that packaging and food production also contribute significantly. Future research should aim to incorporate these aspects for a more holistic assessment. Finally, the focus on China's context means that generalizability to other countries requires further investigation, considering differences in infrastructure, urban design, and consumer behavior.