Divergent Spatial Patterns of Household Food Consumption in China: A Hybrid Spatial Econometrics-Machine Learning Approach

Abstract

Food consumption maps link food supply, consumption as well as food nutrition, offer useful information for analyzing people's food intake and exploring food flow and consumption pattern, and are increasingly regarded as important research tools and data sources for the science community. Based on statistical records from 292 prefectures in China between 2015 and 2020, and considering multi-dimensional driving factors from socioeconomic, climatic, and environmental health, this study builds a hybrid model integrating Spatial Error Model (SEM) and XGBoost machine learning algorithm to simulate spatial distributions of per capita food consumption (PCFC) across urban and rural areas in China. The results indicate that the hybrid model effectively captures both spatial dependence and nonlinear relationships of the PCFC. Validation with independent official data from 2021 to 2023 shows satisfactory accuracy. The results reveal a clear north–south gradient in PCFC, with rural PCFC growing faster than urban areas, reflecting the combined effects of narrowing urban–rural income gap, diffusion of urban dietary patterns, and improvements in rural supply chains. This study is the first-ever endeavor to uncover the spatial distribution of food consumption in China and it offers an insightful approach into the underlying mechanisms of regional dietary imbalances. This study suggests the needs for differentiated policies to promote sustainable development for China's food system and for enhanced actions to improve data coverage and accuracy.