Related Experiment Videos
Mapping rural livelihood strategies in China using deep learning
Zhaxi Dawa1, Wenjuan Yu2, Weiqi Zhou1,3,4,5
1State Key Laboratory of Regional and Urban Ecology, Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences, Beijing, 100085, China.
Scientific Data
|July 21, 2026
Summary
Mapping rural China
Area of Science:
- Rural livelihood strategies
- Socio-ecological systems
- Sustainability outcomes
Background:
- Spatially explicit data on rural livelihood compositions are scarce at national scales.
- Understanding rural livelihood diversity is crucial for socio-ecological system analysis.
- National-scale mapping of rural livelihoods is needed for sustainability research.
Purpose of the Study:
- To develop a Deep Rural Livelihood Model (DRLM) for mapping livelihood strategies in rural China.
- To create settlement-scale probabilistic data on four livelihood strategies: farming-only, farming-dominated mixed, non-farming-dominated mixed, and non-farming-only.
- To provide data supporting analyses of rural transformation and sustainability.
Main Methods:
- Combined satellite imagery (Landsat, VIIRS) with household survey data.
- Utilized quantile-regression XGBoost to expand survey labels to 38,306 rural settlements.
- Employed a dual-branch ResNet architecture with Dirichlet regression for probabilistic mapping.
Main Results:
- Achieved high agreement with expanded settlement-level labels.
- Independent validation showed moderate performance, with better accuracy for dominant livelihood categories.
- Generated settlement-scale probabilistic livelihood information for rural China in 2020.
Conclusions:
- The Deep Rural Livelihood Model (DRLM) successfully mapped rural livelihood strategies across China.
- The generated dataset offers valuable insights into rural transformation and sustainability.
- Probabilistic mapping enhances understanding of complex rural socio-ecological systems.