A study on the refined population spatialization method integrating multi-source data: A case study of Yuxi city

Area of Science:

• Geographic Information Science
• Remote Sensing
• Data Science

Background:

• Precise population spatialization data is essential for accurate earthquake disaster assessment.
• Advancements in remote sensing and social sensing technologies offer new opportunities for high-resolution population mapping.
• Existing population datasets may lack the necessary accuracy for detailed disaster impact analysis.

Purpose of the Study:

• To develop and validate a novel method for high-resolution population spatialization.
• To integrate diverse data sources including social perception big data, building attributes, and multi-source remote sensing data.
• To improve the accuracy of population distribution data for disaster management applications.

Main Methods:

• A Stacking ensemble learning model was constructed using Random Forest, XGBoost, LGBM, Gradient Boosting, AdaBoost, and CatBoost as base learners, with RidgeCV regression as the secondary learner.
• The Partition Density Mapping technique was employed to spatialize population data at a 50m resolution for Yuxi City in 2020.
• Feature importance was analyzed using SHAP (SHapley Additive exPlanations) to identify key indicators of population distribution.

Main Results:

• The proposed population estimation model achieved over 40% higher accuracy than WorldPop and LandScan datasets at the administrative village scale.
• The model demonstrated superior simulation accuracy across regions with varying population densities.
• Residential building area, Points of Interest (POI), and nighttime light intensity were identified as the most significant predictors of population distribution.

Conclusions:

• The integrated data and Stacking ensemble model provide richer and more realistic population distribution information compared to existing datasets.
• The findings underscore the importance of social factors and refined spatial features in population redistribution.
• This enhanced population spatialization method offers a valuable tool for more precise earthquake disaster assessment and planning.