National-scale mapping of environmental noise exposure and socioeconomic disparities in China using a two-stage stacked ensemble framework

Rapid urbanization has intensified environmental noise in China, yet a national-scale environmental noise exposure map remains absent. This study developed a two-stage stacked ensemble learning framework to map acoustic environments across mainland China and assess socioeconomic disparities in noise exposure. We integrated 554,948 observations from 103,102 environmental noise monitoring stations with multisource geospatial predictors, including transport networks, land use, population density, nighttime lights, GDP, building footprints, points of interest, air pollutants, meteorological variables, elevation, and vegetation. The modelling framework combined Random Forest, XGBoost, LightGBM, and CatBoost as base learners, followed by a LightGBM meta-learner and residual calibration. The final model achieved an R 2 of 0.660, RMSE of 4.720 dB, and MAE of 3.631 dB at the national scale, with performance varying across regions and urbanization strata. The predicted 1 km × 1 km noise maps revealed spatial and temporal gradients, including higher exposure in eastern urban agglomerations and sustained evening-nighttime noise in metropolitan cores. Population-weighted L den declined across GDP per capita quintiles, whereas the proportion of the population exposed to L den >53 dB was highest in the lower-middle group and declined across the higher quintiles. These findings provide macro-scale evidence of environmental noise exposure and inequality in China and support differentiated strategies for sustainable urban noise management.