Understanding the health impacts of particulate matter (PM) requires spatiotemporally continuous exposure estimates. We developed a multi-stage ensemble model that estimates daily mean PM_2.5 and PM₁₀ at 1 km spatial resolution across France from 2000 to 2019. First, we alleviated the sparsity of PM_2.5 monitors by imputing PM_2.5 at more common PM₁₀ monitors. We also imputed missing satellite aerosol optical depth (AOD) based on modelled AOD from atmospheric reanalyses. Next, we trained three base learners (mixed models, Gaussian Markov random fields, and random forests) to predict daily PM concentrations based on AOD, meteorology, and other variables. Finally, we generated ensemble predictions using a generalized additive model with spatiotemporally varying weights that exploit the strengths and weaknesses of each base learner. The Gaussian Markov random field dominated the ensemble, outperforming mixed models and random forests at most locations on most days. Rigorous cross-validation showed that the ensemble predictions were quite accurate, with mean absolute error (MAE) of 2.72 μg/m³ and R² of 0.76 for PM_2.5; PM₁₀ MAE was 4.26 μg/m³ and R² 0.71. Our predictions are available to improve epidemiological studies of acute and chronic PM exposure in urban and rural France.

了解颗粒物 (PM) 对健康的影响需要时空连续暴露估计。我们开发了估计日平均PM多级集成模型_2.5和PM ₁₀，在法国各地1公里空间分辨率从2000年到2019年。首先，我们缓解PM的稀疏_2.5通过插补PM监测_2.5在比较常见的PM ₁₀监视器。我们还根据大气再分析的模型 AOD 估算了缺失的卫星气溶胶光学深度 (AOD)。接下来，我们训练了三个基础学习器（混合模型、高斯马尔可夫随机场和随机森林），以根据 AOD、气象和其他变量预测每日 PM 浓度。最后，我们使用具有时空变化权重的广义加性模型生成集成预测，该模型利用每个基础学习器的优势和劣势。高斯马尔可夫随机场在整个集合中占主导地位，在大多数日子里的大多数位置都优于混合模型和随机森林。严格的交叉验证表明集合预测非常准确，PM _{2.5 的}平均绝对误差 (MAE) 为 2.72 μg/m ³，R ²为 0.76; PM ₁₀ MAE 为4.26 μg/m ³和R ² 0.71。我们的预测可用于改进法国城市和农村急性和慢性 PM 暴露的流行病学研究。