Full-coverage high-resolution (Daily, 1-km) PM2.5 dataset in China (2000-present)

We have estimated full-coverage, daily 1-km PM2.5 data from 2000 to 2022 in China using a random forest-based hindcast modeling method. Our modeling method focused on improving pre-2013 PM2.5 estimates because for those years no available PM2.5 measurements can be directly used for constructing the model and evaluating the model performance. In our proposed method, observed predictor information before 2013 was incoporated into the modeling for the first time. Multiple sources were used as inputs, including MAIAC AOD, meteorological data from CMA, reanalysis data from ERA-5, and other land-related data. The daily average data during 2000-2022 are released here and free for non-commercial use. If you want use our dataset, please cite the following publication. 

The estimates in 2021-2022 are separately predicted using the same modeling method developed in the publication below and samples in the corresponding predictive year (sample-based 10-fold cross validation R2 [RMSE] values are 0.91 [8.84 ug/m3] for 2021 and 0.93 [7.42 ug/m3] for 2022, respectively. 

-He, Q., Ye, T., Wang, W., Luo, M., Song, Y., & Zhang, M. (2023). Spatiotemporally continuous estimates of daily 1-km PM2. 5 concentrations and their long-term exposure in China from 2000 to 2020. Journal of Environmental Management, 342, 118145.[url]

-He, Q., Wang, W., Song, Y., Zhang, M., & Huang, B. (2023). Spatiotemporal high-resolution imputation modeling of aerosol optical depth for investigating its full-coverage variation in China from 2003 to 2020. Atmospheric Research, 281, 106481.[url]

Full-coverage daily estimates spanning the years 2015 to Jun 2021 are archived here. These records, organized by month, are available for download in CSV format. For Jul-Dec 2021, please go to 10.5281/zenodo.8084388.

If you want more data (e.g.daily estimates before 2015), have any question, or further collaborate with us, please contact us via [e-mail].

If you want to use monthly estimates from 2000 to 2022, please go to 10.5281/zenodo.8084388.

We also estimate other atmospheric data:

For full-coverage, 1-km, AOD data in China, please go to harvard dataverse. This dataset was imputed based on MODIS MAIAC 1-km AOD retrievals.

本研究基于随机森林（random forest）后向预测建模方法，估算了2000年至2022年中国全域覆盖、逐日1千米分辨率细颗粒物（PM2.5）浓度数据。

我们的建模方法重点优化了2013年之前的PM2.5浓度估算结果，因为该时段尚无可用的PM2.5实测数据可直接用于模型构建与性能评估。在本研究提出的方法中，2013年之前的实测预测因子信息首次被纳入建模流程。本方法采用多源数据作为模型输入，包括MAIAC气溶胶光学厚度（AOD）、中国气象局（CMA）气象数据、ERA-5再分析资料以及其他土地相关数据。本数据集发布了2000年至2022年的逐日平均数据，非商业用途可免费获取。若您希望使用本数据集，请引用如下发表文献。

2021年与2022年的估算结果采用本研究上述发表文献中开发的同一建模方法，结合对应预测年份的样本单独预测得到（基于样本的10折交叉验证的决定系数[R²][均方根误差（RMSE）]分别为0.91[8.84 μg/m³]与0.93[7.42 μg/m³]）。

- 贺祺, 叶涛, 王威, 罗敏, 宋熠, 张明. (2023). 2000—2020年中国逐日1千米分辨率细颗粒物浓度的时空连续估算及长期暴露特征. *Journal of Environmental Management*, 342, 118145.[url]
- 贺祺, 王威, 宋熠, 张明, 黄斌. (2023). 中国2003—2020年气溶胶光学厚度时空高分辨率插补建模及其全域变化研究. *Atmospheric Research*, 281, 106481.[url]

2015年至2021年6月的全域逐日估算数据已存档于此，按月份组织整理，可通过CSV格式下载。2021年7月至12月的数据请访问DOI：10.5281/zenodo.8084388。

若您需要更多数据（例如2015年之前的逐日估算数据）、有任何疑问或希望与我们开展进一步合作，请通过邮箱[e-mail]联系我们。

若您希望使用2000年至2022年的逐月估算数据，请访问DOI：10.5281/zenodo.8084388。

本团队同时估算了其他大气相关数据：

针对中国全域覆盖、1千米分辨率的气溶胶光学厚度（AOD）数据，请访问哈佛数据文库（Harvard Dataverse）。该数据集基于MODIS MAIAC 1千米分辨率AOD反演结果插补得到。