Daily and Annual PM2.5, O3, and NO2 Concentrations at ZIP Codes for the Contiguous U.S., 2000-2016, v1.0

The Daily and Annual PM2.5, O3, and NO2 Concentrations at ZIP Codes for the Contiguous U.S., 2000-2016, v1.0 data set contains daily and annual concentration predictions for Fine Particulate Matter (PM2.5), Ozone (O3), and Nitrogen Dioxide (NO2) pollutants at ZIP Code-level for the years 2000 to 2016. Ensemble predictions of three machine-learning models were implemented (Random Forest, Gradient Boosting, and Neural Network) to estimate the daily PM2.5, O3, and NO2 at the centroids of 1km x 1km grid cells across the contiguous U.S. for 2000 to 2016. The predictors included air monitoring data, satellite aerosol optical depth, meteorological conditions, chemical transport model simulations, and land-use variables. The ensemble models demonstrated excellent predictive performance with 10-fold cross-validated R-squared values of 0.86 for PM2.5, 0.86 for O3, and 0.79 for NO2. These high-resolution, well-validated predictions allow for estimates of ZIP Code-level pollution concentrations with a high degree of accuracy. For general ZIP Codes with polygon representations, pollution levels were estimated by averaging the predictions of grid cells whose centroids lie inside the polygon of that ZIP Code; for other ZIP Codes such as Post Offices or large volume single customers, they were treated as a single point and predicted their pollution levels by assigning the predictions using the nearest grid cell. The polygon shapes and points with latitudes and longitudes for ZIP Codes were obtained from Esri and the U.S. ZIP Code Database and were updated annually. The data include about 31,000 general ZIP Codes with polygon representations, and about 10,000 ZIP Codes as single points. The aggregated ZIP Code-level, daily predictions are applicable in research such as environmental epidemiology, environmental justice, health equity, and political science, by linking with ZIP Code-level demographic and medical data sets, including national inpatient care records, medical claims data, census data, U.S. Census Bureau American CommUnity Survey (ACS), and Area Deprivation Index (ADI). The data are particularly useful for studies on rural populations who are under-represented due to the lack of air monitoring sites in rural areas. Compared with the 1km grid data, the ZIP Code-level predictions are much smaller in size and are manageable in personal computing environments. This greatly improves the inclusion of scientists in different fields by lowering the key barrier to participation in air pollution research. The Units are ug/m^3 for PM2.5 and ppb for O3 and NO2.

该《美国大陆每日及年度PM2.5、O3和NO2浓度按ZIP代码划分数据集（2000-2016，v1.0）》包含了细颗粒物（PM2.5）、臭氧（O3）和二氧化氮（NO2）污染物在ZIP代码级别的每日和年度浓度预测。本数据集采用三种机器学习模型（随机森林、梯度提升和神经网络）的集成预测，以估算2000年至2016年美国大陆1km x 1km网格单元中心点的PM2.5、O3和NO2的每日浓度。预测因子包括空气质量监测数据、卫星气溶胶光学厚度、气象条件、化学传输模型模拟和土地利用变量。集成模型展现出卓越的预测性能，PM2.5、O3和NO2的10折交叉验证R平方值分别达到0.86、0.86和0.79。这些高分辨率且经过充分验证的预测结果，能够以高精度估算ZIP代码级别的污染浓度。对于具有多边形表示的通用ZIP代码，通过平均位于该ZIP代码多边形内部的网格单元预测值来估算污染水平；对于如邮局或大量单一客户等ZIP代码，则将其视为单一点，通过分配最近的网格单元的预测值来估算其污染水平。ZIP代码的多边形形状和经纬度坐标数据来源于Esri和美国ZIP代码数据库，并每年更新。数据包括大约31,000个具有多边形表示的通用ZIP代码，以及大约10,000个作为单一点的ZIP代码。汇总的ZIP代码级别每日预测数据可用于环境流行病学、环境正义、健康公平和政治科学等研究领域，通过与ZIP代码级别的人口统计和医疗数据集相结合，包括国家住院护理记录、医疗索赔数据、人口普查数据、美国人口普查局美国社区调查（ACS）和区域剥夺指数（ADI）。对于因农村地区缺乏空气质量监测站点而代表性不足的农村人口研究，这些数据尤其有用。与1km网格数据相比，ZIP代码级别的预测数据规模更小，且在个人计算环境中易于管理。这极大地降低了不同领域科学家参与空气污染研究的参与门槛，极大地提高了科学家的参与度。PM2.5的单位为ug/m^3，O3和NO2的单位为ppb。