The University of Texas at San Antonio, Department of Civil and Environmental Engineering, BSE 1.202, One UTSA Circle, San Antonio, TX 78249, USA.
Environ Pollut. 2013 Jul;178:411-8. doi: 10.1016/j.envpol.2013.03.035. Epub 2013 Apr 27.
The adverse health effects of high concentrations of ground-level ozone are well-known, but estimating exposure is difficult due to the sparseness of urban monitoring networks. This sparseness discourages the reservation of a portion of the monitoring stations for validation of interpolation techniques precisely when the risk of overfitting is greatest. In this study, we test a variety of simple spatial interpolation techniques for 8-h ozone with thousands of randomly selected subsets of data from two urban areas with monitoring stations sufficiently numerous to allow for true validation. Results indicate that ordinary kriging with only the range parameter calibrated in an exponential variogram is the generally superior method, and yields reliable confidence intervals. Sparse data sets may contain sufficient information for calibration of the range parameter even if the Moran I p-value is close to unity. R script is made available to apply the methodology to other sparsely monitored constituents.
地面臭氧浓度过高对健康的不良影响是众所周知的，但由于城市监测网络的稀疏性，估算暴露量很困难。这种稀疏性阻碍了在最容易出现过度拟合风险的情况下，为验证插值技术预留一部分监测站。在这项研究中，我们测试了多种简单的空间插值技术，用于 8 小时臭氧，数据来自两个城市地区的数千个随机选择的数据子集，这些监测站数量足够多，可以进行真正的验证。结果表明，仅用指数变异函数校准范围参数的普通克里金法是一种普遍优越的方法，并且产生可靠的置信区间。即使 Moran I p 值接近 1，稀疏数据集也可能包含足够的信息来校准范围参数。我们提供了 R 脚本，以将该方法应用于其他监测稀疏的成分。
